Sixteen issues, December 2025 to June 2026, graded against what actually happened by October 2026.
Latest issue · November 2026
The victory-lap issue, plus four new calls now on the clock.
As agents like Muse, Dots and Duet reach non-technical people, buying is the next decision people hand off. Winning means earning the trust of both the human and their agent.
In intent, spec, implement, verify, the last step is the hardest to get right. Formal methods move from niche to default for code that cannot be wrong.
Physics, then chemistry, biology and beyond, with the newest work concentrated inside the labs that hold the most compute.
Services shaped like Buffer, which a model can rely on instead of rebuilding, survive each new model generation.
Said before it happened.
From Second and Third Order Effects of Vibe Coding, reinforced in the impostor syndrome is creeping in.
Claude Opus 4.6 shipped agent teams on February 5. Code with Claude in May added managed multi-agent orchestration. Steve Yegge put Gas Town into maintenance mode and handed the idea to Gas City, an SDK other engineers rewrote.
Evidence: TechCrunch · InfoQ · Welcome to Gas City
From Second and Third Order Effects of Vibe Coding.
Federal prosecutors and the CFTC brought the first prediction-market insider case in April, against a soldier accused of trading on the Maduro operation. House Oversight opened a probe in May. Polymarket referred dozens of military accounts to the Justice Department in August. The Pentagon banned insider betting on October 20.
From Second and Third Order Effects of Vibe Coding.
Mike joined his company's new AI Tools team. Across the market the title is now a posting category with senior pay, and recruiters describe internal champion networks turning into formal teams.
Evidence: Betts Recruiting · Glassdoor listings
From if you're reading this, you're gonna make it.
US software postings rose about 15% after Claude Code launched while overall postings fell 7%. Senior roles drove 71% of the increase. Employment for 22 to 25 year olds in AI-exposed jobs sits 19% below their peers.
Evidence: Indeed Hiring Lab · CNBC
From Second and Third Order Effects of Vibe Coding and Compounding Taste with kAI.
The "SaaSpocalypse" hit in early February, when Claude Cowork plugins set off a software selloff of roughly $285 billion in one session. Mike's own app idea, Roxas, became a Claude skill in April.
Evidence: Tech Startups · Built In
From Second and Third Order Effects of Vibe Coding.
Lovable grew to roughly $500 million in annual recurring revenue as enterprises moved to production, then had a data exposure incident. Consultancies now sell fixed-price migrations from Lovable to AWS.
Evidence: Lovable growth · DevOpsity
From Agentic Convergence-in-Depth, first floated in Agile for Agents.
A July study found Claude reviewing Codex drafts lifted the pass rate from 71.6% to 89.7%. The reverse direction hurt, so the pairing works best one way round.
Evidence: arXiv 2607.21656
From 10 hours with Gas Town and The Fix-Forward Solutions.
Anthropic's run rate climbed from about $10 billion at the end of 2025 to $65 billion in July. Uber used up its annual AI coding budget in four months and set per-engineer caps.
Evidence: CNBC · Uber and Microsoft budgets
True when written, much clearer since.
From Life isn't about work and The Fix-Forward Solutions.
Berkeley researchers found AI intensifies work instead of reducing it. BCG's "AI brain fry" study and reports of developers losing sleep followed.
Evidence: The AI Vampire · Fortune · LeadDev
From What do Humans add to AI Tokens? Perspective.
OWASP researchers say prompt injection remains unsolved. OpenAI's guidance leans on tool permissions and sandboxing instead of training.
Evidence: Infosecurity Magazine
From Agentic Convergence-in-Depth.
Snyk found prompt injection in 36% of nearly 4,000 public agent skills, plus 1,467 malicious payloads.
Evidence: Snyk ToxicSkills
From Agentic Convergence-in-Depth and The Fix-Forward Solutions.
Verification startups raised over $400 million this year. In September, Boris Cherny used Claude and Lean on the Claude Agent SDK and got 16 bug-fix PRs from a couple of prompts.
Evidence: Kleppmann's prediction · Boris Cherny · Axiom raise
From Tokens From Our Past.
Mitchell Hashimoto moved Ghostty off GitHub after a month of near-daily outages. GitHub's CTO blamed agent-driven load and announced a 30x scaling plan.
Evidence: Ghostty is leaving GitHub
From Agile for Agents.
Claude Code now ships persistent tasks, agent-to-agent messaging and memory. The community built several tools for searching past agent sessions.
Evidence: conversation search skill
From What do Humans add to AI Tokens? Perspective.
Stripe launched agent payments and a protocol for agents to pay per API call. Shopify turned on agent storefronts by default.
Evidence: Stripe Sessions 2026
From if you're reading this, you're gonna make it.
Meta runs an AI-enabled coding round that can replace a standard one, and Google is piloting the same. LeetCode rounds still exist alongside them.
Evidence: Refonte Learning
The honest misses and the ones still on the clock.
From Agentic Convergence-in-Depth.
No documented case yet. The closest is a telehealth business run by one or two people with about $401 million in 2025 revenue.
From Agile for Agents.
"AX" is a term with vendor guides, not a field of study.
From The Fix-Forward Solutions.
Funded and newly famous, but not yet standard practice.
From Agile for Agents.
Not due yet. Scheduled agents and cloud routines suggest it is on track.
From Second and Third Order Effects of Vibe Coding.
A few universities revised their curricula. The evidence is thin either way.