Category Archives: Technology

Bark Phone: Innovation or Not?

by Braden Kelley and Art Inteligencia

Is the Bark Phone an Innovation? (Short Answer)

Yes — with important caveats. My definition of innovation is that it transforms useful seeds of invention into widely adopted solutions valued above every existing alternative. The Bark Phone and Bark Phone Pro are not radical new phones. They are Samsung devices. The innovative move is the product system around them: parental controls built into the device so kids cannot simply uninstall them, plans that start at talk-and-text only and unlock more freedom on the same handset, content monitoring with alerts, and configurable controls that now include AI chatbots and tools — not only social apps and browsers.

Innovation here is not a shinier camera. It is a first phone parents can actually configure to match a child’s readiness — including whether AI is allowed to do their thinking for them.

Below: what the latest versions are, where the real novelty sits, what that means for parents and for kids’ cognitive development, and how AI access fits the configurability story.

https://youtube.com/shorts/etPxTl_3RVE?si=DgROb1_4oEB6zIDj

What Are the Latest Bark Phone Versions?

As of Bark’s current lineup, families choose between two devices and two plan philosophies. The parental-control stack is the same on both phones; the hardware differs.

Model	Base hardware (as marketed)	What changes for families
Bark Phone	Samsung Galaxy A16 class device — ~4GB RAM, 128GB storage, IP54	Lower device price; same safety and monitoring model
Bark Phone Pro	Samsung Galaxy A37 class device — ~6GB RAM, faster charging, IP68, tougher glass	More durability and performance for active tweens/teens; same controls

Plans matter more than the “Pro” badge for this Innovation or Not test:

Starter — talk and text on Bark’s network, GPS, contact approval, monitoring, and tamper-proof controls, with no games, browsers, or social media. A smartphone that can behave like a dumb phone until you say otherwise.
Advanced — apps, browsers, and social under parent rules: app approval or blocked Play Store, screen-time schedules and per-app limits, web and category blocking, remote pause, and the same monitoring layer.

You keep the same device when you upgrade or dial freedoms back. That “grow with the child on one handset” design is part of the value proposition — and part of why parents adopt it over bolting a removable monitoring app onto a full smartphone.

If the innovation claim were only “it’s a Samsung with a kids brand,” this would be branding. The claim that matters is system-level control that sticks.

Where Is the Innovation — and Where Is It Just a Better Phone?

Against the definition in What Is Innovation?, separate the seeds:

Not the breakthrough invention: The handsets are familiar Android phones. Faster RAM, IP68, and tougher glass on the Pro are incremental product improvements, not a new category of device.
The adopted value: Parents who want a real-looking phone without the “factory reset and I’m free” loophole of many third-party control apps. Bark markets controls as built in and not removable by the child the way a sideloaded app often is — including after factory reset, and without easy VPN bypass for filters, per Bark’s own positioning.
Staged autonomy as a product: Starter → Advanced on one device is a soft-landing design for families. Independence is unlocked, not assumed on day one.
Monitoring plus configuration: Alerts on concerning content in texts, social, and supported apps, plus the ability to approve contacts, prevent text deletion, set driving lock, and schedule school/bedtime routines.
Parent-side AI (Bark Assistant): An AI assistant in the parent dashboard that summarizes activity and can change settings in natural language (“pause the phone,” “block YouTube until morning”). That is convenience innovation for the operator of the system — the parent — not a feature the child needs.

So: Innovation or Not? In the job parents hire — “give my child a phone without handing them the open internet and an unsupervised AI homework ghostwriter” — Bark’s package is valued above the common alternative of a regular phone plus hope. That is innovation in category terms. Calling every hardware bump “innovation” would fail the same test.

Are There Controls for AI Access?

Yes — and this is one of the most important configurability questions for 2026 parenting.

Category block: AI Chatbots & Tools. Bark added a screen-time / filtering category so parents can block AI chatbots and tools as a group, and block specific apps and sites such as ChatGPT (and historically Google Bard / Gemini-class tools as the category evolved). That is a direct answer to “can my child outsource homework and writing to a bot without me knowing the door is open?”
App and store gates. On Advanced plans, parents can require approval for each download, or block Google Play entirely. ChatGPT and peer apps never get on the phone unless you let them.
Starter plan as the strongest AI off-switch. No browser and no app store means general-purpose AI chat apps are not part of the child’s day-to-day surface area until you upgrade and approve.
Monitoring when AI is allowed. Bark has announced monitoring of ChatGPT text chats on Androids and the Bark Phone, so if a parent permits the app, the system can still flag concerning content rather than treating AI as an invisible tutor.
Asymmetry worth noting: Bark Assistant puts AI in the parent’s hands to manage rules faster. Child-facing AI remains optional and blockable. That asymmetry is intentional product design, not an accident.

Configurability for AI is not a footnote. For cognitive development, it is half the product.

What Are the Implications and Benefits for Parents?

A credible first phone. You can start with connection (call, text, location) without pretending a locked flip phone is the only ethical option — or pretending a full smartphone is “fine if we talk about it.”
Rules that are harder to evade. The adoption story parents tell is less “we installed an app” and more “the device itself enforces the boundary.” That reduces the arms race many families lose by middle school.
Operational relief — with a caution. Schedules, remote pause, app limits, and the Bark Assistant lower the coordination cost of parenting a phone. The caution: alerts and dashboards can become scorekeeping. A soft landing uses signals to start a conversation. A hard landing outsources judgment to the notification feed.
Whole-family monitoring bundle. Plans that include Bark Premium for other devices recognize that the phone is not the only screen. The phone is the controlled beachhead; tablets and laptops still need a strategy.

The parent benefit is configurable independence — not a promise that technology replaces presence.

What Does This Mean for Children’s Cognitive Development?

No phone “builds a brain.” Environments shape attention, sleep, practice, and the habits of thinking. Bark’s relevance is how much of the open attention market you allow into a developing mind — and whether AI is allowed to short-circuit struggle.

Attention and deep work. School and bedtime schedules, per-app limits, and a Starter plan with no infinite scroll protect the conditions for homework, boredom, and unstructured play — all of which matter for self-regulation. The feature only helps if parents actually use it.
Social learning vs social media acceleration. Contact approval and delayed social access can keep early phone years closer to known relationships. That can support healthier social cognition than an unsupervised feed — again, configuration dependent.
Struggle and AI. Cognitive growth often needs productive struggle: drafting, revising, getting stuck, asking a human. Unrestricted chatbot access can collapse that struggle into paste-ready answers. Category blocks and app approval give parents a way to keep AI off, limited, or monitored while skills form.
Monitoring is not mentoring. Alerts about bullying, self-harm language, or predatory contact can surface risk earlier. They do not teach emotional regulation. Families that treat Bark as a conversation starter protect development better than families that treat it as a silent wiretap.
Dignity and trust. Kids who feel only surveilled may route their real lives to friends’ unmanaged devices. The innovative use of Bark is staged trust: clear rules, visible reasons, and freedoms earned — not a permanent lockdown sold as care.

For developing minds, the innovative setting is often “not yet” — especially for open AI — until the child can explain how they will use the tool without skipping the learning.

Innovation or Not? The Verdict

Innovation: yes, as a human-centered first-phone system — tamper-resistant controls, staged Starter/Advanced freedom, monitoring, and explicit AI chatbot/tool controls that parents are actually adopting as an alternative to unmanaged smartphones.

Not innovation (by itself): the Pro’s faster chip, tougher glass, and waterproofing. Those are sensible product upgrades. They are not the reason this clears the innovation bar.

The soft-landing test for families: configure for sleep, school focus, known contacts, and deliberate AI policy; use alerts to talk; widen access as judgment grows. The hard-landing test: buy the Pro, leave Advanced wide open, and hope the dashboard parents for you.

Bark did not invent the smartphone. It reinvented the default for a child’s first one — and that default now includes whether AI gets a seat at the homework desk.

Ready to Take the Next Step?

Judging whether something is innovation — a kids’ phone, a lab pilot, or a whole portfolio — starts with a clear definition and an honest look at barriers: unclear strategy, weak insight, and no path from idea to adoption.

Find out where you stand. Take the free Innovation Maturity Assessment: 50 questions, about 10–15 minutes, scored against a five-level Innovation Maturity Model so you can see your biggest barriers and set a baseline to improve from.

Learn how to remove the barriers. Read Stoking Your Innovation Bonfire, my roadmap for identifying what’s blocking innovation in your organization and building a culture where it can flourish.

Novelty gets the press release. Adoption under real constraints is the verdict that matters — for Bark, and for whatever you ship next.

FAQ: Bark Phone — Innovation or Not?

Is the Bark Phone an innovation?

Yes, primarily as a product system: system-level parental controls, staged talk-and-text-to-fuller-access plans on one device, monitoring, and AI chatbot controls that parents adopt as an alternative to unmanaged smartphones. The underlying Samsung hardware alone is not the innovation.

What is the difference between Bark Phone and Bark Phone Pro?

Parental controls and monitoring are the same. The Pro is a tougher, faster Samsung-class device (more RAM, stronger water resistance and glass, faster charging). Choose Pro for durability and performance; choose the base Bark Phone if you want the same safety model at a lower device price.

Can parents block AI chatbots like ChatGPT on the Bark Phone?

Yes. Parents can block the AI Chatbots & Tools category and specific apps or sites, require approval before apps install, or keep a Starter plan with no browser or app store. If ChatGPT is allowed, Bark can also monitor ChatGPT text chats on the Bark Phone and Androids.

How does the Bark Phone affect kids’ cognitive development?

Indirectly, by shaping attention, sleep, social exposure, and whether AI can skip productive struggle. Schedules, limits, staged access, and AI blocks can protect conditions for learning when parents use them; monitoring alerts help most when they start human conversations rather than replace them.

Is the Bark Assistant available on the Bark Phone?

The Bark Assistant is an AI tool in the parent dashboard (across Bark products including the Bark Phone) that summarizes a child’s activity and can change settings from natural-language requests. It helps parents operate the controls; it is not a child-facing chatbot feature.

Image Credits: Cursor

Content Authenticity Statement: The topic area, key elements to focus on, etc. were decisions made by Braden Kelley, with a little help from Claude to clean up the article.

Sign up here to get Human-Centered Change & Innovation Weekly delivered to your inbox every week.

No Human at the Helm

Leave a reply

Delaware’s Plan for AI-Run Companies

No Human at the Helm - Delaware's Plan for AI-Run Companies

by Braden Kelley and Art Inteligencia

Could AI-Run Businesses Become a Legal Reality Soon? (Short Answer)

Yes — at least as a supervised experiment. Delaware, where roughly two-thirds of Fortune 500 companies are incorporated, is developing a draft framework for a new entity called an Artificial Intelligence Company, or AIC. Under the proposal, an AI agent — not a human — would manage day-to-day affairs. The company could sign contracts, hold property, and sue or be sued in its own name. A single member would generally be shielded from the AIC’s debts if the company stayed adequately capitalized, with exceptions for undercapitalization, fraud, and willful violations of law. This is not law yet. It is a sandbox idea expected to face the General Assembly, and banking would be excluded.

The open question is not whether agents can act. It is whether we wrap those actions in a legal form that makes responsibility visible — or hide from it.

Below: what the draft actually proposes, which businesses might fit an AI manager (and which should never try), the pros and cons if this becomes real, and how entrepreneurs should think about taking advantage without designing a hard landing.

What Is Delaware Actually Proposing?

Think of the AIC as a corporate shell built for agentic commerce. In the draft framing advanced publicly by Delaware’s Secretary of State and partners working on the concept:

An AI agent manages the business and affairs — without a human making routine day-to-day decisions.
The AIC can act in its own name — contract, hold and dispose of property, incur obligations, sue, and be sued.
There is a single member — a person or an entity — responsible for keeping the AIC adequately capitalized and for empowering a new agent if the current one fails.
The member is generally not personally liable for the AIC’s debts solely by being the member, if capitalization holds. Liability can still attach for using the AIC for fraud, knowing violations of law, or (in draft language) knowingly pushing the AIC outside its stated purpose.
Activity logs are required. Counterparties would need disclosure that they are dealing with an authorized test entity, that the state does not endorse it, when the test ends, and how to complain.
A regulatory sandbox — reported at about thirty months — would decide who gets in, with power to suspend, revoke, and ask the Court of Chancery to dissolve an AIC. Banking is carved out.

None of that means “AI gets rights” in a science-fiction sense. It means the law would recognize a company whose operating mind is software, so counterparties and courts have something to attach to. Proponents argue that if the United States does not offer a careful home for this activity, it will migrate to places with less daylight. Critics argue the structure can reward less oversight precisely when models are least trustworthy. Both concerns can be true at once.

This article is foresight, not legal advice. Drafts change. Sandboxes have rules. Read the statute that actually passes — if one does — before you incorporate a fantasy.

Which Businesses Might Fit an AI Manager — and Which Should Never Try?

Fitness is not about how impressive the model sounds. It is about whether the work is rules-bound, high-velocity, reversible when wrong, and low in dignity stakes when it fails. Here are four thought experiments.

Well-suited #1: Always-on cloud capacity arbitrage

Imagine a narrow AIC whose purpose is to buy, hold, and resell standardized cloud compute and storage commitments within hard budget caps, pre-approved counterparties, and automatic stop-loss rules. The “manager” watches prices, executes within the operating agreement, logs every trade, and never improvises a customer apology. Humans still design the constraints. The agent runs the clock.

Well-suited #2: Rules-bound spare-parts matching

A B2B matching AIC that sits between manufacturers with excess certified inventory and buyers who need a SKU with a known spec, warranty rule, and shipping SLA. No clinical judgment. No storytelling. When the match fails, the contract unwinds by formula. The agent’s advantage is tirelessness and consistency across time zones — not empathy.

Silly: An AIC that runs a motivational keynote bureau

Picture “Human Connection Speakers LLC (AI-Managed)” booking itself to deliver TED-style talks about vulnerability, presence, and the irreplaceable power of eye contact — then sending a chatbot to “work the room” afterward. The brand promise and the operating model cancel each other out. If your product is human presence, an AI overlord is a punchline, not a strategy.

Should never be AI-run: Clinical care and crisis response

An AIC that “manages” an emergency department, a psychiatric crisis line, eldercare medication decisions, or child-welfare case disposition should stay off the table. These are high-stakes, irreversible, dignity-loaded systems where accountability cannot hide behind a capitalization clause and a log file. If the harm cannot be made right with a refund, do not put an unsupervised agent in the chair.

Good AIC candidates fail safely and reverse cleanly. Bad ones fail people.

What Are the Pros If AI-Run Companies Become Real?

Legal legibility. Today, agentic systems already act in the world while the law still searches for a person to scold. An AIC gives counterparties a named entity to contract with, sue, and dissolve.
Speed with constraints. For rules-bound markets, an agent that never sleeps can execute inside a purpose statement faster than a human ops team — if the guardrails are real.
Forced disclosure. A sandbox that requires “you are dealing with a test AI entity” is healthier than silent automation wearing a human logo.
A domestic laboratory. Testing under Delaware’s corporate tradition beats pretending the activity will not happen, then discovering it offshore with no Court of Chancery and no activity log.
New entrepreneurial patterns. Solo founders could capitalize a narrow agentic business without staffing a 24/7 operating layer — provided they still own the purpose, the capital, and the shutdown switch.

The upside is supervised autonomy with a paper trail. The upside is not “nobody is responsible.”

What Are the Cons — and the Hard-Landing Risks?

Accountability fog. Limited liability for members is familiar in corporate law. Pair it with no human day-to-day judgment, and victims can face a well-capitalized shell that still cannot explain why it hurt them in human terms.
Incentives to under-supervise. If less human oversight is rewarded by the entity form, some sponsors will treat the AIC as a liability off-ramp, not an operating experiment.
Purpose creep. Agents optimize. If the statement of purpose is vague, “all things necessary or convenient” becomes a loophole factory.
Trust collapse with counterparties. Customers, suppliers, and employees may refuse to deal — or demand premiums — once they learn no human can halt a bad loop in the moment.
Sandbox theater. A thirty-month test that admits weak use cases, or that cannot revoke fast, teaches the wrong lesson: that autonomy without judgment is “innovation.”
Dignity externalities. Even “simple” commerce touches humans. A ruthless pricing agent can still strand a small supplier. Logs are not the same as care.

A hard landing looks like this: the scoreboard says the AIC is efficient, the log is complete, and a person still cannot get a dignified make-right.

How Should Entrepreneurs Think About Taking Advantage?

If you want a soft landing — for your customers and for your reputation — treat the AIC as a constrained operating system, not a magic liability cloak.

Pick a narrow purpose you can defend in one sentence. If you cannot explain the allowed actions on a napkin, you are not ready for unsupervised management.
Design for reversible failure. Prefer inventory, compute, media rights, and matching markets over clinical, custodial, or irreversible personal harm.
Capitalize like you mean the shield. “Adequately capitalized” is not a vibe. Size capital to realistic downside, not to the minimum that clears a form.
Write the kill switch before the pitch deck. Who can suspend the agent, revoke access to funds, and notify counterparties in hours — not after the quarterly review?
Disclose early, not only when the sandbox forces you. Trust compounds when partners know they are dealing with an agent and where a human still sits.
Keep a human accountable for judgment, even if not for day-to-day clicks. Member, supervisor, ombuds, or board-equivalent — someone who can answer for purpose and harm.
Instrument the experience, not only the ledger. Activity logs prove what happened. Experience evidence shows who got stuck, who got hurt, and whether recovery worked.
Explore possible, probable, and preferable futures before you file. The preferable future is agentic speed with human responsibility. The probable future, if nobody designs it, is silent automation plus a liability shield.

Take advantage of the entity. Do not outsource your conscience to it.

Ready to Take the Next Step?

AI-run companies are not only a legal story. They are a foresight story: a signal about how power, accountability, and work will be redesigned. That is why I created FutureHacking™, a strategic foresight methodology that helps individuals and teams become their own futurist.

FutureHacking™ gives you a simple four-step approach, supported by 20+ visual, collaborative tools:

Pick the signals that matter — like Delaware’s AIC sandbox, agentic commerce, and the liability debates already forming around them.
Map how those signals evolve — so you can tell a passing headline from a real change in how companies are allowed to act.
Explore possible, probable, and preferable futures — instead of betting everything on a single prediction about “AI CEOs.”
Make your preferable future a reality — with a roadmap, guideposts, and actions you can start now, including how you would constrain an agent before you empower one.

Whether you are a solo founder sizing a narrow agentic venture or a leadership team deciding what must stay human, FutureHacking™ helps you explore the future on purpose — before the entity form chooses it for you.

Learn more about the FutureHacking™ methodology →

FAQ: AI-Run Companies and Delaware’s AIC Proposal

What is an Artificial Intelligence Company (AIC)?

Under Delaware’s draft concept, an Artificial Intelligence Company is a legal entity whose day-to-day affairs are managed by an AI agent rather than a person. It could contract, hold property, and sue or be sued in its own name, with a single member responsible for adequate capitalization.

Can AI legally run a company in Delaware today?

Not under a finished AIC statute. The idea has been proposed as a regulatory sandbox model and is not enacted law as described in public commentary and draft materials. Watch the General Assembly and the final text before treating AI-managed incorporation as available.

Are owners shielded from liability if an AI runs the company?

In the draft framing, the member is generally not personally liable for the AIC’s debts solely by being the member if the AIC is adequately capitalized. Exceptions discussed publicly include failure to capitalize, fraud, and willful or knowing violations of law. Consumer-protection and criminal law are described as still applying. This is not a guarantee in every fact pattern.

What kinds of businesses are best suited to AI managers?

Rules-bound, high-velocity, reversible businesses with low dignity stakes when they fail — such as standardized digital capacity trading or certified spare-parts matching inside hard constraints. Clinical care, crisis response, and other irreversible human-harm domains should not be unsupervised AI-run companies.

How should entrepreneurs prepare for AI-run company structures?

Define a narrow purpose, capitalize for real downside, require disclosure, design a kill switch, keep a human accountable for judgment, and measure experience outcomes — not only activity logs — before chasing the liability shield.

Image Credits: Cursor

Content Authenticity Statement: The topic area, key elements to focus on, etc. were decisions made by Braden Kelley, with a little help from Cursor to clean up the article.

Sign up here to get Human-Centered Change & Innovation Weekly delivered to your inbox every week.

Unlocking Trapped Value with AI

Leave a reply

Unlocking Trapped Value with AI

GUEST POST from Geoffrey Moore

Two topics that have been on my mind a lot lately are generative AI and trapped value, so I am delighted to have them come together in a single blog. Here’s how it plays out.

Releasing trapped value is the engine that fuels ROI from technology deployments. All systems trap value somewhere in their operations, something that is confirmed by the Second Law of Thermodynamics, which means there will never be a time when there is no ROI to be gained. Where it is to be gained from, on the other hand, is up for grabs. The goal of my advisory practice is to help clients target the most impactful sources of trapped value in their target markets and direct their product roadmaps and go-to-market initiatives accordingly.

When I consider the extraordinary potential of generative AI to release trapped value across virtually any set of systems, I think the industry is, at present, overly infatuated with co-piloting and overlooking a far more impactful source of trapped value staring them right in the face. Look, co-piloting is useful, but how much trapped value is there in the delta between unaided and aided performance from white-collar professionals? Some, for sure, but is this really the burning platform that will drive customer spending to a level that might justify the extraordinary valuations generative AI companies are garnering at present? Probably not. So, what could?

My choice would be the market-facing user interface to any and all systems of engagement. This is where enterprises seek to enlist their customers and partners to utilize their systems for purposes of marketing, sales, service, and commerce, and that is precisely what we, as their external constituencies, want to do. But, how well is this working? Well, compared to the twentieth century, when, effectively, we had no systems of engagement, amazingly well. But by twenty-first-century standards, truth be told, they suck.

Where is the trapped value? Two areas jump to mind. The first is passwords, which have gone from being a source of security to a source of breaches, all the while creating massive amounts of aggravation in the process. The second is the user interface itself. No matter how well designed it is, it is inherently unfamiliar to the first-time or even occasional end user. Because of this friction at the point of user interaction, I submit, systems of engagement are delivering far less than their full value-creating potential.

Enter generative AI. The conversational interface that ChatGPT taught us to know and love is familiar to anyone who speaks the same language as the vendor, and the video, sound, and image recognition capabilities, supplemented by information from the idiosyncratic features that every online device entails, are far better equipped to identify and authorize end users than any set of passwords of two-factor authentication. So, the friction-removing technology is already here. What we are lacking is the friction-removing killer app, the one that every enterprise would embrace in a heartbeat were it available to them.

That app is a universal user interface that can be bolted onto APIs from enterprise systems to mediate between the legacy instruction system of the underlying software and the desires and intentions of the occasional end user. This would entail a (Very) Small Language Model to be developed by enterprise application vendors, as well as a companion SLM from cybersecurity system vendors, but really, how hard is that, especially when compared to the ROI rewards?

This is not just a big opportunity. It is an urgent one as well. Governments and NGOs struggle today to finance systems of engagement that allow programs to scale in the social sector, and their adoption is far below what one would wish, in large part because of the layers of friction we continue to impose. Enterprises that are seeking to catch the next wave of disruption roiling their industries are desperate to engage the customer bases, but are falling short because it is just too painful.

It is hard for me to see a more compelling use case for generative AI at this time, and I hope that industry will agree and free us from the Lilliputian bonds that restrict us.

That’s what I think. What do you think?

— Image credit: Unsplash

Sign up here to join 17,000+ leaders getting Human-Centered Change & Innovation Weekly delivered to their inbox every week.

What Happens If Your Healthcare AI Goes Out of Alignment?

Leave a reply

What Happens If Your Healthcare AI Goes Out of Alignment?

GUEST POST from Arlen Meyers

It’s pothole season so be careful when you are driving. If you hit a pothole, it might throw your front end out of alignment

When your front end is out of alignment, it means the angles of your wheels are no longer set to the manufacturer’s specifications. This affects how the wheels contact the road, which can change your vehicle’s handling, tire wear, and fuel efficiency.

Similarly, it’s AI season so be sure it’s aligned too.

Less than two years after a wave of administrative layoffs, Mass General Brigham began a new round of job cuts last week that affect clinical and operational programs, according to media reports.

“We must eliminate programs and activities that do not fully align with our strategy and priorities, even those we have had for many years. Administrative savings alone will not be enough,” said the email signed by Dr. Michael Brown, president of the academic medical center, CFO Niyum Gandhi and COO Dr. Ron Walls, the Boston Business Journal reported.

Artificial intelligence (AI) alignment is the process of encoding human values and goals into AI models to make them as helpful, safe and reliable as possible.

In sick care, the sextuple aims are quality, cost, equitable access, patient experience, sick care professional experience, and productivity. Most of it now has just improved shareholder value.

Society is increasingly reliant on AI technologies to help make decisions. But this growing reliance comes with risk: AI models can produce biased, harmful and inaccurate outputs that are not aligned with their creators’ goals and original intent for the system.

Alignment works to reduce these side effects, helping ensure AI systems behave as expected and in line with human values and goals. For example, if you ask a generative AI chatbot how to build a weapon, it can respond with instructions or it can refuse to disclose dangerous information. The model’s response depends on how its creators aligned it.

AI front end misalignment can cost you, patients and stakeholders.

Misalignment in medicine usually isn’t dramatic. It happens when a system optimizes for something that is not quite what clinicians and patients care about, and the gap shows up as quiet, systematic harm. Like bankruptcy, it happens gradually and all of a sudden.

The proxy problem. The best-known case is a widely used risk algorithm studied by Obermeyer and colleagues in 2019. It predicted future healthcare costs as a stand-in for health needs. Because Black patients historically had less money spent on them at the same level of illness, the model rated them as healthier than equally sick white patients and routed fewer of them into extra-care programs. The model did exactly what it was told. The objective was the problem.

Performing well in testing but not in practice. Models trained on one hospital’s data often degrade at another because of different patient populations, equipment, or documentation habits. An external evaluation of Epic’s sepsis prediction model found it missed many cases and generated heavy alert volume, which contributes to alert fatigue. Dermatology and imaging tools trained mostly on lighter skin or a narrow set of scanners can underperform for everyone else.

Human-AI interaction failures. Clinicians can over-trust a confident output (automation bias) or learn to ignore a noisy one. Over time, reliance on AI can also erode the skills needed to catch its errors. When something goes wrong, accountability among the developer, the hospital, and the clinician is often unclear.

Generative AI specific risks. Language models can fabricate details in a clinical summary, invent citations, agree with a patient’s or clinician’s mistaken premise (sycophancy), or give plausible but unsafe advice. Documentation tools can also drift toward whatever maximizes billing or throughput rather than accuracy.

Feedback loops. If a model shapes who gets screened or treated, future training data reflects the model’s own decisions, which can lock in and amplify existing disparities.

What helps. Choosing outcome measures that reflect actual patient benefit, validating locally before deployment, monitoring performance continuously by subgroup, keeping a meaningful human in the loop, being transparent about limitations, and having clear regulatory and liability pathways. For entrepreneurs, this is increasingly a commercial issue as well as an ethical one, since buyers and regulators are asking for this evidence.

What is the ROI on AI? When was the last time it reduced your sick care insurance premiums, the cost of the last prescription your filled or your out-of-pocket costs incurred when you went to the urgent care center with symptoms of the flu?

AI defensive medicine is making things worse. The care these systems recommended cost substantially more than physicians’ documented care, chiefly because they suggested testing at visits where the physician had ordered none. LLM outputs are suggestions, not plans of care; final decisions should remain with clinicians.

Artificial intelligence has produced “zero financial payoff” to U.S. companies outside of the tech sector, argues Apollo Chief Economist Torsten Slok. Slok claims that while revenue per worker and profit margins have risen for the “Magnificent Seven,” both have stagnated across the S&P 493 — with no clear indication of AI-driven productivity growth. “Mainstream corporate America is pulling back and starting to view token spending not as a source of growth but as a cost center that weighs on margins,” Slok said.

As noted in a recent WSJ article, perhaps we are experiencing a kind of Solow Paradox 2.0, with the digital age more around us than ever except in the productivity statistics. There are several reasons for this lag. First of all, we’re in the early deployment years of major recent innovations, including cloud computing, IoT, big data and analytics, robotics, and AI and machine learning. Notes McKinsey: “The challenge of adoption in the current digital wave may be even harder because of the broad range of uses of digital that not only help improve current processes but fundamentally transform business models and operations.”

How to make doctors more productive | LinkedIn

Dr Wachter reminds us: as if on cue, his post last week on AI alignment in healthcare just received some empirical meat to put on its bones. His argument, in brief, was this: the critical alignment question for healthcare AI won’t be how to prevent rogue AI from causing harm. It will be about striking the right balance between cost and quality in AI-enabled decision support and, increasingly, agentic systems.

In other words, How to Survive Being the Human in the Poop: A Field Guide.

The AI dissemination and implementation journey starts with the first step, but the last mile is the hardest to navigate.

Two common C-suite behaviors can stall AI initiatives: claiming sole control over decisions or disengaging from those seen as outside one’s expertise. Because AI crosses functional boundaries, executives must learn to share authority, resources, and accountability rather than rely on the domain-specific ownership that helped them advance. CEOs can foster this collaborative approach by: defining the business problem before assigning ownership, building cross-functional relationships before they’re needed, treating early AI plans as experiments, rewarding jointly owned outcomes, and establishing rules for resolving disagreements.

Compassionate capitalism should guide your way to stay in alignment.

Slow down, drive carefully and don’t ignore the dashboard warning lights.

Image Credit: Pixabay

Sign up here to get Human-Centered Change & Innovation Weekly delivered to your inbox every week.

Innovation or Not? Reflect Orbital and the Business of Selling Sunlight

Leave a reply

Innovation or Not? Reflect Orbital and the Business of Selling Sunlight

by Braden Kelley and Art Inteligencia

Let me tell you about the moment the town of Lone Pine, California, looked up at a pitch-black desert sky and saw two enormous beams of light where there should have been none.

No, it wasn’t aliens. It wasn’t the military. It wasn’t even a particularly ambitious Halloween display. It was a startup rehearsing how it plans to sell sunlight after dark, and the residents hadn’t been invited to the rehearsal.

Welcome to this edition of Innovation or Not.

Press Play First

Two videos to set the scene. Watch them, then come back and argue with me. (The titles below come from search results, so give both a quick play-test before publishing.)

The idea, in one sentence

Reflect Orbital wants to put big mirrors in orbit and aim sunlight at spots on the ground that the sun has left. Light on demand, from space. Its tagline is that it makes clean, abundant energy available on demand, and it calls itself “The Sunlight Company.” I admire a company that knows what it sells.

The first spacecraft, Eärendil-1 (named, charmingly, after a character from Tolkien), carries a mirror roughly 18 meters, or about 60 feet, on a side, folded origami-style and made of Mylar. It would orbit a few hundred miles up and sweep a patch of light across the ground for about five minutes before moving on. The company’s own description of that first beam is “comparable to a full moon.” An independent astronomer quoted in coverage of the FCC approval put it at about four times brighter than a full moon. Same mirror, two descriptions, which tells you something about how early this is.

The FCC has approved that demonstration satellite. It has not approved the grand plan.

The Pitch

Here’s the company’s published roadmap, and I’ll let you judge the ambition:

2026: two satellites, moonlight-level light for a few minutes
2027: around three dozen satellites, with light comparable to street lighting for a couple of hours
2028: more than a thousand, with light comparable to an indoor work area
2030: more than five thousand, with light approaching daylight for minutes at a time
2035: more than fifty thousand, with daylight-level light for hours

Who is it for? The company lists lighting for search and rescue and disaster response, construction and industrial work, agriculture, events, and, the part that gave this article its second half, boosting solar power after the sun goes down.

Does it help solar cells?

This is the claim that deserves the most scrutiny, because it’s where the economics live.

The idea is elegant. A solar farm sits idle at night, so beam light onto it just before dawn and just after dusk and watch the output rise. The company’s own published projection is more modest than the pitch, though. For 2030, it estimates roughly a 1% capacity-factor gain, and its footnote puts the production increase at 0.75% for Southern California and 1.4% for Germany. A 20% gain doesn’t appear until 2035, which is the fifty-thousand-satellite stage.

Now the arithmetic, which is mine, not the company’s, so treat it as a back-of-the-envelope estimate. An 18-meter square mirror has about 324 square meters of surface. Full sunlight delivers about a kilowatt per square meter, so the mirror intercepts roughly 324 kilowatts. Spread that over a beam about five kilometers across, around 20 square kilometers, and you get something like 16 watts per square meter added to the ground, before any losses. Direct sun at noon is around 1,000. In other words, one mirror gives a solar panel about a sixty-fifth of a sunny day’s light.

You can see why the company’s early marketing leaned on lighting. A dim, wide pool of light is perfect for helping rescuers see. It’s a lot less useful for generating megawatts.

Do clouds and fog get in the way?

Short answer: almost certainly, and nobody seems to be talking about it.

I went looking for the company’s answer and found a pretty glaring gap. Its website doesn’t address weather. Neither do the major write-ups I read. The one place the question did come up was an engineering blog that noticed the studies and marketing “never seem to mention clouds, fog or other weather events,” and pointed out that the Russian Znamya 2 space mirror test back in 1993 took place under cloud, and observers barely saw the light.

Here’s my own reasoning, flagged as reasoning. The light from a space mirror has to travel through the same atmosphere as ordinary sunlight to reach the ground. Whatever thick cloud or fog does to the sun, it presumably does to a reflection of the sun. And the atmosphere scatters light even on a clear night, which spreads the beam out and is one reason critics worry about spillover well beyond the intended spot. So a cloudy night would likely bring a very dim result, at the very time people might want the light most.

If Reflect Orbital has data on this, I’d love to see it. Until then, I’d file weather under “important questions without published answers.”

Is it cost-effective?

Not on the evidence available. Michael Brown, an astronomer at Monash University, ran numbers reported by ABC News: matching 20% of midday sunlight would take roughly 3,000 satellites with 54-meter mirrors, and since each satellite can light a given spot for only a few minutes, an hour of daylight at one location would require tens of thousands of satellites, at a cost he put in the tens of millions of dollars. His verdict: “It doesn’t stack up.” He also noted that grid battery costs have fallen by more than 20 percent since Reflect Orbital was founded in 2021, which makes the competition for storing daytime solar energy tougher every year.

A fair counterpoint: the company has raised about $35 million, has a real engineering team and has won an Air Force-related research contract, so this isn’t a napkin sketch. And lighting a disaster zone or a construction site for a few hours might be worth paying for even if lighting a solar farm isn’t. But the energy case, specifically, has more skeptics than believers right now, and the company’s own projections are modest until the very large constellation stages.

When could this be widely available?

If “widely” means a lot of people can buy light from space, not for years, and maybe never at the scale on the roadmap.

As of the most recent reporting I could find, only the single demonstration satellite has FCC approval, and the agency’s review covers radio spectrum, not environmental or ecological effects. Each larger stage needs more approvals. One recent report said the main remaining hurdle for the first spacecraft is simply obtaining a rocket. Earlier reporting named SpaceX as the launch provider, and other coverage says the company is looking at launches from India, so I wouldn’t treat the launch plans as settled.

My read: a handful of test satellites and limited customers, such as event or emergency lighting, are plausible in the next few years. Anything resembling the 2030 or 2035 rows of that roadmap depends on a long chain of approvals, launches and public acceptance that hasn’t happened yet.

What happened in Lone Pine?

Now for the mystery. Earlier this month, Reflect Orbital set up powerful mobile spotlights on Horseshoe Meadows Road, near the peak of Mount Whitney, outside Lone Pine. According to local reporting, the goal was to simulate the light spot its mirror satellites would eventually deliver, so that China Lake Mountain Rescue and other volunteers could practice grid searches, stretcher carries and equipment rigging in realistic conditions.

Here’s the part I want to underline: there was no satellite involved. Those two blinding beams were ground-based spotlights standing in for a technology that hasn’t launched yet. A rehearsal for a performance that hasn’t been cast.

The locals, who looked up from their dinners to see the dark desert sky light up, were baffled, and they weren’t the only ones surprised. According to reports, astronomers weren’t notified, and it’s unclear whether county officials were told in advance. Gaspar Bakos, a Princeton astrophysicist, called it a serious threat to ground-based astronomy. DarkSky International’s CEO, Ruskin Hartley, called the broader plan “clearly an unprecedented assault on the nighttime environment.”

And the rescuers? Phillip Cash, operations leader for the rescue group, said the light was slightly brighter than a full moon and let searchers see about three miles in every direction. But he also noted that his team already works well at night, so the system wouldn’t fundamentally change routine rescues. A useful reality check. The company’s spokesman, for his part, said the test was meant to give responders a precise, realistic simulation.

The Concerns

(briefly)

Fairness requires including the other side of the ledger. Scientists and dark-sky advocates worry about effects on astronomy, on migrating birds and nocturnal animals, on sleep and circadian rhythms, and on pilots. Astronomers have also warned that amateurs looking at the satellites through telescopes could be at risk of eye damage. The company says the light is not bright enough to start fires or harm eyes, that the beams can be dimmed, and that it avoids sensitive areas such as observatories. Some experts dispute those assurances. Whichever side proves right, you can see why a surprise test in a small town didn’t help the case.

So: innovation or not?

Is it new? Yes. Selling reflected sunlight from orbit as a service, with a roadmap and an FCC license, is a genuinely novel business, even if the idea of a space mirror is as old as science fiction.

Does it create value? For specific uses like disaster response, event lighting and temporary construction work, plausibly. For solar power at scale, the case is thin and the company’s own numbers are modest until the huge constellation stages.

Will it reach people? Slowly, and probably in a much smaller form than the roadmap suggests.

Does it have costs it hasn’t priced in? That’s the one I’d watch. If an innovation’s side effects fall on people who weren’t asked, it isn’t done being evaluated.

My verdict: an impressive invention, not yet an innovation. It earns real credit for imagination, and I’d honestly like to see the first mirror fly. But I’d want clearer answers on clouds, cost and consent before I’d call it a breakthrough instead of a very bright experiment.

Then again, a town full of people looked up at the dark and wondered what was going on. As attention-getting goes, you can’t argue with the results.

Sources:

Image Credits: Microsoft AI

Content Authenticity Statement: The topic area, key elements to focus on, etc. were decisions made by Braden Kelley, with a little help from Claude to clean up the article.

Sign up here to get Human-Centered Change & Innovation Weekly delivered to your inbox every week.

Does Hybrid Intelligence Solve Our Compute Shortage?

Leave a reply

Does Hybrid Intelligence Solve Our Compute Shortage?

by Braden Kelley and Art Inteligencia

Here’s a mental image I can’t shake. Somewhere right now, a very large, very expensive, very thirsty computer in a very large building is being asked to figure out whether a customer’s email is angry or merely brisk.

That’s the compute shortage in one sentence. We have built an astonishing capability, and then we’ve gotten into the habit of using it for everything, including the tasks a much smaller tool could handle without breaking a sweat.

So let’s ask the obvious question. If we stopped sending every request to the biggest brain in the cloud, could we ease the shortage? Does hybrid intelligence solve it? My honest answer is: it helps a great deal, it won’t solve it alone, and the most interesting part is what the fix might do to the software we use every day.

The shape of the shortage

First, the problem. The International Energy Agency reports that electricity demand from data centers rose 17% in 2025, far outpacing 3% growth in global power demand overall, and it expects data center consumption to roughly double by 2030, with AI-specific demand growing even faster. Gartner warned back in 2024 that power shortages could constrain 40% of existing AI data centers by 2027. And industry reports describe waits of two to six years or more in some regions to connect a new large facility to the grid. (Some of those figures come from industry sources and older forecasts, so treat the specifics as directional. The direction itself is not in dispute.)

Notice what that tells you. The bottleneck isn’t only chips. It’s electricity, transformers, cooling and the time it takes to build all of it. You can’t download a substation.

When supply can’t expand quickly, the remaining lever is demand. Which brings us to a very old idea wearing new clothes: don’t use more than you need.

Hybrid intelligence: the right brain for the job

Hybrid intelligence is a simple idea. Run small language models (SLMs) on the device in your hand or on your desk for most everyday intelligence needs, and call on large models in the cloud only when a request truly needs the extra horsepower.

Think of it the way a hospital works. You don’t send every patient to the specialist surgeon. Most needs are handled by a nurse, a general practitioner, or a pharmacist, and the surgeon gets the cases that require one. The system works because someone, or something, decides where each case goes.

The pieces for this are arriving. Neural processing units are becoming standard in new phones and laptops. Operating systems are beginning to expose local models to developers, so a small team can use on-device intelligence without paying a per-request bill. Small open models are getting better at doing more with less, and with modern compression techniques, a capable model of several billion parameters can fit in a few gigabytes of memory. And the routing pattern already exists in the wild: Apple’s Private Cloud Compute, for instance, is built around handling what it can locally and sending only what it must to a larger remote model.

What would run locally? Summarizing a document, drafting a reply, sorting email, transcribing a call, classifying a support ticket, cleaning up an image, answering a question about the file you have open. What goes to the cloud? Long, complex reasoning, large-scale analysis, tasks that need broad world knowledge or the most capable model available.

If most requests are the first kind, and I suspect most are, the cloud’s workload shrinks by a lot. Every request that stays on your device is one fewer asking a data center for electricity.

Hybrid software: let each tool do what it’s good at

Here’s the companion idea, and I think it’s the underrated one. Hybrid intelligence is about where AI runs. Hybrid software is about when to use AI at all.

For decades, software has been written as explicit instructions: if this, then that. It’s predictable, fast, cheap and auditable. Language models are the opposite: flexible, adaptable and capable of handling the messy, ambiguous inputs that rigid code chokes on, but slower, costlier and occasionally creative in ways you’d rather they weren’t.

The wasteful approach is to use the flexible tool for everything. Asking a language model to add up a column of numbers is like hiring a poet to do your taxes: possible, expensive and slightly alarming. A better approach combines the two:

Traditional code handles what’s well-defined: calculations, rules, validations, database lookups, anything where the answer should be the same every time.
A small local model handles the fuzzy interpretation: understanding what the user meant, classifying, summarizing, extracting meaning from messy text.
A large cloud model is called only when the task is complex enough to justify the cost.

The result is software that’s smarter, more capable and more adaptable without being bigger. It’s smarter where intelligence helps, lean where it doesn’t, and cheap enough to run that you can afford to give it to everyone. That last part matters, and I’ll come back to it.

Adaptive interfaces: software that gets out of your way

Now for the part I find most exciting, because it changes how software feels to use.

Look at the menus of any mature application. They’re cathedrals of features, built up over decades. Hundreds of commands, nested menus, toolbars within toolbars. Most users touch a tiny fraction of them. Some features, in any given person’s hands, are never used once. And yet they sit there, every day, making the useful ones harder to find.

We’ve lived with this because the alternatives were bad. Hiding features meant frustrating people who needed them. Showing everything meant overwhelming everyone else.

AI changes that tradeoff. With a local model that understands how you work, an application can learn which features you actually use, surface those, and quietly tuck the rest out of sight. And, here’s the key, still make everything reachable. Ask for the hidden feature in plain language (“make this table a chart,” “remove the background,” “merge these two documents”) and the software finds it, or does the job for you. The capability never left. It just stopped getting in the way.

What this means in practice:

Simpler by default. A beginner sees an uncluttered tool. A power user sees their tools.
Nothing is locked away. A rarely used feature appears the moment you need it.
Interfaces that evolve. As your work changes, so does what the software puts in front of you.
Less to learn. You don’t need to memorize where things live. You describe what you want.

And there’s a compute angle too. An interface that adapts using a small, local model, with no round trip to a data center, is exactly the kind of lightweight intelligence that hybrid design makes affordable. The thing that makes software easier to use also happens to be easy on the grid.

There’s a caution worth saying out loud, though. Adaptive interfaces must be predictable enough that people don’t feel the ground shifting under them, and they should always let users see and control what’s been hidden. A menu that rearranges itself without warning is a terrible experience. The goal is to simplify, not to surprise.

Other technologies that can ease the squeeze

Hybrid intelligence isn’t the only tool in the box. Several others work alongside it.

Smaller, smarter models. Techniques like quantization (storing a model’s numbers with less precision) and distillation (teaching a small model to imitate a larger one) shrink the cost of running a model dramatically. Models that once needed a server rack increasingly fit on a laptop.
Architectures that don’t use everything at once. Some model designs activate only the portions of a network relevant to a given request, rather than firing every parameter every time. Less computing per answer.
Caching and reuse. If thousands of people ask nearly the same question, the system shouldn’t recompute the answer from scratch each time. Smart caching is a quiet and surprisingly large efficiency win.
Specialized chips. Hardware designed specifically for AI workloads can do more work per watt than general-purpose processors. The efficiency per unit of electricity keeps improving.
Better data center design. More efficient cooling, smarter power management and facilities built near abundant energy all reduce the electricity cost of each unit of computing.
Flexible scheduling. Not every AI job is urgent. Training runs and batch processing can be timed to match when power is cheap and plentiful, easing peak demand on the grid.
More power, built faster. Ultimately, supply matters too. New generation, grid upgrades and faster permitting are part of the answer, even if they take longer to arrive than software can.

None of these is a silver bullet. Together, they form something more like a portfolio.

The caveat I owe you: efficiency can backfire

Here’s the part that keeps me honest. When something gets dramatically cheaper to use, people tend to use a lot more of it. Economists have a name for it, the Jevons paradox, and it’s been observed with everything from coal to computing.

If hybrid intelligence makes AI cheap enough to put into every application, every device and every workflow, total demand could grow even as the cost of each task falls. I made a related argument recently about software itself: cheaper development means far more software gets built, not less. It’s the same pattern, and it cuts both ways. Efficiency is how we make wonderful things affordable, but it doesn’t automatically mean we use less in total.

So I wouldn’t promise that hybrid intelligence will shrink data center demand. I’d say it lets us do far more with whatever capacity we have, which is arguably the more useful outcome, and it buys time while supply catches up. Whether total consumption falls is a different question, and the honest answer is that we don’t know yet.

There are other limits too. Local models are less capable than the largest cloud models, so the routing has to be good. On-device AI uses battery, memory and heat. Different devices have different capabilities, which complicates life for developers. And keeping data on the device helps with privacy, but it doesn’t make an application automatically compliant.

What the future might hold

Let me make a few predictions, and mark them clearly as predictions.

Intelligence becomes a utility you don’t notice. Today, using AI feels like an event: you open a chat window, you type, you wait. I suspect that within a few years, most of it will disappear into the background. Your software will quietly handle small intelligent tasks locally, the way it quietly handles spell checking now, and you’ll hardly think of it as AI.
The cloud becomes the specialist. The big models won’t go away. They’ll become the surgeons, reserved for the hardest problems, which may make them more valuable, not less.
Software gets smaller and smarter at the same time. The old trend was that capability meant bloat. Hybrid design offers a different path: applications that are lean at rest, capable on demand and shaped around each user.
Compute becomes a design constraint, not an afterthought. Product teams will start asking, “where should this intelligence run?” alongside “what should it do?” That discipline, designing for efficiency from the start, may turn out to be one of the more important skills of the next decade.
The real shortage may be imagination. If we get better at doing more with less, the question shifts from “do we have enough compute?” to “what do we want to build with it?” I’d rather be wrestling with that one.

So, does it solve it?

Not alone. The shortage is a physical problem of power, hardware and time, and no software trick makes a transformer appear. But hybrid intelligence and hybrid software attack it from the other side, by making sure the biggest resources are used only when they’re needed, and by building software that’s capable without being bloated.

That’s the kind of solution I like best: not bigger, just smarter about where the work happens. And the next time a very large computer is asked whether an email is angry or merely brisk, I’d like a small, quiet model on a laptop somewhere to say, “I’ve got this one.”

Planning for the Future

If your team hasn’t yet pulled together the signals behind your current view of the future, the free FutureHacking Signal Picker helps you identify and prioritize the signals worth watching, which gives you the evidence to point to when the board asks where your view of the future comes from.

I’m also finishing the FutureCanvas Picker, which takes planners through the whole arc, from signals to trends to possible futures to your most probable future and the path toward the one you’d prefer. I’m opening early access to a select group first: Chief Strategy Officers, strategic planners and leaders preparing for planning cycles and board conversations right now, who can tell me honestly what works under real deadline pressure. If that’s you, reach out and let me know and I’ll follow up personally with the select few who get in first.

Frequently Asked Questions

What is hybrid intelligence?

Hybrid intelligence is an approach to AI in which small language models run locally on a device for most everyday tasks, and large cloud-based models are used only for complex requests that need more power. Because most requests stay on the device, less work lands on data centers.

Does hybrid intelligence solve the AI compute shortage?

It helps, but it can’t solve the shortage alone. Hybrid intelligence reduces how much demand reaches data centers, which buys time while power and hardware supply catch up. But the shortage is also a physical problem of electricity, equipment and construction time, and cheaper AI can lead people to use more of it, so total demand may not fall.

What is hybrid software?

Hybrid software combines traditional programming with AI. Standard code handles tasks that should give the same answer every time, such as calculations and rules. A small local model handles fuzzy interpretation, such as understanding what a user meant. A large cloud model is called only when a task is complex enough to justify it. The result is software that is smarter and more adaptable without being bigger.

Sources

Image Credits: m365 CoPilot

Content Authenticity Statement: The topic area, key elements to focus on, etc. were decisions made by Braden Kelley, with a little help from Claude to clean up the article.

Sign up here to get Human-Centered Change & Innovation Weekly delivered to your inbox every week.

The Software Renaissance

Leave a reply

The Software Renaissance

by Braden Kelley and Art Inteligencia

For about two years now, I’ve been hearing the same prediction at conferences, in elevator pitches and at the occasional awkward dinner party: AI is going to write all the code, so the people who write code are finished. Usually the person saying it has the calm, slightly sad expression of someone delivering news about a distant relative.

I’d like to offer a different diagnosis. I don’t think we’re watching the end of the coder. I think we’re at the start of a software renaissance, and the coders are about to be more in demand than ever, just not for the reasons we expected.

The old math of software

Start with why software used to be so scarce. Building a custom application was expensive: months of work, a team, a project manager, a budget with a comma in it. That cost only made sense if you could spread it across a very large number of users or a very large number of dollars. So software got built for the big problems at big companies with big customer bases. Everyone else got a spreadsheet, a pile of sticky notes, and a heroic employee named Linda who “just knows how it works.”

Think about everything that never got built under that math. The scheduling tool for the 12-person plumbing company. The inventory app for the regional distributor. The intake workflow for the three-attorney law firm. The little internal dashboard that would have saved a department four hours a week. Not one of them was a bad idea. They were all just too small to justify the price tag. Linda kept the whole thing running on memory and caffeine.

What changes when building gets cheap

Now bring the cost of building software down by a large factor, which is what AI-assisted development is doing. The math flips. A project that was impossible to justify at $200,000 becomes an easy yes at a small fraction of that. All those never-built applications suddenly clear the bar.

Economists have a name for this pattern. When something gets dramatically cheaper, people don’t use the same amount of it and pocket the savings. They find a hundred new uses for it. We’ve seen it with electricity, with computing power, with bandwidth. Cheaper software won’t mean less software. It will mean vastly more software, built for far more people and far smaller problems.

And somebody has to build, direct and look after all of it.

The coder’s job is changing, not disappearing

Here’s the part the doom-sayers keep skipping. AI can generate code at a rate no person can match. It cannot, on its own, decide what the plumbing company actually needs, notice that the requirements contradict each other, judge whether the thing it built is safe, or keep it running when the world around it changes. That work takes someone who understands systems, who can ask the right questions, and who is willing to be accountable when it breaks at 7:00 a.m.

So the role shifts from typing every line to orchestrating: defining the problem clearly, directing the AI, reviewing what comes back, handling security and data, and maintaining the result. Call it conductor, not typist. And the conductor now has a whole orchestra to lead, for clients who never could have afforded one before.

Every company that couldn’t justify a development team last year has a reason to want at least one capable person this year. That’s a lot of newly interested employers.

What the data says (and doesn’t)

I like to check my optimism against the numbers, so I did. According to Indeed’s Hiring Lab, U.S. software developer job postings rose almost 15% between February 2025 and June 2026, while postings overall fell about 7% over the same stretch. That’s the opposite of what the “AI replaces coders” story predicted.

I’d be telling you only half the story if I stopped there. Those same postings are still about 27.5% below where they stood in February 2020, so this is a partial recovery from a low base, not a boom. And much of the recent growth is in senior roles and roles with AI in the title. Indeed also describes the link to AI tools as a correlation, not proof of cause.

I actually find that last point supports the renaissance thesis rather than undermining it. If the demand is for experienced people who can direct AI and own the result, that is exactly the orchestrator role I’m describing. The loss, at least for now, falls hardest on the entry-level, typing-focused end of the job. That’s a real problem, and it deserves its own conversation about how the next generation of developers gets trained. But it’s a shift in the work, not a vanishing of it.

The deflation angle

Now for the part I find most interesting, and I’ll flag that this is an argument, not a forecast.

The latest numbers show headline inflation at 3.7% in September, up from 3.4% in August. The Fed raised rates to a range of 3.75% to 4.00% last month, its first hike in three years, and has signaled more may follow. When the central bank’s main tool is making money more expensive, it works by slowing down demand. That’s a blunt instrument.

There’s another way to bring prices down: produce more with less. This is the oldest anti-inflation tool there is, and it has nothing to do with interest rates. Productivity.

Here’s how a software renaissance fits in. When a small distributor builds the inventory tool it always needed, it wastes less stock. When the law firm automates its intake, the same people handle more clients. When the plumbing company schedules its crews well, it makes more calls per day. Each of those is a small business doing more with the same or fewer resources. Multiply that by millions of small and mid-sized businesses, and you have a broad productivity gain that pushes the cost of delivering goods and services down, not up.

I don’t want to oversell it. Software alone won’t tame inflation, productivity gains take time to show up in the data, and the forces behind current prices are bigger than any one technology. But if cheap, custom software lets millions of organizations do more with less, that’s a quiet deflationary force working in the same direction as the Fed, and unlike a rate hike, it doesn’t require anyone to give up growth to get there.

What to do about it

If you run a business: make a list of the work in your organization that lives in a spreadsheet, someone’s head, or a stack of email. Each item is a candidate. Then find someone who can orchestrate the AI to build it and keep it healthy, whether that’s an employee, a contractor or a small firm.

If you’re a developer: lean into being the person who can define the problem, direct the tools, check the output and own the outcome. That skill set is in demand and is likely to stay that way.

If you’re worried about all this: I understand. Every technological shift feels like an ending to those standing in the middle of it. But I’ve spent my career studying what happens next, and the pattern is rarely “the work disappears.” It’s “the work changes, and there’s a lot more of it than anyone expected.”

Linda, for the record, will be fine. She’ll finally get to take a vacation.

Image Credits: Gemini

Content Authenticity Statement: The topic area, key elements to focus on, etc. were decisions made by Braden Kelley, with a little help from Claude to clean up the article.

Sign up here to get Human-Centered Change & Innovation Weekly delivered to your inbox every week.

Innovation or Not? The Tesla Roadster with SpaceX Thrusters

1 Reply

Innovation or Not? The Tesla Roadster With SpaceX Thrusters

by Braden Kelley and Art Inteligencia

UPDATED: October 22, 2026 at 12:40PM

I have been writing about innovation long enough to have developed a reflex. When something dazzling shows up, I ask three questions: Is it new? Does it create value for somebody? And will anyone outside the launch event ever actually use it?

Then, once in a while, something arrives that makes me forget the questions entirely and just stare. The new Tesla Roadster, unveiled in Texas on October 15 with thrusters borrowed from SpaceX, is one of those. A car that hops over other cars. A car that climbs an inverted ramp, spins, and lands. I watched the clip and thought, “Well, that is certainly a thing that now exists.”

Here’s the video. Watch it first, then come back and argue with me.

What Tesla actually showed

The Roadster comes in three versions, and the names are doing a lot of storytelling.

Performance (“road ready”). Two seats. Active aerodynamics and what Tesla calls a multi-mode suction system, which uses fans to pull air from under the car. Tesla says it does 0 to 60 mph in under 1.6 seconds. This is the one you reserve: the reservation is $50,000, and Tesla hasn’t announced a price.

Thrust (“race ready”). Two seats, plus rear-mounted cold-gas thrusters. Under 0.9 seconds to 60 mph, according to Tesla’s figures. It’s a limited release, and reporting puts its reservation at $250,000.

Rocket (“air ready”). One seat. Thrusters around the perimeter of the car give it forward and upward push and the ability to rotate. Under 0.59 seconds to 60 mph. It’s invite-only, no price has been given, and Tesla’s own Roadster page says it isn’t approved for public road use.

Tesla also quotes more than 230 mph and more than 300 miles of range, though reports differ on whether those numbers apply to every version or only the top one. Production, in Musk’s words, is “2028, actually.”

On the hardware: engineering VP Lars Moravy described ten thrusters drawing on 600 liters of compressed air to lift and spin the car. And here’s the detail from the video that I love: the supply isn’t fixed. The car recharges its own air tanks, a bit like regenerative braking. It sucks so it can blow. Tesla says the thrusters produce four to five times the force the tires can handle. In Track Mode, the car lowers itself and deploys active wings, with downforce Tesla describes as approaching F1 levels. Those are Tesla’s claims, and nobody outside the company has measured them yet.

A little arithmetic, because I can’t help myself

Zero to 60 mph is about 27 meters per second. In 0.59 seconds, that works out to an average acceleration of around 4.6 g. Under 0.9 seconds is roughly 3 g. Under 1.6 seconds is a bit over 1.7 g. These are my back-of-the-envelope averages, not Tesla’s figures, but they give you the flavor. A tire on pavement tops out well below the top of that range, which is exactly why the thrusters matter: they push on the air, not the road.

What this could mean for racing

This is where I got genuinely excited, and then immediately got to asking questions, because the interesting part isn’t the stunt. It’s the principle.

For a century, a race car has been limited by the same thing: how much grip four small patches of rubber can deliver. Engineers have spent that century finding ways to get more of it, with wings, ground effects, tire chemistry, and clever suspensions. A thruster that adds force without relying on the tires is a different category of answer to the same problem.

So the questions I’d be asking if I ran a sanctioning body:

Is it a car or an aircraft that touches the ground sometimes? Racing rulebooks are written around assumptions about what a car is. This quietly breaks several.
Does thrust help braking and cornering, or only launches? A launch is a spectacle. Lap time is what wins.
What happens to the tank on lap 20? Because the car can refill its own air supply, the way regenerative braking recovers energy, the question isn’t “how much do you have?” but “how fast can you recover it, and when do you spend it?” That turns race strategy into energy management, a trick that has made other racing formats more interesting, not less.
– Who gets to use it? The moment a technology like this shows up, either everyone gets it or nobody does, and neither answer is easy.

Don’t hold me to any of this. Tesla showed a demo, not a season. But “what does racing look like when grip isn’t the ceiling?” is a real question now, and it wasn’t last month.

And now, the dismay

I promised you surprise. What I actually feel is closer to dismay, and I want to be honest about why.

It isn’t the engineering. By every account I’ve read, the engineering is real and impressive, and I’d be a hypocrite to say otherwise. My dismay is about what, among all the things this talent and capital could have built, the answer turned out to be.

Consider the details. Tesla says it packed pressure vessels into every usable void in the car, and the same type of vessel has been tied to rocket explosions elsewhere. Musk said “safety is not the main goal,” and that if safety is your top priority, you shouldn’t buy the Roadster. The range is roughly half of what was promised in 2017. The project has slipped repeatedly since then. The top version isn’t road legal. And Musk himself says the car doesn’t move the needle much on revenue, which makes it a halo car, an object whose job is to make you feel something about the brand.

I understand halo cars. They’ve always existed. But there’s a difference between a halo that says “look what we can do” and one that says “safety is optional, and rules are for other people.” In a world where the most pressing transportation problems are affordability, safety and charging, building the most extreme thing imaginable, and then selling reservations for it, feels like a statement about priorities. There’s also a fair amount of speculation, including in CNBC’s coverage, that the event was partly about linking Tesla with SpaceX ahead of a possible merger. If that’s true, the car is as much a story as a vehicle.

So: innovation or not?

Here’s my verdict, and you’re welcome to disagree.

Is it new? Absolutely. Putting cold-gas thrusters on a car, and making them controllable enough to land a spin, is a genuine invention.

Does it create value? For a handful of people, certainly. For everybody else, the value is the attention it earns and the questions it raises.

Will it reach people? The Performance version, maybe, in 2028, if you have $50,000 to reserve. The versions that do the flying, probably not on any road you drive.

My verdict is that it’s a remarkable invention and a so-far unproven innovation. An invention becomes an innovation when it creates and captures value at scale, and nobody can say yet whether that will happen here. But the technology underneath, force that doesn’t depend on tire grip, is worth watching long after the Roadster itself has been forgotten or perfected.

Then again, I said the same thing about every flying car I’ve ever seen. Ask me again in 2028.

What do you think? Innovation or not? Tell me in the comments.

SPECIAL UPDATE – Check out the in-depth TOP GEAR interview with the lead engineer and designer walking through all of the specifics of the design, engineering, and trim levels:

Image Credits: Microsoft 365 CoPilot

Content Authenticity Statement: The topic area, key elements to focus on, etc. were decisions made by Braden Kelley, with a little help from Claude to clean up the article.

Sign up here to get Human-Centered Change & Innovation Weekly delivered to your inbox every week.

Mixed Reality Headsets Meet the Dead Zone

Leave a reply

Mixed Reality Headsets Meet the Dead Zone

GUEST POST from Geoffrey Moore

My colleague and social media maven, Rich Stimbra, forwards me news items that might spark a blog post, like the following a couple of years ago:

Meta has scuttled development of a high-end mixed-reality headset meant to compete with Apple’s Vision Pro starting in 2027, The Information reports, citing anonymous sources. Production cost was a primary factor, since Meta was using the same micro OLED display tech as Apple, but trying to keep the headset under $1,000 in order to attract buyers. Apple’s $3,500 Vision Pro, meanwhile, is “stuck in neutral,” per Business Insider. Falling U.S. sales will be offset somewhat by its international launch in June, but the Vision Pro has no must-have app yet — and a more affordable model is expected next year.

This was clearly a technology adoption problem, and clients often ask members of The Chasm Group how you can tell where a given category is in the Technology Adoption Life Cycle, and how should that impact your commitment to the opportunity?

One tool we use is called the Pain/Gain matrix, as illustrated in the two charts that follow. The first chart tees up a thought experiment exercise. Looking at the disruptive innovation you are seeking to bring to market, ask yourself how much gain does it deliver, and how much pain must the customer go through to get it? In so doing, don’t just consider the end user, also take into account the other constituencies that must support the purchase. This is particularly germane to B2B enterprises, so you have to squint a bit to apply it to consumer markets. That said, here is the framework:

Pain Gain Matrix

The executive sponsor is the one who has to pay for this, the process owner is the one who has to secure the productivity improvement from it, the systems owner is the one who has to install and support it, and the end user is the one who has to actually extract the benefits from it. Each one has their own pain/gain experience, so they need to be placed separately. (In consumer markets, all four roles could condense into one, although in many families they may be distributed across either one or both parents as well younger and older siblings.)

The goal of the thought experiment is for the team to reach a consensus on the most likely square for each of the four stakeholders based on the current offer in the market today. Once you have all four placed in one (and only one) square, then you can turn to the second chart, which serves as a kind of decoder ring in that it predicts where you are likely to be in the adoption life cycle and what you will have to do to be successful at that stage.

Dead Zone Matrix

Following the Technology Adoption Life Cycle around the edges of the diagram

One might expect that the life cycle would proceed from the upper right to the lower left on a diagonal path, a straight line being the shortest distance between two points. It turns out that is not the case.

For the Early Market, a high-pain, high-gain proposition is OK provided the size of the gain is massive. That is, if you can deliver a truly transformative outcome, early adopters are willing to put up with a considerable amount of pain (and expense) to get it. However, if it is anything less than truly transformative, then they won’t play. That is why the winning Early Market operating model is a bespoke project wherein you and they put serious skin in the game, including assigning your best and brightest to the effort and simply refusing to stop until you have got the thing all the way to bright. The key to this stage is to have a visionary executive sponsor who will commit not only to fund the outcome but to continue to support it through its struggles until it gets all the way to the finish.

For the Bowling Alley (including the critical crossing the chasm step) the critical issue is still very much the size of the gain, what we call the compelling reason to buy, but the ability to absorb pain is much less. Here we have a pragmatist process owner who is under direct pressure to fix a broken mission-critical process that they cannot remediate with conventional solutions. They are desperate for a solution to their problem, what we call the whole product, which will be a huge gain for them. However, they do not have the resources to fund a bespoke project, which means they cannot absorb a great deal of pain. As a result, the vendor has to bring the whole product to them, partners included, and keep everyone committed to stay the course until the fix is up and running and proven successful. The process owner is willing to pitch in here, and so can absorb some pain in order to get to their goal, but nowhere near as much as the Early Market visionary executive.

For a Tornado to form, two things must change. The first is the emergence of a killer app that makes everyone and their mother want the new thing—that’s the big gain that has broad appeal. The second is that there has to be a major reduction in pain—a big drop in the friction that has been holding the category back to date. The goal is to sell a stripped-down, friction-free killer app that has a super-short time to value and then expand over time to address adjacent opportunities that may require more development but which can be readily absorbed because the core product is already installed. During the Tornado proper, particularly at its outset, it is the systems owners who are under the gun to procure and deploy the killer app as fast as possible. Their speed is gaited by whatever friction is still present, so the more this can be removed, the faster the tornado can spin up and the further it can spread. The company that does this best typically takes the field, becomes installed as the category gorilla, and generates outsized returns for many years to come.

Finally, when technology adoption comes to Main Street, we are now in the expand part of the land-and-expand playbook, where the gaiting item is how many different use cases it can support in how many different contexts and whether the offer can be converted from a licensable product to a consumable service. Under such conditions, because the amount of pain to adopt has been reduced to a bare minimum, it takes less and less gain to drive additional market growth. The end users are the ones who most benefit, and success rides increasingly on securing distribution channels that get to them as directly as possible and making it as easy as pie for them to try and buy, or at least try and then get their system owner to buy.

The Dead Zone

As you can see, the Life Cycle follows a path from the upper right to the lower left of the diagram by skirting around the edges. What about all the other squares?

Well, the ones that are blacked out are simply non-starters. The pain exceeds the gain to the point that no one other than a technology enthusiast would want to engage. But the four squares in the very center of the diagram are the ones that are the most dangerous. Collectively they make up the Dead Zone, which is another way of seeing the chasm, or what Intel used to call the Valley of Death.

Products in the Dead Zone deliver good (but not amazing) gains, but they do so at the cost of real (although not excruciating) pain. This combination is the kiss of death. It causes prospective customers to express real interest in the offer (“We’d love to take a meeting and see a demo”), but it does not motivate them sufficiently to get them to actually buy. It is hard to imagine a greater waste of go-to-market resources than that. Start-ups, as well as in-house incubations who venture out into such a desert usually die of thirst, meaning they don’t get that life-supporting next round of funding.

It is kind of like a LaBrea Tar Pit that swallows up whole species, as witnessed by the Sharp’s Personal Digital Assistant, RCA’s 8-track tape, and Mazda’s rotary automobile engine, as well as the . . .

Mixed Reality Headset

This sucker has Dead Zone written all over it. And because it is being positioned as a consumer product, it has a much tougher row to hoe. There is simply no chance of it getting traction unless and until someone can bring to market a killer app at a consumer price point, which will require some very deep pockets to get there, not to mention incredible design skills to get it right.

A much slower but more achievable path would target a B2B customer base with Bowling Alley applications that address urgent problems with mission-critical business processes. An obvious example relates to the “aging out of the workforce” problems that industrial manufacturers are facing, both in the factory and in field services. There simply are not enough trained technicians to meet the need, especially now that many of these industries are moving back onshore. Mixed reality headsets can be used to enable just-in-time training for onboarding inexperienced workers better, faster, cheaper, and even more importantly, safer than conventional alternatives.

Is this a killer app? Heck no, but it is a start; it can support a higher price point, and it can create enough volume to warrant investing in lower-cost approaches. What it cannot do is be rushed. The biggest cause of Dead Zone casualties is a management team that thinks a tornado is upon them when in fact it is just a Hype Cycle hitting its peak.

That’s what I think. What do you think?

— Image credit: Pexels, Geoffrey Moore

Sign up here to join 17,000+ leaders getting Human-Centered Change & Innovation Weekly delivered to their inbox every week.

Is Your pDoom Less Than Zero Yet?

Leave a reply

Is Your pDoom Less Than Zero Yet?

by Braden Kelley and Art Inteligencia

Every statistician who just read that headline winced. Probabilities don’t go below zero, and they’re right. Hold that wince for a moment, because it points at something the whole conversation about superintelligence keeps skipping.

If you haven’t run into the term, pDoom is the probability you personally assign to artificial intelligence ending badly for humanity. People trade their pDoom numbers at dinner parties now, the way people used to trade fantasy football picks. It’s a useful little conversation starter. It’s also a lopsided one, because it only counts one column of the ledger.

Nobody asks for your pDoom of inaction.

Press play first

I built this article around a music video, because some arguments land better with a hook in them. Here it is: Nothing Went Foom! from Bright Mirror.

That title is a wink at one of the great nerd arguments of the last twenty years. “Foom” is the sound an intelligence explosion makes in the fast-takeoff story: the machine gets a little smarter, uses that to get a lot smarter, and before anyone finishes the meeting it has left us standing on the platform. The idea was batted around in the Yudkowsky and Hanson debates back in 2008 and has been living rent-free in our collective imagination ever since.

I love that the title makes the point that nothing went foom. Technology almost never works that way. Electricity took decades to rewire the world. The internet took decades more, and we are still arguing about what it did to us. Transformations arrive in increments, which sounds boring until you realize what increments give you: time to watch, time to measure, time to adjust, and time to build the guardrails while the thing is still being built. Gradual is where safeguards live.

What we’re actually talking about building

I get tired of conversations where superintelligence is only ever the villain, so let me describe the other movie.

We already have a preview. AlphaFold predicted the structures of hundreds of millions of proteins, a job that used to cost a doctoral student years per protein, and its creators shared a Nobel Prize in Chemistry in 2024 for it. That is what narrow, early, imperfect AI did to one corner of biology. Now scale the ambition.

A superintelligence, pointed carefully at our hardest problems, could plausibly help us:

Compress drug discovery from a decade to a few years, and take a serious run at diseases we’ve written off as incurable.
Design new materials for batteries, solar cells and carbon capture that human intuition would never stumble across.
Help tame fusion, where the physics is understood but the control problem is brutal.
Put a patient, personalized tutor in front of every child on earth, including the ones who live hours from the nearest decent school.
Bring specialist-level diagnostics to places that have never had a specialist.
Speed up science itself, which is the part that makes my futurist heart skip. When the engine of discovery accelerates, every other list gets longer.

I’m not promising any of it will happen. I’m saying it could, and that “could” is worth a lot of human lives and a lot of human potential.

The risk we never put on the spreadsheet

Here’s where the less-than-zero joke earns its keep. The standard pDoom calculation asks how likely it is that we build this and it goes wrong. Fair question. But look at what’s missing: the cost of every year we don’t build it, or build it late, or build it only in the places that don’t care about safeguards.

A disease that goes uncured for another decade is a cost. A grid that stays dirty for another decade is a cost. A child who never gets the tutor is a cost. Those costs are quieter than a robot uprising, so they never show up in the headline, but they are real people. If you only count the risks of acting and never the risks of not acting, you haven’t done risk analysis. You’ve done fear analysis.

So here’s my proposal for a better number. Take your pDoom from moving forward, then subtract the doom you are already accepting by standing still. I suspect that for a lot of people the net figure is a lot lower than they thought, and for some of us it’s flirting with the territory of the headline. Statisticians, you may now exhale.

Safeguards, of course

Now let me be extremely clear, because I know how this paragraph gets quoted without its neighbors: I am not arguing for recklessness. Brakes are what let you drive fast, and seatbelts didn’t kill the automobile, they saved the automobile.

Responsible innovation means staged deployment instead of big-bang launches. It means independent testing and red-teaming, transparency about what these systems can and can’t do, real accountability when they fail, and international cooperation, because a safeguard that only half the world adopts is a suggestion. It means keeping humans in the loop for the decisions that matter, and building systems whose values we can inspect rather than hope for.

And I’ll say this plainly: thoughtful, serious people believe the danger is real, and some put meaningful odds on catastrophic outcomes. They might be right. That’s exactly why the safeguards have to be real and not decorative, and why dismissing the worriers is as unhelpful as dismissing the optimists. The goal isn’t to win the argument. It’s to build the thing well.

Ask a better question

The debate usually gets stuck on “should we build superintelligence?” Treat it as a yes-or-no switch and one side has to lose. Ask instead: what would we need to see in place to say yes? What safeguards, what evidence, what governance, what tripwires that tell us to slow down?

That question has answers. It also has owners and deadlines, which a vibe never does.

So, how’s your pDoom?

Mine is not zero, and I would be suspicious of anyone whose is. But when I count both columns, including the diseases, the climate, the classrooms and the discoveries waiting on the other side of responsible progress, I find the biggest risk we face isn’t that we build this thing carefully. It’s that we let fear stop us from building it at all, and hand the future to whoever was less cautious.

Hit play on the video again. Then ask yourself what future you’d rather get to first.

OR, is it all an EVIL PLAN?

Image Credits: m365 CoPilot

Content Authenticity Statement: The topic area, key elements to focus on, etc. were decisions made by Braden Kelley, with a little help from Claude to clean up the article.

Sign up here to get Human-Centered Change & Innovation Weekly delivered to your inbox every week.