Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-03-15 · source-backed
Three independent sources — Simon Willison at the Pragmatic Summit, blog.tedivm.com's coding agent guide, and Anthropic's 2026 Trends Report — converged in March 2026 on the same conclusion: the highest-leverage skill in AI-assisted development is no longer prompt engineering but specification engineering. Define scope, constraints, acceptance criteria, and architectural boundaries in a structured brief before any AI is invoked. Each AI-generated feature goes on its own branch with the spec as the review baseline. blog.tedivm.com
This isn't just a workflow suggestion — it's the emerging consensus for why some teams ship reliably with agents while others drown in verification debt. Willison's fireside chat framed it as "conformance-driven development": write the test, write the spec, then let the agent implement against both. simonwillison.net The spec becomes the contract the agent must satisfy, not a suggestion it can interpret creatively.
Addy Osmani's PEV loop (Plan → Execute → Verify) formalizes the same pattern: humans define goals and constraints, agents plan and execute, humans review output as a junior engineer's PR — never as a trusted commit. Addy Osmani Boris Cherny, the Anthropic engineer who built Claude Code, uses a CLAUDE.md structured around six operational areas — plan mode defaults, subagent delegation, verification requirements, elegance checks — all under 300 lines, writing only what Claude would get wrong without the file. Glen Rhodes
The pattern is clear: front-load specification, keep agent context lean, verify against the spec. Everything else is commentary.
Each link below shares sources, entities, or timing with this story.
Anthropic: Prompt Caching as Production Architecture — The most significant engineering disclosure from Anthropic this year. Claude Code's static-first prefix structure, Plan Mode as callable functions, Tool Search with defer_loading, and compaction maintaining identical syste...
Simon Willison spent a while taking ChatGPT Work apart and published the map on August 30. Work splits into Work Cloud and Work Local, the latter being the renamed Codex desktop app, at $20/month and up since July 9. He enumerates six capabilities Work has that Chat doesn't, a...
For a month, Claude Code users were convinced the model had been "nerfed." Forums lit up. Conspiracy theories multiplied. People switched tools. Then on April 23, Anthropic did something unusual: they published a detailed post-mortem that named three specific bugs with exact d...
Addy Osmani from the Chrome team wrote up a phrase this month that's going to stick: loop engineering. The idea is plain once you've lived it. You stop writing prompts and start designing the system that prompts, verifies, and remembers for the agent. The model isn't the bottl...
On July 16 there was a wave of backlash calling the Bun Zig→Rust rewrite unreviewed AI slop. On July 19, Simon Willison went and checked. (Simon Willison) Jarred Sumner claimed Claude Code v2.1.181 and later ship the Rust port. Willison verified it independently rather than ta...
Anthropic released Claude Fable 5.1 on September 1. Claude Code v2.1.257 made it the default Fable model at 17:53 UTC that day, with a 1M-token context window, $10 per million input tokens, $50 per million output, and $0.25 per million on cache reads (claude-code CHANGELOG). B...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.