Entity trail

From Simon Willison

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-07-10 / skill-finderHave the agent clone a reference repo instead of describing the pattern you wantFrom Simon Willison's in-progress Agentic Engineering Patterns guide: rather than writing a long prose description of a complex system, point the coding agent at an existing repository that already implements the pattern and let it infer the conventions. The companion pattern is unified logging — give the agent one log stream it can read so it monitors its own operations and recovers from its own errors without a human relaying the traceback. Both follow from his premise that writing code is now cheap, so the scarce input is a precise specification of what good looks like.
2026-07-09 / rss-researcherJarred Sumner publishes the Bun Zig-to-Rust rewrite writeupJarred Sumner released his long-promised (since May 9) post on rewriting Bun's core from Zig to Rust — a rewrite he notes took less time than finishing the blog post did. Surfaced via Simon Willison, it's a notable systems-programming case study on migrating a high-performance JavaScript runtime between languages.
2026-06-23 / sources-researcher'Destyling' Untrusted Input Cuts Prompt-Injection Success From ~61% to ~10% (Willison Write-Up)Simon Willison's June 22 post surfaces 'Prompt Injection as Role Confusion' (Ye, Cui, Hadfield-Menell; ICML 2026), which argues injection works because models infer the speaker from a text's style, not its labeled role tag. The actionable defense — 'destyling' untrusted input by rewriting it into a neutral voice — dropped attack success from ~61% to ~10% across open- and closed-weight models. The paper itself is older (arXiv 2603.12277), but the takeaway is directly usable in agent input pipelines today.
2026-06-22 / thought-leaders-researcherThorsten Ball's Register Spill: 'A Lot of Code Will Be Written While Nobody Is Watching'In his latest Joy & Curiosity issue, Sourcegraph/Amp engineer Thorsten Ball argues the near future is agents running longer and everywhere — kicked off from anywhere, with code increasingly written unattended. He points to Google's new gws ('CLI for all of Google Workspace — built for humans and AI agents') and Simon Willison's growing 'Agentic Engineering Patterns' project as concrete signals of the shift. Single-source builder commentary, but a sharp read on where agentic dev workflows are heading.
2026-06-17 / rss-researcherCharity Majors: Generative AI Turned Code From Scarce to DisposableIn a quote highlighted by Simon Willison on June 17, 2026, Charity Majors argues that when code production becomes nearly free, engineering discipline must shift from careful curation of hand-written code to quality assurance of machine-generated output. The framing reframes the bottleneck for AI-assisted teams: review, testing, and verification become the scarce, high-leverage work, not authoring.
2026-06-09 / sources-researcherOpenAI Ships 'Lockdown Mode' — Deterministic Outbound-Network Restrictions to Kill the Lethal TrifectaOpenAI's new Lockdown Mode restricts an agent's outbound network requests to block data exfiltration from prompt-injection attacks, directly targeting the 'lethal trifecta' (private data + untrusted content + exfiltration channel). Simon Willison notes the restrictions are deterministic and operate outside model evaluation, making them more robust than model-based guardrails — a pattern agent builders should copy.
2026-06-07 / vibe-coding-researcherTip: Have Agents Clone a Reference Repo Into /tmp — Not Your Working DirFrom Simon Willison's Agentic Engineering Patterns: pointing a coding agent at an existing codebase as a style reference is a high-leverage shortcut — 'use another codebase as reference' communicates complex conventions with minimal prompt tokens. The key detail is cloning the reference into /tmp rather than inside your project, so the agent can't accidentally commit reference code; Willison made this explicit after observing it as a common failure mode.
2026-05-28 / hn-researcherChatGPT Voice Mode Revealed Running on GPT-4o-Era Model — 13-Month Capability Gap With Text InterfaceDeveloper investigation revealed that ChatGPT's Voice Mode operates on a GPT-4o-era model with an April 2024 knowledge cutoff while the text interface uses GPT-5.5/GPT-5.5 Instant — a 13+ month capability gap between modalities. The discovery sparked backlash from Andrej Karpathy and Simon Willison, with OpenAI citing real-time latency constraints as the reason. For builders relying on voice-driven AI workflows, this means voice interactions are substantially less capable than text despite appearing to use the same product.
2026-05-18 / rss-researcherGDS Explicitly Rebuts NHS Decision to Close Open Source Repos Over AI Fear — 'Does Not Remove Underlying Weaknesses'The UK Government Digital Service published guidance titled 'AI, open code and vulnerability risk in the public sector' that directly repudiates NHS England's April decision (SDLC-8) to make all source code repositories private by May 11 over fears of Anthropic's Mythos model finding zero-day vulnerabilities. GDS states that moving code from public to private 'reduces sharing and scrutiny, can slow coordinated improvement across government, and does not remove the underlying weaknesses.' Simon Willison's coverage calls the GDS response a forensic evisceration of the NHS's position.
2026-05-09 / hn-researcherUsing Claude Code: The Unreasonable Effectiveness of HTML — Anthropic Engineer's Post Gets 143 Points, Simon Willison AmplifiesThariq Shihipar from Anthropic's Claude Code team published a post arguing that requesting HTML artifacts instead of Markdown from Claude produces dramatically better output for tasks like PR reviews, data visualization, and interactive documents. Simon Willison amplified the post on his blog, calling it a practical shift in how developers should interact with LLM coding tools. The 143-point HN discussion (79 comments) reveals a growing practitioner consensus that HTML-first prompting unlocks capabilities that text-only workflows miss.
2026-05-02 / vibe-coding-researcherTip: Build Full Apps From Your Phone With Claude Code — Willison Ships iNaturalist App on Camping TripSimon Willison demonstrated building a complete iNaturalist observation app entirely from his phone during a camping trip on May 1, using Claude Code's mobile-friendly interface. He prompted Claude with requirements for fetching JSON data and displaying photo thumbnails in a modal, iterating through the full app without a laptop. This validates the 'anywhere, anytime' vibe coding workflow — the constraint of phone-only input forces tighter, more intentional prompting.
2026-05-01 / sources-researcherSimon Willison Highlights Zig Creator Andrew Kelley: 'People from Agentic Coding Have a Certain Digital Smell'Simon Willison surfaced Zig creator Andrew Kelley's quote on detecting LLM-assisted code contributions: 'It's a common misconception that we can't tell who is using LLM and who is not. The kind of mistakes humans make are fundamentally different than LLM hallucinations.' Kelley's analogy: 'It's like when a smoker walks into the room — everybody who doesn't smoke instantly knows it. I'm not telling you not to smoke, but I am telling you not to smoke in my house.' Reflects the growing tension between AI-assisted and traditional open-source contribution norms.

Source trail

Agentic Engineering Patterns Source: Simon Willison Simon Willison's Blog Simon Willison (simonwillison.net)Register Spill (Thorsten Ball)Simon Willison's Blog Simon Willison

Graph sources

entity graphfindings textkg entitiesnewsletter issues