AI Digest — July 12, 2026
Quick Notes
- Your company has rules nobody consciously voted on—they live in meetings and processes; AI dissolves scarcity but not judgment, leaving you to sort which rules still matter from those running on inertia. https://natesnewsletter.substack.com/p/ai-native-company-rules
- Anthropic is extending Claude Fable 5 access through July 19 across all paid plans, allowing subscribers to use up to half their weekly usage limit before switching models or using credits after exhausting it. https://simonwillison.net/2026/Jul/12/bump/#atom-everything
- DRI stands for Directly Responsible Individuals: someone ultimately accountable for project success; this remains uniquely human since machines cannot take accountability for their actions, echoing IBM’s 1979 stance that computers must never make management decisions. https://simonwillison.net/2026/Jul/12/directly-responsible-individuals/#atom-everything
- shot-scraper 1.11 releases with minor improvements around command option consistency and fixing the server: mechanism when the service takes longer than a second to start serving traffic. https://simonwillison.net/2026/Jul/12/shot-scraper/#atom-everything
- sqlite-utils 4.1.1 includes mainly a fix for an edge case discovered while experimenting with the 4.1 release, specifically addressing questions about ON DELETE behavior that regular Claude chat spotted. https://simonwillison.net/2026/Jul/12/sqlite-utils/#atom-everything
- The video argues companies like Anthropic ship fast by moving repeatable human interactions to code: decisions become documents agents can act on, product management moves into the terminal, reviews become evals with agents building on them over time. https://www.youtube.com/watch?v=hYcOFTMesGc
Structured Summaries
Tooling Updates
Two notable updates ship today: shot-scraper 1.11 improves command option consistency and ensures the server: mechanism works reliably even when service startup exceeds a second; sqlite-utils 4.1.1 addresses an edge case in ON DELETE handling that surfaced from user experiments with the earlier release.
https://simonwillison.net/2026/Jul/12/shot-scraper/#atom-everything https://simonwillison.net/2026/Jul/12/sqlite-utils/#atom-everything
New Models & Product Changes
Anthropic extends Claude Fable 5 access through July 19 on all paid plans, maintaining the boosted weekly rate limits at 50% higher levels; after consuming half your weekly limit you may continue with Fable using credits or switch models. OpenAI’s GPT-5.6 appears not to require similar restrictions as Anthropic applied to Fable originally citing compute constraints and demand gauging—OpenAI seems confident they won’t need access restrictions on GPT-5.6 in the same way, fueling suggestions that Anthropic should keep Fable permanently available rather than cycling it off.
https://simonwillison.net/2026/Jul/12/bump/#atom-everything
Research Insights
Organizational structure around AI agents demands rethinking accountability: DRI (Directly Responsible Individuals)—the person ultimately accountable for a project’s success or failure—must remain human because machines cannot take responsibility for their actions, echoing IBM’s legendary 1979 training slide. Companies like Anthropic and OpenAI ship at fast cadence by converting repeatable coordination into code: decisions become documents agents can reference rather than vanish in meetings; product managers work directly in terminals with engineers rather than through ticketing systems; reviews evolve into evals that agents improve over time. Trust stays human, but repeatable parts of business coordination move toward code to avoid making humans the rate limit on execution speed.
https://simonwillison.net/2026/Jul/12/directly-responsible-individuals/#atom-everything https://www.youtube.com/watch?v=hYcOFTMesGc
Research Insights: Organizational Change
Every organization already runs on implicit rules written by scarcity conditions rather than deliberate choice—engaging engineering time through documentation rationing, carrying context in meetings that moved too slowly to digitize, approval chains protecting against costly mistakes. AI has dissolved much of that scarcity without removing judgment, leaving operators with rules enforced by outdated conditions and no clear way to distinguish what still matters from what runs on inertia alone. Effective organizational transformation separates value, rule, runtime check, and human appeal into different owners; it watches prohibitions only when paired with replacements; copying someone else’s constitution enforces nothing—the work must be legible to agents. Five questions help rewrite resented rules: name the behavior, find its scarcity condition, assess if that scarcity persists, articulate what could break a person-enforced block, and climb the enforcement ladder from value through instruction, reminder, hard block, to human-owned decision where no rung six exists.
https://natesnewsletter.substack.com/p/ai-native-company-rules
🔗 View this digest on the web: https://ai-digest-b7u.pages.dev/digests/2026-07-12-evening/