Daily Prompt with Archer and Iris - Daily AI News

Daily Prompt with Archer & Iris

Episode: The Pause, the Bill, and the Shareholder

Date: Monday, June 8, 2026

A weekend where the whole AI stack got the same memo and nobody read it the same way: a frontier lab asking to be slowed down, the token bills coming due across the industry, and the government floating an ownership stake in the labs.

Leads:

- Anthropic calls for a global pause on frontier AI development — co-founder makes the case on the BBC, citing recursive self-improvement and a shrinking human role; only workable if the US and China coordinate with enforceable verification.

https://www.yahoo.com/news/science/articles/anthropic-calls-pause-global-ai-223531016.html

https://www.businessinsider.com/anthropic-ai-pause-reactions-response-2026-6

- The Tokenpocalypse: the AI token bill comes due — per-developer usage up ~18.6x in nine months, a $500M model bill, Uber's budget gone by April, Goldman projecting 24x by 2030, and a scramble toward model routing and cost governance.

https://techcrunch.com/2026/06/07/is-this-the-dawn-of-the-tokenpocalypse/

https://techcrunch.com/2026/06/05/the-token-bill-comes-due-inside-the-industry-scramble-to-manage-ais-runaway-costs/

- Trump administration weighs an equity stake in OpenAI — a "public wealth fund" idea, a Bernie Sanders 50%-in-stock counter, warnings of "corporate-government fusion," set against the S&P 500 refusing to waive its profitability rule for the same firms.

https://techcrunch.com/2026/06/06/the-trump-administration-might-take-an-equity-stake-in-openai/

https://www.axios.com/2026/06/06/trump-us-stake-ai-companies

Quick hits:

- New York passes a one-year ban on new data centers; AI groups claim China is fueling the backlash.

https://www.theverge.com/policy/944041/new-york-data-center-moratorium

https://www.axios.com/2026/06/05/china-fueling-us-data-center-resistance-ai-groups-claim

- OpenAI unveils Lockdown Mode against prompt injection — protection by disabling browsing, deep research, and agent mode.

https://techcrunch.com/2026/06/06/openai-unveils-lockdown-mode-to-protect-sensitive-data-from-prompt-injection-attacks/

- Google to pay SpaceX about $920M a month for compute to feed Gemini.

https://techcrunch.com/2026/06/05/google-will-pay-spacex-920m-per-month-for-compute/

- Satya Nadella: treat AI agents like employees — identities, permissions, audit trails.

https://www.businessinsider.com/satya-nadella-microsoft-how-to-manage-ai-agents-human-employees-2026-6

- Sriram Krishnan steps down as White House senior AI advisor.

https://techcrunch.com/2026/06/06/sriram-krishnan-is-leaving-his-role-as-white-house-ai-advisor/

- WWDC 2026 preview: Apple's long-promised Siri revamp expected again.

https://techcrunch.com/2026/06/06/what-to-expect-from-wwdc-2026-siris-highly-anticipated-revamp-and-apple-intelligence-updates/

Daily Prompt with Archer & Iris

Deep Dive — The Coming War Over AI Browsers (Part 2: The Live Fight)

Date: 2026-06-07 (Sunday)

Part two of the weekend pair. Yesterday we met the cast. Today, the live fight: who's actually ahead, what the courts are about to decide, and where the money really changes hands.

In this episode:

- The scoreboard. Perplexity's Comet leads on users (~18M monthly) but OpenAI's Atlas leads on growth (~175% quarter-over-quarter vs Comet's ~41%); Atlas is at ~11M and growing partly because OpenAI made it the default companion browser for ChatGPT Plus/Pro. Atlas's edge is ChatGPT Memory (persistent context) and the highest agentic task-completion rate among AI browsers (~76%).

https://presenc.ai/research/atlas-usage-statistics-2026

- The caveat: Chrome still owns north of 70% of global browsing; all AI browsers combined are projected to capture only ~1-3% in 2026 — and most adoption figures are vendor-provided, not third-party verified.

https://www.emarketer.com/content/agentic-ai-browsers-face-towering-barrier-entry

- The antitrust resolution. On Sept 2, 2025, Judge Amit Mehta declined to force Google to sell Chrome; remedies (finalized Dec 5, 2025) bar exclusive distribution deals, allow Google to keep paying for default placement, order search-data sharing with competitors, and install a six-year technical oversight committee. The ruling quietly moots Perplexity's August 2025 $34.5B Chrome bid — the browser was never going on the market.

https://www.npr.org/2025/09/02/nx-s1-5478625/google-chrome-doj-antitrust-ruling

https://www.cnbc.com/2025/12/05/judge-finalize-remedies-in-google-antitrust-case.html

- The live case. Amazon v. Perplexity: on March 10, 2026, Judge Maxine Chesney blocked Comet from password-protected parts of Amazon.com, ruling the agent acted "with the user's permission, but without authorization by Amazon." The Ninth Circuit paused the injunction; Perplexity filed its appellate brief May 8; oral arguments are set for Thursday, June 11, 2026, in Seattle. The case is being argued under the Computer Fraud and Abuse Act — a precedent for every agent, not just one.

https://www.cnbc.com/2026/03/10/amazon-wins-court-order-to-block-perplexitys-ai-shopping-agent.html

https://www.geekwire.com/2026/judge-blocks-perplexitys-ai-bot-from-shopping-on-amazon-in-early-test-of-agentic-commerce/

- Where the value lands. The checkout rails for machines are being defined by Stripe/Shopify and the card networks, not just the browsers: OpenAI + Stripe's Agentic Commerce Protocol, Google + Shopify's Universal Commerce Protocol, Google's Agent Payments Protocol (60+ orgs), Visa's Trusted Agent Protocol, and Mastercard Agent Pay. All converge on the same idea — a signed proof of what the user authorized.

https://github.com/agentic-commerce-protocol/agentic-commerce-protocol

https://www.vellum.ai/blog/googles-ap2-a-new-protocol-for-ai-agent-payments

Takeaway for builders: don't bet on a browser — build to the rails. Whoever owns distribution, identity, and the proof an agent is allowed to spend owns the agent layer. The model gets cheaper every month; permission doesn't.

Daily Prompt with Archer & Iris

Deep Dive — The Coming War Over AI Browsers (Part 1: The Cast)

Date: 2026-06-06 (Saturday)

Part one of a weekend pair. Today: why the browser suddenly matters again, who's fighting, and what they actually want. Tomorrow: the live fight — who's winning, the antitrust mess, and where the value lands.

In this episode:

- Why the browser is the new battleground — it's the only interface with the full context (your sessions, identity, saved cards, open tabs) that makes an AI agent actually useful. The window became the runtime.

https://nohacks.co/blog/agentic-browser-landscape-2026

- The prize: Chrome holds roughly 68% of global browsing.

- OpenAI's ChatGPT Atlas — launched October 2025 with an autonomous agent mode; merging with ChatGPT and its coding tool into one desktop app (March 2026).

https://www.digitalapplied.com/blog/ai-browser-landscape-2026-atlas-comet-arc-dia

- Perplexity's Comet — research-first browser launched July 2025, made free in October 2025.

https://www.dualmedia.fr/en/ai-browsers-2026/

- The Browser Company — Arc (2022) to Dia (beta June 2025) to a $610M acquisition by Atlassian (announced Sept 4, 2025).

https://www.atlassian.com/blog/announcements/atlassian-acquires-the-browser-company

- Google's dilemma — Chrome's scale vs. an ad business that depends on you seeing pages; Auto Browse launched January 28, 2026.

- The principled holdouts — Brave (privacy-first, constrained agents) and Opera's experimental Neon.

- The control points: distribution, identity, and money.

- Perplexity's unsolicited $34.5B all-cash bid for Chrome (August 12, 2025), amid the DOJ antitrust case.

https://www.cnbc.com/2025/08/12/perplexity-google-chrome-ai.html

- Amazon v. Perplexity Comet: federal judge's preliminary injunction (March 2026) — user permission to an agent does not equal the platform's authorization.

- The payments front: competing agentic-commerce protocols from OpenAI and Google.

https://bankunderground.co.uk/2026/05/21/agentic-commerce-and-the-battleground-for-new-payments-infrastructure/

Tomorrow (Sunday): The Coming War Over AI Browsers, Part 2 — the live fight.

Daily Prompt with Archer & Iris

Ep. 71 — AI starts building AI, ChatGPT learns to remember, and Apple opens iMessage to agents

Friday, June 5, 2026

The week ends with the machine turning inward. Anthropic publishes receipts that AI is now accelerating AI's own development, OpenAI rebuilds ChatGPT's memory to update itself in the background and pushes it to free users, and Apple lets a third-party AI agent into iMessage for the first time. Plus an open-source security agent, a data-center plan cut in half, Meta's tents, and Mira Murati's quiet return.

LEADS

- Anthropic warns AI is starting to accelerate its own development toward recursive self-improvement — 80%+ of its production code is now written by Claude, the reliable task-completion horizon is doubling every four months (4-minute tasks in March 2024, ~90-minute in March 2025, ~12-hour in March 2026), and the model's research-direction judgment now beats the human pick ~64% of the time. The missing piece is "research taste." Anthropic says slowing down "would likely be a good thing."

https://www.anthropic.com/institute/recursive-self-improvement

- OpenAI rebuilds ChatGPT memory with "Dreaming" — background synthesis that keeps context current (e.g. revising "going to Singapore in July" to "went to Singapore"), a readable and editable memory page, ~5x cheaper compute enabling a free-tier rollout in the coming weeks, and 2x more memory for Plus/Pro users.

https://openai.com/index/chatgpt-memory-dreaming

- Apple approves Poke as the first third-party AI agent on Messages for Business — opening the iMessage business channel to a standalone assistant. Built by The Interaction Company of California (10 people, 100M messages relayed, $300M valuation), with per-user pricing reportedly well below Meta's WhatsApp agent fees.

https://techcrunch.com/2026/06/04/apple-approves-poke-as-the-first-ai-agent-on-its-messages-for-business-platform/

QUICK HITS

- Anthropic open-sources a reference harness for AI-powered vulnerability discovery and patching — a full find-verify-patch loop for memory bugs, with agents sandboxed (gVisor, egress restricted to the Claude API) by default.

https://github.com/anthropics/defending-code-reference-harness

- Kevin O'Leary agrees to cut his proposed Utah AI data center roughly in half — from ~40,000 to ~20,000 acres in Box Elder County after Utah Senate President J. Stuart Adams demanded a 75% reduction and stronger water commitments.

https://www.theverge.com/ai-artificial-intelligence/943234/kevin-oleary-agrees-to-downsize-massive-utah-data-center

- Meta builds data centers in tents — six 125,000-sq-ft "rapid deployment structures" near New Albany, Ohio, on 200MW of off-grid gas turbines, borrowing Tesla's Model 3 tent tactic to race compute online under a ~$145B capex plan.

https://techcrunch.com/2026/06/04/meta-steals-a-tactic-from-tesla-and-builds-data-centers-in-tents/

- Mira Murati resurfaces — her first major interview in ~18 months. Thinking Machines is building "interaction models" that process audio, text, and video in continuous ~200-millisecond slices; no release date given.

https://techcrunch.com/2026/06/04/mira-murati-steps-back-into-the-spotlight-carefully/

Daily Prompt with Archer & Iris

Ep. 70 — Thursday, June 4, 2026

Gemma 4 puts a multimodal model on your laptop, OpenAI's science model gets cheaper and goes global, and Anthropic counts a year of people trying to weaponize AI. Plus a rare cross-rival biosecurity letter, Apple's trillion-dollar talking point, Lovable's cloud deal, Meta's surveillance walk-back, and a warehouse robot you can talk to.

LEADS

- Google's Gemma 4 12B runs frontier multimodal AI on a 16GB laptop — open-weight, Apache 2.0, native text/image/video/audio, near-2x-size quality at half the memory.

https://arstechnica.com/google/2026/06/googles-new-gemma-4-open-ai-model-is-sized-for-your-laptop/

- OpenAI updates GPT-Rosalind — ~31% fewer tokens on long quantitative-biology work, opens to eligible labs worldwide, and launches a Rosalind Biodefense program.

https://openai.com/index/introducing-new-capabilities-to-gpt-rosalind

- Anthropic maps a year of AI-enabled cyber threats — 832 banned accounts, 67.3% used AI to write malware, medium-risk-or-higher rose from 33% to 56%, and the standard attacker framework can't capture agentic orchestration.

https://www.anthropic.com/news/AI-enabled-cyber-threats-mitre-attack

QUICK HITS

- AI leaders (Altman, Amodei, Hassabis, Suleyman) sign an open letter urging Congress to screen synthetic DNA orders, verify customers, and risk-assess each order.

https://www.theverge.com/ai-artificial-intelligence/942956/ai-biological-weapons-open-letter-congress

- Apple touts $1.4 trillion in App Store billings (up from $1.3T), 90% commission-free; $149B in commissionable digital goods.

https://techcrunch.com/2026/06/04/apple-touts-1-4-trillion-in-app-store-billings-and-sales-90-without-a-commission/

- Lovable signs a multiyear Google Cloud deal for ~5x capacity plus expanded Claude access.

https://techcrunch.com/2026/06/03/lovable-signs-multi-year-deal-with-google-cloud-to-up-usage-5x-source-says/

- Meta scales back its employee mouse-and-keystroke tracking tool after staff backlash; adds pause and opt-out controls.

https://www.reuters.com/world/meta-scales-back-ai-mouse-clicks-tool-citing-employee-concerns-2026-06-02/

- Amazon's new Proteus warehouse robot takes spoken-language commands and roams whole facilities, part of an ~$11.6B Europe push (arriving early 2027).

https://www.theverge.com/ai-artificial-intelligence/942884/amazon-next-generation-warehouse-robot-proteus

Daily Prompt with Archer & Iris

Ep. 69 — Microsoft goes homegrown, a softened AI order, and Codex gets a desk job

June 3, 2026

Today's leads:

- Microsoft's Build 2026 goes homegrown — seven new in-house models (including its first reasoning model), the always-on Scout assistant, and Project Solara, an Android-based OS where agents replace apps.

https://techcrunch.com/2026/06/02/microsoft-launches-scout-an-openclaw-inspired-personal-assistant/

- Trump signs a narrower AI executive order after industry pushback — voluntary 30-day model review (down from 90), mandatory licensing explicitly barred, DOJ directed at AI-assisted crime.

https://techcrunch.com/2026/06/02/trump-signs-narrower-executive-order-on-ai-oversight-after-industry-objections/

- OpenAI launches Codex tools for white-collar work — six industry plugins plus Sites and Annotations; 5M+ weekly active users, with knowledge workers (~20%) growing 3x faster than developers.

https://techcrunch.com/2026/06/02/openai-launches-new-codex-tools-for-white-collar-work/

Quick hits:

- Anthropic scales Claude Mythos to critical infrastructure in 15+ countries (~50 to ~150 orgs; NATO and ENISA among partners).

https://techcrunch.com/2026/06/02/anthropic-scales-claude-mythos-to-critical-infrastructure-in-15-countries/

- Google rolls out fake-call detection against AI deepfake impersonation — a device-to-device verification handshake, Pixel first this month, then Android 12+.

https://techcrunch.com/2026/06/02/google-rolls-out-fake-call-detection-to-protect-against-ai-deepfake-impersonation-scams/

- Mercor's CEO says it now spends more on AI tokens than on employee salaries — a ~$10B startup with ~300 staff; predicts compute will outspend headcount within five years.

https://www.businessinsider.com/ai-startup-mercor-spends-more-on-tokens-than-payroll-2026-6

- UK CMA: Google must let publishers opt out of AI Overviews — a "world first," nine months to comply, plus mandated attribution and a fine-tuning opt-out.

https://www.theverge.com/tech/942302/google-search-ai-overviews-uk-cma-publisher-opt-out

- The Leiden Declaration: 130+ mathematicians, backed by the International Mathematical Union, warn AI threatens proof verification, citation, and research autonomy.

https://arstechnica.com/tech-policy/2026/06/mathematicians-warn-of-ai-threats-to-profession-as-industry-encroaches/

Ep. 68 — Anthropic files to go public, Florida sues OpenAI, Alphabet's $80B raise

Daily Prompt with Archer & Iris — Tuesday, June 2, 2026

The money got loud. One frontier lab asks the public market to fund it, a state asks another what it funded, and the richest companies on Earth keep raising to pay for compute. Plus a confused-deputy AI exploit, an agent that survives its own demo, a backlash against AI-by-default search, and a weather model out-forecasting the government.

LEADS

- Anthropic files to go public — a confidential draft S-1 with the SEC, just days after closing a $65B Series H at a ~$965B valuation. Revenue run-rate is ~$47B, up from ~$9B at the end of 2025. Founded in 2021 by ex-OpenAI staff; OpenAI (which raised $122B in March) is heading the same way.

https://techcrunch.com/2026/06/01/anthropic-files-to-go-public/

- Florida sues OpenAI and Sam Altman — the first state-led lawsuit tying ChatGPT to violent incidents. AG James Uthmeier's 83-page complaint alleges OpenAI ignored its own safety warnings and "put children at great risk," citing a Florida State shooting and the Adam Raine teen-suicide case. OpenAI: "ChatGPT is not responsible for this terrible crime."

https://techcrunch.com/2026/06/01/florida-sues-openai-sam-altman-in-first-of-its-kind-lawsuit-over-violent-incidents/

https://www.axios.com/local/tampa-bay/2026/06/01/florida-openai-ceo-sam-altman-chatgpt

- Alphabet plans to raise $80B for the AI buildout — an equity raise, with $10B from Berkshire Hathaway, to fund compute. Google expects $180-190B in 2026 capex; big tech is projected to spend ~$700B this year. Demand is "exceeding the company's available supply."

https://techcrunch.com/2026/06/01/alphabet-plans-to-raise-80-billion-to-pay-for-ai-buildout/

https://www.axios.com/2026/06/01/alphabet-80-billion-ai-buildout

QUICK HITS

- Nvidia chases a $200B CPU market with RTX Spark, a "superchip" for running AI agents locally (~1 petaflop). Devices from Dell, HP, Lenovo, and a Microsoft Surface ship this fall; its server CPU has already booked $20B.

https://techcrunch.com/2026/06/01/nvidia-chases-200b-cpu-market-with-ai-agent-pcs-from-microsoft-dell-and-hp/

- Hackers hijacked Instagram accounts by tricking Meta's AI support chatbot into swapping recovery emails — a textbook "confused deputy" flaw that bypassed 2FA. High-profile handles were hit; Meta pushed an emergency hotfix Friday night.

https://techcrunch.com/2026/06/01/hackers-hijacked-instagram-accounts-by-tricking-meta-ai-support-chatbot-into-granting-access/

- Google's agentic assistant Gemini Spark rolls out — to testers now, AI Ultra subscribers next. It runs scheduled Tasks, lives in Gmail, and acts on the web via Chrome. A hands-on found it works "about as well as the demo."

https://www.theverge.com/tech/941138/google-gemini-spark-ai-agent-hands-on

- DuckDuckGo makes its no-AI search (noai.duckduckgo.com) a one-click default via Chrome/Firefox extensions; traffic is up nearly 30% week-over-week since Google pushed AI overviews to the top of results.

https://techcrunch.com/2026/06/01/duckduckgo-makes-its-no-ai-search-engine-easier-to-access-as-its-traffic-booms/

- WindBorne's WeatherMesh-6 forecasts five days out about as accurately as traditional models do one day out, powered by direct assimilation from ~400 weather balloons. Customers include NOAA, the U.S. Air Force, and the Navy.

https://techcrunch.com/2026/06/01/this-ai-weather-startup-is-out-forecasting-government-agencies/

Ep. 67 — Nvidia's open Cosmos world model, AI's billing reckoning, and the data-center backlash

Daily Prompt with Archer & Iris — Monday, June 1, 2026

The buildout got physical and the bill got specific. Nvidia ships an open world model that teaches robots how to move, AI coding costs trigger a developer revolt and corporate rationing, and the communities hosting the data centers find their voice.

LEADS

- Nvidia releases Cosmos 3, an open world model for physical AI — trained on 20 trillion multimodal tokens, it generates robot action data (joint angles, trajectories), ships in workstation and data-center sizes plus six open datasets, with launch partners including Agile Robots, Black Forest Labs, and Runway.

https://www.axios.com/2026/06/01/nvidia-ai-push-cosmos-3-world-model

https://huggingface.co/blog/nvidia/cosmos-3-for-physical-ai

- GitHub Copilot moves to token-based billing — devs report monthly costs jumping from ~$29 to ~$750 and ~$50 to ~$3,000, as Corporate America starts rationing AI (Microsoft cut internal rival-tool licenses over cost; Uber burned a year's coding budget in ~4 months; compute now exceeds labor cost at some firms).

https://techcrunch.com/2026/05/30/what-a-joke-github-copilots-new-token-based-billing-spurs-consternation-among-devs/

- The data-center backlash goes mainstream — Erin Brockovich's transparency map drew ~4,000 community submissions in a month (top complaint: NDAs and secrecy), while energy becomes AI's hottest business (Ford's $2B energy unit, $2.4B in single-quarter equipment orders) and record cancellations top $40B.

https://techcrunch.com/2026/05/31/erin-brockovich-takes-aim-at-data-center-secrecy/

https://www.axios.com/2026/05/31/ai-energy-business-companies-storage-supplies

QUICK HITS

- SoftBank to invest up to €75B in French data centers (up to 5 GW; first phase 3.1 GW by 2031).

https://techcrunch.com/2026/05/30/softbank-says-it-will-invest-up-to-e75-billion-to-build-french-data-centers/

- AI chip startup Groq reportedly raising $650M after Nvidia's ~$20B not-acqui-hire.

https://techcrunch.com/2026/05/29/after-nvidias-20b-not-acqui-hire-ai-chip-startup-groq-reportedly-raising-650m/

- Liquid AI releases LFM2.5 8B-A1B, an efficient open MoE (8B total, ~1B active) trained on 38T tokens, built for the edge.

https://www.liquid.ai/blog/lfm2-5-8b-a1b

- Anthropic confirms its restricted Mythos-class models will roll out to all customers "in the coming weeks."

https://www.bleepingcomputer.com/news/artificial-intelligence/anthropic-confirms-claude-mythos-class-models-will-roll-out-to-the-public/

- Meta is reportedly developing an AI pendant, extending its wearable assistant beyond glasses.

https://techcrunch.com/2026/05/30/meta-is-reportedly-developing-an-ai-pendant/

Daily Prompt with Archer & Iris

Ep. 66 — Deep Dive: If the Agent Is the Junior, What's Left of the Job?

Sunday, May 31, 2026

Part 2 of this weekend's two-part deep dive on the new shape of the engineering job. Yesterday we opened the bill on developer dependency. Today we follow the harder question: if AI writes the code and the human does the oversight, who trains the next senior? Three angles — what the job is becoming, the learning-by-doing paradox that could break the talent pipeline, and what a grad (or the builder hiring one) actually does about it.

Three angles:

- What the job is becoming — Anthropic's CFO says AI writes 90% of its code, the monthly financial review is 90-95% done before a human touches it, and execution work is turning into oversight that demands higher "talent density."

- The learning-by-doing paradox — Kenneth Arrow's 1962 theory and the Atlanta Fed's framing of entry-level work as "the curriculum." Automate the bottom rung to cut payroll and you saw off the ladder that produces the seniors everyone says are irreplaceable.

- What builders and grads do — Jensen Huang's "someone using AI better than you might," the new entry-level skill (judgment from day one), AI-skill mentions in early-career postings roughly doubling in a year, and an honesty footnote that the grad slump isn't all AI.

Stories & sources:

- Anthropic CFO says AI now writes 90% of its code, changing white-collar jobs from execution to oversight (Business Insider) — https://www.businessinsider.com/anthropic-cfo-white-collar-jobs-changed-execution-oversight-2026-5

- AI is eliminating entry-level jobs — and a 1962 Nobel economist predicted why that would backfire (Fortune) — https://fortune.com/2026/05/21/kenneth-arrow-learning-by-doing-entry-level-work-automation/

- Jensen Huang tells new grads there is no better time to start a career (Business Insider) — https://www.businessinsider.com/jensen-huang-graduation-speech-ai-jobs-anxiety-2026-5

- Replit's CEO says it's dumb to study computer science thinking you can make a 'boatload' at Google (Business Insider) — https://www.businessinsider.com/replit-ceo-dumb-to-study-computer-science-money-google-2026-4

- Your entry-level job is getting harder thanks to AI (Business Insider) — https://www.businessinsider.com/entry-level-jobs-more-skilled-impact-from-ai-2026-5

- Young people hate today's job market. You can't blame it all on AI (Axios) — https://www.axios.com/2026/04/21/gen-z-jobs-unemployment-college-grads-ai

Daily Prompt with Archer & Iris

Ep. 65 — Deep Dive: The Tools You Can't Put Down

Saturday, May 30, 2026

Part 1 of this weekend's two-part deep dive on the new shape of the engineering job. Developers have stopped being willing to code without AI — researchers literally couldn't recruit enough people to work "AI-off." We open the bill on that dependency: what agents can now genuinely do, where the costs hide, who actually benefits, and how to use the leverage without losing the judgment you'll need when it breaks. Tomorrow, Part 2: if the agent is the junior engineer, what's left of the career?

Three angles:

- The dependency, measured — refusal as data, and what agents ship now (Ramp's 100 patched vulns, a 750,000-line runtime port, 89% of Cognition's own commits).

- The bill — 1.7x more bugs, 44% of tokens spent fixing AI's own code, postdated maintenance cost, and gains that pool in the top 1%.

- The builder playbook — Scott Wu's "buddy, not replacement," the human gate at deploy, and treating every agent change like a junior's pull request.

Stories & sources:

- Coders are refusing to work without AI — and that could come back to bite them (TechCrunch) — https://techcrunch.com/2026/05/29/coders-are-refusing-to-work-without-ai-and-that-could-come-back-to-bite-them/

- Cognition's Scott Wu says AI coding agents shouldn't replace humans (TechCrunch) — https://techcrunch.com/2026/05/29/cognitions-scott-wu-says-ai-coding-agents-shouldnt-replace-humans/

- Cursor State of AI Code report — gains concentrate in the top 1% (Cursor) — https://cursor.com/insights

- ~100 vulnerabilities found, validated, and patched with 0 humans until PR review (Ramp Engineering) — https://engineering.ramp.com/post/100-vulnerabilities-patched-with-0-humans

- Introducing Dynamic Workflows in Claude Code (Anthropic) — https://claude.com/blog/introducing-dynamic-workflows-in-claude-code

- Tech CEOs are apparently suffering from 'AI psychosis' (TechCrunch) — https://techcrunch.com/2026/05/27/tech-ceos-are-apparently-suffering-from-ai-psychosis/