AI.

The New Canadian Dream

Karpathy walked back vibe coding. Aesthetics, judgment, and taste were always the load-bearing part.

Six months later, the position between the watch and the EHR is still unowned. The Plaid analogy is the bet structure.

Gas Town landed for the developer. The operator's bench is still tmux and a notebook.

OpenAI's S-1 is the governance doc. Read it before the next contract.

The agent-handoff failure mode Air Canada paid $650 to document publicly.

The agent stack did not collapse. The wrong prediction collapsed. Salesforce has the receipt.

On the AGI-by-2027 claim: DeepSeek's $5M training run made the debate structural, not theatrical.

AI's capital gravity is pulling investment away from every adjacent tech sector.

When your AI becomes a design partner, the design-team org chart collapses the same way the dev team's did.

The AI-in-everything wave broke earlier than consensus expected. McDonald's and Klarna are the receipts.

The AI-for-prior-authorization pilots are the CMS mandate bet. Here is how to classify them.

The AI-generated travel itinerary product was always going to be the booking funnel. Expedia's 10-K confirmed it.

AI for hospital staffing is the only optimization that addresses both the burnout data and the org-design problem.

AI in airline operations: where the unsexy money is.

AI in claim adjudication is the under-discussed health-AI win the prior-auth mandate is about to mainstream.

The AI-in-corporate-travel category is the second-most underbuilt.

Revenue management is now an inputs problem.

AI radiology is real. AI-in-radiology-as-product is mostly not. The FDA data explains the gap.

Amazon cut 14,000 corporate roles and blamed culture. The receipts show it was the manager ratio math.

Amazon is doing to healthcare access what Booking did to travel. Here is how it ends.

California AB 1018 is the automated-decisions safety act the hiring-AI vendors didn't see coming.

California SB 53 is what SB 1047 would have become. The operator compliance posture is the same.

Canada just funded 10,000 AI scribes. The scribe is the wedge. The workflow layer is the business.

Canada's AI regulatory vacuum is not a feature. It is a liability at the investor table.

What 'Claude for Education' tells us about Anthropic's roadmap.

Colorado AI enforcement is active. Three health-AI bills converged. The patchwork accelerated.

Colorado passed the first comprehensive US state AI law. Operators in travel and health need to read it.

The cottage was already the office.

The Deel/Rippling lawsuit is the first corporate espionage case where the evidence is AI-generated data trails.

The demo video died when Google AI Overviews recommended eating rocks. Here is what replaced it.

DOGE is the largest public-sector AI displacement test in history. The operator lessons are transferable.

Duolingo cut contractors in January 2024 and called it AI. The actual AI memo came 15 months later.

The eval is the product. The Texas AG healthcare AI settlement is the consequence of ignoring that.

The FDA just shifted from product approval to process approval. Your AI-cleared moat got drained.

The Fed's 50bp cut changed the cost of capital for AI infrastructure. Here is what that means for your build-vs-buy math.

US federal AI regulation is now a four-year on/off cycle. Operators who built for the EO are overspending.

Founder Mode is not a management style. It is a reaction to AI making middle management structurally visible.

Healthcare-AI procurement is its own skill. The Texas AG settlement proved why.

IBM replaced hundreds of HR staff with AI and confirmed the back-office pattern is structural, not experimental.

Illinois banned AI therapy. The prohibition names the liability every mental-health startup avoided naming.

Itinerary rebooking is the right shape of agentic travel. Expedia finally agrees in its 10-K.

Local models are real. Local-models-as-product mostly are not. DeepSeek is the dividing line.

The MCP gambit, two years in: Claude Cowork is the product built on the bet.

On the morning routine I stopped having.

The 'multi-agent' marketing flag appears whenever Shopify-style memos run ahead of actual deployments.

Newsom vetoed SB 1047. California AI regulation will return and it will look different when it does.

The next wave of foundation models will get the deployment layer wrong. The capex cycle guarantees it.

NYC Local Law 144 enforcement exposed a compliance gap. Most hiring-AI vendors failed the audit.

The open-source agents will win the long tail. EU GPAI enforcement will accelerate it.

OpenAI o1 introduced slow thinking. The operator question is where slow thinking earns its 4x premium.

OpenAI Operator is the demo of what every travel and healthcare booking flow needs in 18 months.

Specs are the new prompt.

The retraining-cycle question your AI vendor cannot answer, and what that tells you.

The 'rich agent / poor agent' divide is coming.

The routing layer in healthcare AI is bigger than the diagnosis layer. Six million unattached Canadians prove it.

What the safety teams got right in 2024 shows up in the Claude Constitution and the 36-AG pushback.

The Techno-Optimist Manifesto is the document that made every operator choose a side.

Texas passed TRAIGA. Three states now have comprehensive AI laws. The patchwork is the policy.

The long tail is what an agent does that an agent can't.

Tool use is the only benchmark that matters after Claude 4 can code for hours unsupervised.

On Toronto winters, briefly.

The Trump election reset the US AI regulatory calendar. Canadian operators should not breathe easy.

I trust eval results from companies that do not sell the model being evaluated. The Texas AG settlement is why that matters.

UnitedHealth invested $1.5B in AI. No Canadian payer invested anything equivalent. That gap compounds.

The 'we use [foundation model X] under the hood' admission.

What the year of vibes actually changed: vibe coding is a real productivity shift, not a discourse moment.

30% gone to AI Overviews. The OTA equity story repriced.

55% / $36.6B / six IPOs. The number every health-AI deck cites is the same number that calls the top.

Agent deployment crossed 50%. The skills gap is the ceiling.

The agent is not the user. Stop building UIs that pretend it is.

Why the agent layer skipped the orchestration startups.

The agent runtime will be the operating system, not the model alone.

The agent-to-agent commerce question.

The "agents aren't production-ready" take aged badly. The 54% receipt.

'Agents eat SaaS' is a clean thesis. Most enterprises can't build a clean thesis.

AI deployment is a data-engineering problem in a model-engineering costume.

AI-driven 'patient journey' tools mostly aren't.

AI for pharmacy operations is undersold.

Why the 'AI killed the search engine' framing is premature.

AI Overviews don't kill clicks. They kill the attribution chain that monetized the click.

AI pricing creates two consumer outcomes. Regulation only sees one.

AI doesn't read your hotel brand. It reads your API.

AI scribes and the documentation-burden question.

AI slop is in your procurement pipeline. Audit before you debate.

AI travel-planning crossed 50%. The OTA is the checkout now.

AI triage in the ED keeps not landing. The deployment gap is the reason.

AI workers are unfortunately slowed down by law. At least in some domains.

Airbnb made AI the fourth pillar. Specialized beat general where trust was scarcest.

Ambient scribes are the one healthcare-AI category where evidence and adoption moved together.

Ambient scribes were never going to be a product category.

Amodei's compressed century stops at deployment.

Amodei skipped the deployment timeline.

Anthropic open-sourced the plumbing. Your roadmap just got shorter.

Anthropic ships every two weeks now. Your eval window just got eaten.

Aschenbrenner cuts both ways now.

Aschenbrenner wrote a forecast. The market read a budget.

What the biotech foundation models will look like.

Booking became underwriting last winter.

Canada is 44 of 47. The number is the citation, not the country.

Canada tried again. The U.S. 2016 playbook now applies.

Why I cancelled my GitHub Copilot in 2025.

The chatbot was an employee. Air Canada paid $650 for that ruling.

The clinical-decision-support liability question.

Cluely raised $5.3M for a posture. The posture became the category.

Copilot crossed 100M. Your TMC just became substitutable.

The cost-per-token graph is not the cost-per-task graph. The capex numbers prove it.

DAX Copilot is a CFO product.

The defensible asset isn't the drug, it's the dossier.

Denial works because appealing is expensive. Claimable inverts that.

DMA and Google AI agree on one thing. The OTA is the spread to compress.

The drug commoditizes. The data doesn't.

The EHR bundling clock starts when the wedge works.

Energy is the new chip. The grid is the four-year overhang.

The essay isn't a forecast. It's a discourse-shaping instrument.

Even the AGI bulls are quietly long adoption risk.

Expedia's 10-K named agentic AI. The board owns the threat now.

Most enterprises can't build a clean thesis. Expedia builds 1,500 agents a year.

Famous or indispensable. Travel platforms have to pick one.

The fine-tune is dead. Long live the prompt engineer.

The first wave of agent failures will look like infrastructure failures.

Founder Mode, Chesky version. Airbnb is the receipt.

Founder Mode is a permission slip. Manager Mode is a default.

Four bolts went unrecorded. The CMES record-keeping was the safety control.

Goldman published the bear case. Every AI deck answers to Covello now.

Healthcare AI adoption isn't a growth story. It's a redistribution schedule.

Human-in-the-loop is the only AI promise that Klarna had to honor.

The imaging RCTs are unimplemented. The LLMs without evidence are everywhere.

The single industry that's invisible to current AI tooling.

The Intelligence Age is the next round's pitch.

Intelligence got cheaper. Judgment didn't.

LLMflation is good news for everyone except AI-health startups.

The 'long-context window' was always a transition primitive.

Long-stay travelers are an unbuilt platform tier.

MAI-DxO is the procurement question now. Where do physicians + tools beat it?

MAI-DxO reset the room. Position your deck against it.

Microsoft leased a nuclear reactor. The grid became the new chip.

When the model is free, the EHR is the moat.

The NYT failed the blind test publicly. The trust stack failed privately.

The next travel category is opacity-as-a-service.

OpenEvidence is Nadler's second act. 1M physician questions a day.

Opus 4.7 ships with design. Your B2B moat just got reshaped.

The OTA endgame is being the agent's vendor, not the user's.

The OTA isn't a destination. It's a click-out tax.

The system that caused overtourism is being asked to fix it.

Per-token pricing isn't a billing change. It's a product-architecture verdict.

Pick a calendar. Defend a budget. AI 2027 or Normal Tech.

Pick a substrate. The decade's moat is downstream.

The post-app stack, sketched.

The post-search internet, briefly.

Provider AI runs ahead. Payer AI lags. Their interop runs slowest.

The reasoning model wars are about evaluation, not capability.

Software 3.0 is a mental model. Don't ship it as a roadmap.

Stargate is industrial policy. AI capex moved up a floor.

Stop benchmarking on benchmarks the lab made.

The synthetic-data debate is a distraction.

The model card no one reads is now the governance document everyone will require.

The post-Stack Overflow developer exists.

Topol named the longitudinal-data wedge.

The travel hacker is now a vintage profession.

Two travel companies are quietly shipping agentic features. One will call it existential.

United deployed AI to catch skiplaggers. The Texas jury said skiplagging is legal.

The unowned position is between the watch and the EHR.

Urgency-led conversion is a leading indicator. The retention haircut is the trailing one.

Urgent care has a competitor with no CAC.

Vibe coding was real. The walkback is the lesson.

The wearables stack is now upstream of biology.

What the open-weights community actually shipped in 2025.

Yegge burns $100 an hour in parallel. The unit economics are still the question.

Zitron called it 'subprime AI'. File the receipt.

I have always been here. Just one layer up now.

The AI content problem is not misinformation. It is that operators cannot tell what is real anymore.

I read the Claude Constitution so you don't have to. It's mostly fine.

The IDE was always a waiting room.

We have not seen magic yet.

The agent layer landed last quarter. The operator layer didn't.

I Have Always Been Here

A late new year's note about finally writing this stuff down.

Neo-nomads: Between 'Anyware' and 'Anyplace'

A 2007 master's thesis. Full text, figures, bibliography, and PDF.