The week in one beat
Snap, Nebraska, and Alaska in one week: executives finally say “AI” on layoff calls, courts suspend lawyers for machine fiction, and a state court chatbot steps back before it misstates the law again. The transformation is visible — and nobody’s pretending it’s theoretical anymore.
Week 1 · Apr 14–20
This Week in Dysfunction
Snap lays off 1,000 people and actually says the quiet part out loud
On April 15, Snap axed 1,000 employees and closed over 300 open roles — about a quarter of its planned workforce. The headline was the headcount. The story was the reason. CEO Evan Spiegel told employees that AI now generates more than 65% of Snap’s new code, and small teams with AI tools can do what large teams used to. Stock went up 11%. Estimated annual savings: $500 million. Every other company’s board meeting this month will feature this slide.
Historic precedent
Nebraska lawyer: “I was on my anniversary, my computer broke.” Court: suspended.
Omaha attorney Greg Lake submitted an appellate brief in a divorce case with 57 defective citations out of 63 — including 20 AI hallucinations: fictitious cases, fabricated quotations, nonexistent statutes. When the justices asked how it happened, he blamed a broken computer and an uploaded wrong draft. The Nebraska Supreme Court found his explanation “lacks credibility” and suspended him indefinitely. U.S. courts have now imposed at least $145,000 in sanctions on attorneys for AI citation errors in Q1 2026 alone. The AI did the crime. The lawyer did the time.
Hall of shame
92,000 tech jobs gone in 2026 — and AI is the stated reason in almost half
The layoffs.fyi counter hit 92,000 tech workers cut so far this year, approaching 900,000 since 2020. What’s different in 2026: roughly 47.9% of cuts are now directly attributed to AI automation by the companies announcing them. That’s a new number. Oracle, Atlassian, Amazon, Nike — all in the same cycle. Entry-level and generalized IT roles are contracting fastest. AI engineering roles remain the one sector where salaries are rising.
Structural, not cyclical
Tool Reviews
Claude Code (Anthropic) — In the Wild
Agentic coding · Terminal-based · Production deployments
Anthropic’s own Boris Cherny confirmed in January that “pretty much 100%” of code written inside Anthropic is now AI-generated. Claude Code is the tool doing that work. In production, it’s legitimately good for multi-file refactors and tasks where context needs to persist across a codebase. The terminal-native interface is a feature, not a bug — it lives where engineers live. The ceiling: it still needs a human who can tell if the output is architecturally sound. Snap’s 91% longer PR review times for AI code are a real warning: the bottleneck has moved from writing to reviewing, and most orgs haven’t restructured around that yet.
Deploy-readiness: ★★★★☆
Alaska Court’s AI Legal Chatbot — Public Legal Aid Pilot
Civic AI · Self-represented litigants · Pulled back
The Alaska Court System quietly scaled back its AI chatbot after it struggled with the nuances of state-specific law and delivered inconsistent guidance to self-represented litigants. “Hallucination of legal facts” was cited as the specific risk. They’re reverting to human-verified resources until further refinement. Filed under: great idea, premature deployment. The Nebraska lawyer and Alaska courthouse are two data points in the same week. The pattern is legible.
Deploy-readiness: ★★☆☆☆
The week in one beat
Big Tech earnings week. Mass layoffs tied openly to AI for the first time. A lawyer suspended for hallucinated citations he claimed never happened. OpenAI chasing ad money. Two weeks of the transformation eating itself alive — and the numbers are finally honest about it.
Week 2 · Apr 21–30
This Week in Dysfunction
Meta spends $20 billion in one quarter on AI infrastructure. Then lays off 8,000 people.
Meta announced Q1 capex of $19.84 billion and raised its full-year guidance to $125–145 billion. In the same week: 10% workforce reduction, ~8,000 jobs, effective May 20. The memo cited “running more efficiently” and “offsetting other investments.” The investments are the data centers. The efficiencies are the people. Shares fell 5–7% after hours anyway, because revenue growth guidance for Q2 was flat. The market wanted faster. The employees got the answer first.
Maximum cognitive dissonance
OpenAI wants your attention — and is building the pipes to sell it
OpenAI is now projecting $2.5 billion in ad revenue for 2026 and up to $100 billion annually by 2030. An early ad pilot generated $100M annualized within two months. The pitch is that chatbot ads are premium real estate because users state explicit intent — you don’t infer the need, the user says it. The product risk: ads and trust don’t mix well in AI assistants. The moment a user wonders if the recommendation is sponsored, the utility model collapses. Anthropic’s consumer products remain ad-free. For now, that’s a differentiator.
Watch the trust erosion
Sam Altman’s house was firebombed. Twice. In one week.
In early April, the San Francisco home of OpenAI CEO Sam Altman was attacked twice in a single week. A 20-year-old man was charged with traveling from Texas to SF, firebombing the residence at night, then showing up at OpenAI HQ trying to break through the front doors — with a manifesto naming AI executives as targets. Altman’s response: “The fear and anxiety about AI is justified… power cannot be too concentrated.” The most remarkable part is that the CEO of the most powerful AI lab on earth said that out loud, publicly, and meant it.
The vibe has shifted
Tool Reviews
GPT-5.5 (OpenAI)
Frontier reasoning · Agentic tasks · “Super app” positioning
OpenAI dropped GPT-5.5 this week framing it as a step toward a unified “super app” — ChatGPT, coding tools, and browser capabilities in one interface. Reasoning and speed improvements are real and measurable. The super app framing is telling: OpenAI is betting on consolidation rather than specialization. The concern from an engineering deployment standpoint: rapid release cycles mean the model you build on in April isn’t the model you’re running in July. Stability is underrated.
Deploy-readiness: ★★★★☆
Adobe CX Enterprise (formerly Experience Cloud)
Martech platform · Persistent AI agents · “Coworkers”
Adobe killed the Experience Cloud brand and replaced it with CX Enterprise — built around persistent AI agents called “Coworkers” that orchestrate tasks across creative and marketing systems continuously. It is a real architectural shift, not a rename with new buttons. The problem every enterprise buyer will hit: governance, data quality, and ROI measurement are all still the user’s problem. Adobe ships the agent; you figure out what it’s allowed to do. Promising for mature marketing orgs. Dangerous for ones that still have messy data pipelines.
Deploy-readiness: ★★★☆☆
Market Signals — Earnings Week Edition
All four Mag 7 companies beat estimates on April 29. All four saw stock reactions that were mixed-to-negative. The gap between “beat expectations” and “met the vibe” has never been wider.
| Company | What they said | What it meant |
|---|---|---|
| Alphabet | “Google Cloud up 48% in Q4. Accelerating.” | The one company where AI spending is visibly translating to revenue. Profit up 81%. Shares +6% after hours. This is what winning looks like right now. |
| Microsoft | “AI business at $37B annual run rate, up 123% YoY.” | Beat estimates. Shares dipped. Nadella used the phrase “eval-max their outcomes in the agentic computing era.” That sentence is not a product. Investors noticed. |
| Meta | “Advantage+ AI targeting is working. Capex raised to $125–145B.” | Beat on revenue. Q2 growth guidance was flat. Raised capex by $10B in the same breath. Shares fell 7%. The market is asking: at what point does the spend have a ceiling? |
| Amazon | “AWS re-acceleration. Bedrock adoption growing.” | AWS hit $37.6B, beating estimates. EPS of $2.78 vs $1.63 expected — a monster beat. Tariff exposure on retail side caused hesitation. The cloud story is clean. The rest is noise. |
The actual signal buried in every call
Combined 2026 AI capex for these four companies alone: ~$650 billion — larger than the GDP of most countries. Every earnings call this quarter was really a supply chain call: is compute capacity keeping pace with demand? Azure said no — demand exceeds supply. Google Cloud suggested yes — and it’s the one with the best results. That asymmetry is the signal. Companies that built their own silicon (Google TPUs, Meta MTIA) are starting to separate from those renting it.