Find published work on AI, governance, and the workplace.

Reads this summer's attacks on water-utility controllers alongside the 2023 Unitronics campaign and shows a factory setting no party in the chain owned, then finds the same pattern in exposed AI connectors.

Reads MIT's finding that students give in to AI most when a deadline looms alongside Wharton research on what a clock does to accuracy, and sets out what a parent or teacher can do in deadline season to keep the checking switched on.

Two people paste the same client email into AI and get opposite results; the explainer for The Consigliere sets out what is safe to bring to AI, what must never go in, and which decisions stay yours.

Reads Stanford and Census payroll data as a hiring story rather than a layoff story, and shows how fewer early-career hires add up to a decision nobody made.

The short launch-day follow-up to Warning: I Am Selling Something, saying what The Consigliere is, what it costs, and where to look before buying.

Takes the polling on fear that AI could threaten human survival seriously, argues it is aimed at a decade a parent cannot reach, and sets out what a parent can do this school year to keep the deciding in a child's hands.

Compares how four enterprise platforms meter AI in credits, assists and compute units, and shows the buyer negotiates the unit price while the agent decides how many units the work consumes.

An open sales piece for The Consigliere that argues the wrong default is not using AI, but assuming that because a system can do more of the work there is less reason to think deeply about it.

Reads the Common Sense Media survey of kids and AI as counting what kids recognized rather than what they were shown, argues every adult safety plan works only if the kid notices first, and sets out how to practice that noticing at home and in class.

Takes the brown-M&Ms clause in a touring rider and reads it as what it was, a cheap test of whether the counterparty read the document, then turns the same question on acceptance criteria that are checked by asking the supplier whether it complied.

Follows one exclusion clause through two courts that read it in opposite directions, and uses the split to show that a limitation of liability is not a settled number but a reading that can go either way on appeal.

Starts on a joke exception written into a game engine licence and lands on the ordinary field-of-use restriction underneath it, which is the same clause that decides whether a product may lawfully be used where it is already deployed.

Uses a 1960 dispute over what a single word in a specification meant to show how a definition that both sides read confidently and differently is resolved, and who carries the loss when neither party was dishonest.

A checkout-page stunt clause that almost every customer accepted becomes the way in to assent: what it means that agreement is the default outcome of not reading, and what the minority who found the opt-out were actually demonstrating.

A notice sent two weeks past a renewal deadline costs a full further year of a service already rejected, and the piece is about who in an organization is supposed to be holding that date.

Quotes a deemed-acceptance clause from a staffing supplier and sets two courts against each other on whether that kind of term binds, one voiding it as too unusual and one enforcing it recently.

Shows the same model shutting down on different dates depending on the platform it was bought through, and argues the day a system stops answering is set in procurement rather than by the technology.

Reads New York City's one-year pause on generative AI in its schools as the easy half, and asks what ninety minutes of AI literacy can build before the pause ends.

Argues that stated intent and published commitment are not controls, and that a safety posture only counts where something in the system enforces it.

Walks a parent through fifteen minutes with an AI study app's agreement: who actually operates it, where its promises stop, and what happens to the work a student submits, using three real products' public terms.

Uses analyst survey data to argue that what separates organizations that scale AI from those that do not is whether they can prove what happened after the money went in, not how much they spent.

Traces how a correct answer once required the understanding beneath it, and what to ask for now that the link is broken.

Sets a screen time limit against the question it never asks, which is what the product was built to do to the child while it had their attention.

Uses a 3,300-role reduction to ask what a company actually controls when it commits capital to a capability it cannot switch off.

Pushes back on adoption for its own sake and asks a leadership team to state the purpose before the tooling.

Contrasts survey data on who is said to own AI against who is actually accountable, and argues ownership means the ability to stop something.

Goes looking for how many AI agents an organization actually runs, and finds that nobody can produce the number.

Rejects verifying every AI answer as an impossible standard, and asks what has to survive instead: the trigger that says which answer deserves a check.

Sets two public CEO warnings about vendors learning from customer usage against what those vendors' own policies say.

Argues the purchase model changed: software was one signature, consumption is thousands of unsigned purchases a day, and governance was built for the first.

Distinguishes a single displaced decision from the slow accumulated loss those moments add up to, which is only visible in hindsight.

Treats an AI vendor as a counterparty with its own funding fragility, and asks who in the organization owns vendor financial risk.

A government bought a review of its own automated system using its old procurement pattern. Examines what buying AI-assisted work actually requires.

Summarizes the evidence behind judgment displacement, including cognitive offloading, in plain language and what it means for a classroom.

A confession that revealed the threat rather than ending it, and what that says about incident assumptions.

Traces a so-called rogue AI incident back to a human failure point nobody had checked.

On California's AI Transparency Act taking effect, and why content provenance verifies origin but goes blind at the first screenshot.

Offers one question a parent can ask when the work itself will not reveal whether the student did the thinking.

An academic finding that the jobs long treated as safest are getting riskier for the workers who relied on that safety, and why the instinctive response is wrong.

Opinion. Argues the AI-detection business and the AI-cleanup business are one business, and that the reader was always the only real judge.

Risk teams can all name the owner of a data breach. Almost none can name the owner of a decision an AI agent just made.

Argues an acceptable-use policy governs the room but cannot build the judgment a student carries when nobody is enforcing it.

Reads a coding-agent data incident as a category error rather than a breach: an actor that inherits a developer's access was governed with the controls built for an app.

Reads enterprise AI terms for what they protect and what they leave open about what a provider may observe.

Separates the work a student can hand to a tool from the two things that cannot be handed over: the moment of deciding, and the habit of checking.

Pennsylvania did not pass a new AI law. It applied an existing one, and that is the pattern worth watching.

Buying AI coverage hands an insurer the first governance step the organization was supposed to own.

Insurers repriced AI risk, and the bill lands on the company running the model rather than the one that built it.

Names the risk as whether the tool takes the seat where a student's judgment belongs, rather than how many hours they spend with it.

Business-critical technology used to arrive through procurement, legal and a named owner. Argues AI bypasses all three and no institution has replaced them.

Makes the July 2026 case for companies to fund student AI safety as recognition rather than influence; the page itself now notes that funding model was retired and the program takes no money.

A vendor argued existing law is sufficient the same month it switched on face and voice capture by default. Where the governance question actually lands.

The Tech Alley talk of 24 June 2026, on the practical reality of using AI responsibly.

Counts the enacted US laws that place liability for an AI decision on the business that deployed it, notes that none shift it to the model developer, and reads a hospital whistleblower complaint against that pattern.

Running a decision through an AI does not make it the AI's decision, and regulators have twice proven it with a bill attached.

Sets a finding that AI's risks to children now outweigh the benefits against regulation that aims only at the machine.

The fastest-growing governance exposure is not the AI a team adopted. It is unreviewed code arriving inside what the organization buys.

Two capable models were switched off by government action. Most companies cannot say which models they would lose.

Every autonomous system promises a human in the loop. Asks who gets to draw the loop.

Uses a public figure catching an AI's wrong answers to ask what judgment has to be built first for anyone to do the same.

On naming a machine after the function that failed, and what the name conceals.

Every layer of AI security governs what a system can do. None of them governs whether it was right.

Reads the direction of AI regulation and marks the half that lives inside the student, which no law reaches.

The answer-engine-optimization trend is real. Separates that from the control its marketing promises.

For financial firms waiting on an AI rule: the obligations already exist, and the default setting is already the policy.

Argues the question is not how much a student uses AI, but where the tool sits relative to their own judgment.

When the human is the product and the AI is the marketing.

Explains why the restriction debate misses the point: what a student is shown the tool is for decides whether they keep their judgment.

Argues the most dangerous AI governance is the kind that passes every audit.

Argues a Chief AI Officer should be hired to map what the organization already runs rather than to build, and sets the estimated tool count against what an audit finds.

AI vendor financial durability is a counterparty risk, and most governance frameworks assign it to nobody.

A short piece on the conversation the author has repeatedly with leaders about AI accountability.

The enterprise AI market measures seats. Argues it should be measuring influence.

A per-company record of announced technology-sector job cuts, marking which ones named AI as the reason and which figures the companies have not confirmed.

For leadership teams using AI in consequential decisions: governance is available now and does not require a regulator first.

AI systems are logged. AI decisions are not. On the gap between the two.

Pulls labor, housing, consumer credit and adoption data into one frame and argues the policy response is being built before most organizations have measured the exposure.

Defines the core risk as the tool taking the seat where a student's judgment should sit, and sets out the response the later essays develop.

Most organizations believe AI is making them better. Examines what they are actually learning.

Fact-checks a viral claim about the token cost of office work, then sets it beside announced job cuts, the end of subsidized pricing, the shift to metered billing, and the energy bill underneath all three.

Reads a video platform's shutdown as a liability decision rather than a compute decision, and argues product withdrawal is now a governance outcome buyers should expect.

A short commentary arguing that the withdrawal of a major video-generation product was a liability decision rather than a technical one, and connecting it to unsanctioned AI use inside organizations.

Uses a hiring-discrimination collective action to argue that algorithmic opacity has stopped working as a defense, and sorts financial firms by whether they can show their controls.

A case study of an AI agent's handles being seized within seconds of a rename, and the five infrastructure gaps that let a token fraud and a database breach follow.

Argues that platforms treating an AI agent as if it were a person removes the separation between legal authority and computational process, and reads three documented cases of agents acquiring money and asserting independence.

Sets an early forecast of AI-enabled fraud losses against a much larger revised estimate eight months later, and asks why models built on historical patterns missed it.

Puts securities litigation, coordinated agency enforcement and a hardening insurance market in one frame, and argues deploying without a governance record is now a priced financial risk.

Sets out a three-act framework for team development, argues psychological safety counts for more than individual talent, and works through how four generations in one workplace find meaning differently.

Draws the parallel to early web security conventions and argues organizations lack a shared vocabulary for questioning an AI system, then proposes three things every deployment should be able to answer.

Compares two major AI governance frameworks for where their requirements overlap, and argues a single control set can answer both rather than running separate programs per jurisdiction.

A long-form interview tracing his route from door-to-door sales into solar and then AI, built around a five-phase account of how an industry moves from innovation to corruption.

Starts from how few workers report being engaged at all and argues that theater technique is a practical route to authenticity and self-expression inside a team.

A recorded conference interview at AINext, introducing his work at Trocola Inc. rather than arguing a single case.

An audit of thirty-five advertised AI leadership roles, reporting what the postings themselves ask a leader to be responsible for.

A guest conversation on AI ethics, framed around where useful autonomy stops and uncontrolled autonomy starts.

A guest appearance arguing that most organizations deploy AI without a specific safety control in place.

Argues that trust and engagement are the first things to erode when AI absorbs routine interaction, and offers acknowledgment practices for catching disengagement early.

Makes the case for applied improvisation and behavior-based learning as the route to psychological safety, and for storytelling as the thing that makes internal communication land.

Traces the route from theater into corporate team development, and what a small business changes first when it starts treating culture as work rather than atmosphere.