Workflow Intel

Triage queue

Workflows

ActionableINVESTADOPT-CHEAPTRACKWAIT
All topicspromptingorchestrationcontexttoken-efficiencyother

prompting

Note on 24th September 2026

Simon Willison's Weblog (LLM / AI engineering) · Sep 24, 2026

ADOPT-CHEAPModel: ARCHIVE
ev0cred2dur2com0cmp1app2

Claim: Coding agents make software engineering harder and require extraordinary discipline and knowledge to unlock their potential.

Evidence: Personal observation from time spent working with coding agents; no quantified data, reproducible tests, or comparative metrics provided.

Rationale

Bare assertion from experience, no reproducible data or method. Relevant to Timothy's CC+Codex workflow (applicability=2) but too vague to act on. Hype markers present (superlative "extraordinary", unquantified "full potential"). No commoditization angle. Archive as context for discipline discussion, not actionable triage item.

How NVIDIA engineers and researchers build with Codex

OpenAI — News · May 12, 2026

WAITModel: ARCHIVE
ev1cred2dur1com0cmp0app2

Claim: NVIDIA engineers and researchers use Codex with GPT-5.5 to ship production systems and turn research ideas into runnable experiments

Evidence: Assertion of internal NVIDIA practice; no quantified results, before/after metrics, or reproducible signal provided

Rationale

Baseline tier 3, downgraded to 2: bare assertion of practice with zero quantified evidence. Unverified internal claim from NVIDIA (plausible but unsubstantiated). No metrics, no reproducible signal, no methodology—just a statement that teams do this. GPT-5.5 reference (unfamiliar model name) is plausible per guidance, so credibility is lowered only for lack of evidence, not novelty of model name.