codex · release log
release log
See what changed in the codex this week
131 releases since 2026-03-03. Daily ingests, prompted batches, depth passes. Subscribe via RSS.
Week of 2026-10-05
- 2026-10-09Org structure after AI, multi-model routing, scope expansion, and the Haiku cost cliff+4 insights
- 2026-10-08Public AI work, skill canonicalization, and model commoditization+3 insights
- 2026-10-07Production agent architecture, codebase knowledge graphs, and per-use-case evals+3 insights · +1 operator
- 2026-10-06Inference economics, agent budget risk, and what scale does to quality+4 insights
- 2026-10-05Vibe coding vs. formal design, and Google's alarm clock for the next big update+3 insights
Week of 2026-09-28
- 2026-10-04Agents self-organize, and why hard spending caps must ship on by default+2 insights
- 2026-10-03Agent worms via shared infra, and what human value shifts to when AI does the building+4 insights · +3 operators
- 2026-10-02Agents dissolve management, evals demand data, and attribution collapses under AI search+3 insights
- 2026-10-01Agent deployments correct the theory, and laptops catch the cloud+3 insights
- 2026-09-30Attribution collapses, and one measurement illusion falls with it+4 insights
- 2026-09-29Agent authorship changes what you optimize for in technology+3 insights
- 2026-09-28Agents shift the floor, software factories reveal the ceiling+4 insights · +1 operator
Week of 2026-09-21
- 2026-09-27AI demand's fat middle, engineering as system design, decision models as eval judges+3 insights
- 2026-09-26Vocabulary and agency: two uncaptured dimensions of Mollick's capability-overhang framework+2 insights
- 2026-09-25Specification bugs, forced judgment, and the executive AI gap: what shipped September 18-25+4 insights
- 2026-09-24Criteria drift: why eval agents miss the failures your product team keeps finding+1 insight
- 2026-09-23Extended reasoning limits, publishing cost collapse, and SERP displacement+3 insights
- 2026-09-22Decision model economics and agent governance+3 insights
- 2026-09-21Why AI pacing proposals have no number+1 insight
Week of 2026-09-14
- 2026-09-20Workflow codification as AI sales prerequisite, and counter-positioning as the early-stage moat+2 insights
- 2026-09-19Selection as the scarce resource, harness margin data, domain-expert builders, and outrageous growth+4 insights · +1 operator
- 2026-09-18AI writing tools as mechanical editors, self-generated context injection, and the embedded evaluator proposal+4 insights · +3 operators
- 2026-09-17Code economics upend the developer pipeline, and a model welfare warning from Microsoft AI's CEO+4 insights · +3 operators · +1 pattern
- 2026-09-16AI code quality floors, agent accountability gaps, and channel structure durability+10 insights · +1 operator
- 2026-09-15AI training priority leaks, automation as work-sorter, and what AI detectors actually measure+3 insights
- 2026-09-14AI code quality gates, multi-model security audits, agent attack forensics, and horror launch timing+4 insights
Week of 2026-09-07
- 2026-09-13Autonomous agents drift without quality gates, CAPI outlasts platform tracking restrictions, and open-weight models pace through competition+3 insights
- 2026-09-12Parallelization crossed human consumption, agents can render Blender scenes, and AI hours are invisible in marketing teams+3 insights
- 2026-09-11Coding agents changed mobile platform economics, and October is the wrong month for horror games+3 insights · +2 operators
- 2026-09-10Night-shift economics, AI-as-advertiser pricing, and capability thresholds for agent adoption+4 insights
- 2026-09-09Steam timing myths, professional bias in AI code review, and eval tooling that measures the wrong things+3 insights · +2 operators
- 2026-09-08AI capability stakes, off-site brand reputation in AI search, and software's missing collapse floor+4 insights · +2 operators
- 2026-09-07LLM convergence effects, AI adoption blind spots, chatbot advertising structure, and data center debt at credit-market scale+4 insights
Week of 2026-08-31
- 2026-09-06AI creative distribution, evals as harness feedback, and MVVM deferral in SwiftUI+3 insights · +2 operators
- 2026-09-05Positioning stability under AI churn, GET-only sandbox bypasses, software factories replacing PRs, and listicle freshness signals+4 insights
- 2026-09-04Vibe coding at production scale, Meta's data barter, and agents as trace-first systems+3 insights · +1 operator
- 2026-09-03AI quality floors, citation graphs, salaried ceilings, and prompt governance+4 insights · +1 operator
- 2026-09-02Expert at the wheel, reasoning on a budget, and substitutability as the real ranking risk+3 insights · +1 operator
- 2026-09-01Automate the right half, measure the megawatt, and date your listicle+4 insights
- 2026-08-31Auto mode blocks its own cleanup, product pages hide the attack surface, and agents lower the defect shipping threshold+4 insights
Week of 2026-08-24
- 2026-08-30A negative mean time to exploit, a six-criterion citation framework, and the 48-to-1 widget ratio+3 insights · +1 operator
- 2026-08-29Vendor financing hidden in AR, a replaceability score for content, and the research window that decides Black Friday+3 insights
- 2026-08-28Encoding taste from 30,000 edits, rule complexity that resists full automation, and the disclosure gap in AI writing+3 insights
- 2026-08-27The 37-point citation gap, a dissolved familiarity tax, and reasoning that is just text+3 insights
- 2026-08-26Google's spam filter, an agent's sleep cycle, and the sentence that replaces a grammar textbook+3 insights
- 2026-08-25Cascading constraints, stubborn humans, and what you learn when you refuse to take the shortcut+4 insights · +2 operators
- 2026-08-24When the frontier model gets more expensive, harness investment finally pays off+1 insight
Week of 2026-08-17
- 2026-08-23Instruction over review, widgets over social+4 insights
- 2026-08-22Distribution gates, consent asymmetry, and the local AI cost inflection+3 insights
- 2026-08-21English as interface, discipline as casualty: two frames for understanding what agentic tools change+3 insights
- 2026-08-20Triage before tooling: annotate traces to name failure modes before automating evals+1 insight
- 2026-08-19Small local models rival the cloud via first-principles reasoning, websites earn relevance as AI training sources, and indie games win by amplifying one formula+3 insights
- 2026-08-18Readers test authenticity, third-party sources supply 80% of AI citations, and test-time training changes the per-user model math+4 insights
- 2026-08-17Agent intent is unresolvable, local dev is gone, and LinkedIn citations reward the person not the brand+5 insights · +1 operator
Week of 2026-08-10
- 2026-08-16Community citation floors, category-leader multiples, and a reasoning-trace theft that hit all three major APIs+3 insights · +1 operator
- 2026-08-15Hypothetical classification, SOTA buyer data, system prompt injection, and citation mechanics+4 insights · +1 operator
- 2026-08-14Human-agent trust failures, evidence-free GEO tactics, and the meaning cost of AI rewrites+4 insights · +2 operators
- 2026-08-13AI category repricing and the retrieval-first engineering sequence+3 insights
- 2026-08-12AI valuations, community signals, and festival genre fit+4 insights · +1 operator
- 2026-08-11Agentic engineering at scale, from two directions+4 insights · +1 operator
- 2026-08-10Willison names two new AI evaluation failure modes; Crawshaw argues open source is the agent extension model+3 insights · +1 operator
Week of 2026-08-03
- 2026-08-09Indig measures AI Mode's commercial shift; Willison names the PDF cost trap and the attribution gap in AI agent attacks+3 insights
- 2026-08-08Solis scores content on six AI-era dimensions; Willison names the publish-unhappy threshold; Tunguz extends zero-trust to internally-deployed AI agents+3 insights
- 2026-08-07AI Mode kills the evaluate-and-compare loop; frontier evals move to multi-hour autonomous tasks; delegation and AI ROI share the same missing prerequisite+4 insights · +2 operators
- 2026-08-06AI fails as ghostwriter; tier routing sustains GPU consumption under supply constraints; evals fail upstream; DRI eliminates PM layer+4 insights · +1 operator
- 2026-08-05Don't be a meat proxy; AI search recommendations require depth not breadth; ephemeral sandboxes flip agent spawn economics+3 insights · +1 operator
- 2026-08-03AI-cited layoffs mask structural corrections; the terminal becomes a coordination primitive for agents and humans; vibe-coding flips the SaaS build-vs-buy calculation+3 insights · +1 operator
Week of 2026-07-27
- 2026-08-02Harness engineering beats model selection; AI subscription apps face a retention crisis+5 insights · +1 operator
- 2026-08-01Stateless tools, longer queries, and the eval containment gap: three cards on what the agent era breaks and what it fixes+3 insights
- 2026-07-31Ownership beats resell at the model and chip layer, autonomous runs need front-loaded briefs, and Friendslop demand compounds each cycle: three cards from three operators+3 insights
- 2026-07-30Harness beats model, MoE collapses inference costs, and two new attack surfaces from machine-speed AI: four cards from two sources+4 insights
- 2026-07-29AEO as brand channel, and agentic safety from two angles: four cards from three sources+4 insights
- 2026-07-28Agent security architecture and the open-weight coalition: three cards from two sources+3 insights
- 2026-07-27Scaffolding ceilings, relay markets, and a field experiment on what AI Overviews cost publishers+3 insights · +1 operator
Week of 2026-07-20
- 2026-07-26Rewrites, linting, and ownership: four cards on what the agent era makes cheap and what it makes durable+4 insights
- 2026-07-25Harnesses and context zones: two new operators on what sits between the model and the work+4 insights · +2 operators
- 2026-07-24Agents get a computer, open models catch up, and nobody corrects the CEO+5 insights · +1 operator
- 2026-07-23Three production disciplines from the Claude Code team and the first confirmed real-world AI sandbox escape+4 insights
- 2026-07-22Operating discipline is the lever: frontier models, three productivity tiers, and the signals that separate PMF from novelty revenue+5 insights · +1 operator
- 2026-07-21AI agents collapse the seat-based SaaS expansion model, splitting software into infrastructure that AI feeds and applications that AI feeds on+1 insight
- 2026-07-20B2B pricing pages drive agent fallback to third-party directories, chained web_fetch links enable private data exfiltration, and visual benchmarks have decoupled from frontier model rankings+3 insights
Week of 2026-07-13
- 2026-07-19Token spend breaks PC-era parity, third-party sources dominate AI search citations for SaaS, and CTV compounds B2B brand+4 insights
- 2026-07-18AI adoption stalls at guardrails and process, not model limits, plus two techniques from Bun's 535k-line Rust migration+3 insights
- 2026-07-16The harness battleground, agents dissolving architectural understanding, and open AI infrastructure following the Docker playbook+3 insights · +1 pattern
- 2026-07-15Steam's slop filter, the momentum-over-base formula, and why AI content floods benefit real creators+3 insights · +1 operator
- 2026-07-14The 41-day model window, the colander retention curve, and how agents are quietly draining ERP value+3 insights
- 2026-07-13Agent accountability, preflight memory, and the database lesson for inference infrastructure+3 insights
Week of 2026-07-06
- 2026-07-12Deployment bottlenecks, commodity pricing, and the end of prompting tricks+3 insights · +1 operator
- 2026-07-11Context that AI can't see, and the two jobs knowledge work now has+4 insights · +1 operator
- 2026-07-10Game over-exposure is a myth, and engineering ownership starts at problem definition+2 insights
- 2026-07-09Adversarial agents rewrote a million lines in 11 days, AI worldview maps to alignment not geography, and tech workers split personal optimism from institutional trust+3 insights · +1 operator
- 2026-07-08AI baseline creep drives burnout, overclaiming poisons adoption, and product builders hold four beliefs that collectively stall distribution investment+4 insights · +2 operators
- 2026-07-07AI Overviews decouple citation from recommendation, benchmark-shaped content earns 3.3x more AI citations, and frontier models self-route tasks to cheaper subagents+4 insights · +1 operator
- 2026-07-06Harness-specific model drift, cross-model review practice, concurrent agent norms at OpenAI, and the SaaS category rotation+4 insights
Week of 2026-06-29
- 2026-07-03Agent routing before model choice, schema gaps causing retry loops, and AI search fragmentation across three non-overlapping engines+4 insights · +1 operator
- 2026-07-02Product design before evals, hidden tokenizer costs, and exponential pace mismatches+4 insights · +1 operator
- 2026-07-01Domain expertise as the agentic multiplier, long-horizon run economics, and the Sicko theory of Steam demos+4 insights · +1 operator
- 2026-06-30AI attribution shifts, compute cost inversions, and agentic diffusion beyond engineering+3 insights · +2 operators
- 2026-06-29What landed today, 2026-06-29+2 insights · +2 operators
Week of 2026-06-22
- 2026-06-28What landed today, 2026-06-28+3 insights
- 2026-06-27What landed today, 2026-06-27+3 insights
- 2026-06-26What landed today, 2026-06-26+3 insights · +1 operator
Week of 2026-06-15
- 2026-06-18What landed today, 2026-06-18+4 insights · +2 operators
- 2026-06-17What landed today, 2026-06-17+3 insights · +1 operator · +1 pattern
- 2026-06-16What landed today, 2026-06-16+3 insights · +3 operators
- 2026-06-15What landed today, 2026-06-15+3 insights · +3 operators
Week of 2026-06-08
- 2026-06-14What landed today, 2026-06-14+4 insights · +2 operators
- 2026-06-13What landed today, 2026-06-13+3 insights
- 2026-06-12What landed today, 2026-06-12+5 insights
- 2026-06-11What landed today, 2026-06-11+4 insights
- 2026-06-09What landed today, 2026-06-09+3 insights · +1 operator
Week of 2026-06-01
- 2026-06-04What landed today, 2026-06-04+5 insights · +1 operator
Week of 2026-05-25
- 2026-05-27What landed today, 2026-05-27+5 insights · +2 operators · +1 pattern
- 2026-05-25What landed today, 2026-05-25+4 insights
Week of 2026-05-18
Week of 2026-05-11
- 2026-05-17AEO splits into three jobs, the judgment ceiling holds, and listening compounds+9 insights
- 2026-05-16Five new insights: recovery floors, AI consensus, and task-driven minimalism+5 insights · +2 operators
- 2026-05-15Earn the citation, hold the floor, retire the position metric+4 insights
- 2026-05-13The reliability trap, stripping what does not compound, and context isolation+7 insights · +1 operator · +1 pattern
- 2026-05-11AI implementation realism and AEO benchmarks, 2026-05-11+4 insights · +2 operators
Week of 2026-05-04
- 2026-05-10Context engineering quantified, AEO continuity, volume versus curation+4 insights
- 2026-05-09AEO becomes a page-experience problem; agentic architecture settles into named patterns+11 insights · +4 operators · +1 pattern
- 2026-05-07Verification convergence: four operators, one platform answer+9 insights · +1 operator
- 2026-05-06Verification, judgement, scarcity, voice, three converging frames in one week+15 insights · +4 operators · +2 patterns
- 2026-05-05Research scan v1-v6, depth-pilot to new-operator workflow
- 2026-05-04Quality pass, 79 cards rewritten / stub-flagged