Choose your first delegated use case from real accounts ↗
Use a visible defect, an independent check and a bounded workspace to choose a first agent task. A practical guide grounded in sourced deployments.
What’s new todayLast update
18 additions A week of agents in use.
Week of 2026-10-12 · 38 ideas to explore
Claude Code adjusts a live Cloudflare search challenge ↗
Willison used Claude Code to narrow a Cloudflare challenge to multi-parameter searches. When the MCP interface could not edit the relevant rules, the agent switched to the API.
Claude Code builds a local video compression interface ↗
Willison wanted to optimize a phone recording before publishing it. He used Claude Code for web to build a browser tool around FFmpeg compiled to WebAssembly, with multiple output presets.
Claude Code builds and browser-tests a GIF optimizer ↗
Willison asked Claude Code for web to compile Gifsicle to WebAssembly and build a preview interface. The agent used Rodney to exercise the page and fix UI bugs during implementation.
Claude Code ports an inpainting model into a browser ↗
Willison delegated a Moebius model port to Claude Code using local repositories and a research note. He tested the browser UI and returned errors while the agent iterated on ONNX and WebGPU support.
Claude Code uses Playwright to record a presentation clip ↗
Willison needed a video of an interactive canvas animation for a keynote slide. A local Claude Code session used Playwright to open the page, time clicks and produce a recording.
Claudius checks suppliers in a browser during shop operations ↗
In the next office-shop phase, Claudius gained browser access, CRM and better inventory context. It checked supplier prices and delivery details while human purchase approval remained mandatory.
Claudius runs an office shop and loses money ↗
Anthropic and Andon Labs let a Claude-based agent manage a real office shop. It searched for products, set prices and requested restocking, but the researchers report losses and persistent mistakes.
Codex builds a blog inside an existing Datasette site ↗
Willison published the Codex session used to build the Datasette project blog. The agent inspected the existing site, moved conflicting database routes and implemented blog behavior with local checks.
Codex builds a rate limiter for an over-crawled site ↗
When crawlers hammered datasette.io, Willison used Codex to build a configurable IP rate-limiting plugin. His release note includes the configuration used on the live site.
Codex fixes Markdown leaking into social previews ↗
Simon Willison gave Codex a specific defect in his Django blog: social preview descriptions exposed raw Markdown. Codex changed templates and added a passing test in a configured PostgreSQL environment.
Codex ports a Python HTML parser to JavaScript ↗
Willison supplied an existing parser and an independent compatibility suite, reviewed an API plan, then let Codex implement a JavaScript port through repeated tests and commits.
Cursor implements and revises a hybrid search service ↗
Dodds used Cursor with GPT-5.4 to add lexical retrieval to his semantic search. Production blocking exposed an architectural flaw, so later agent work moved search into a unified Worker.
Cursor simplifies an audio pipeline into a one-shot sandbox ↗
Dodds delegated an audio pipeline migration to Cursor. He rejected an initial design that preserved a separate service and used another agent run to make the sandbox part of the existing queue Worker.
Deep research gathers an incomplete orchestrion location dataset ↗
Willison ran o4-mini-deep-research to find surviving orchestrions and return structured location records. The model used web search and code execution; he explicitly considered the resulting list incomplete and unverified.
Robin proposes retinal disease experiments and analyzes lab data ↗
FutureHouse used Robin to connect literature research, experiment planning and analysis for dry age-related macular degeneration. Human researchers ran the lab work; the system proposed follow-up candidates from the results.
Watch Claude Code build a terminal-session sharing tool ↗
Willison recorded a real build with Claude Code for web, turning pasted terminal output into shareable HTML. The companion article explains the repository references and the workflow behind the recording.
Watch Codex repair a Datasette plugin compatibility failure ↗
Willison recorded upgrading datasette-checkbox for a changed permissions API. He first repaired a test that missed a real page failure, then gave Codex the failing runner and upgrade documentation.
What this week's sourced agent use cases teach us ↗
An editorial review of the October 14 collection: tested patches, browser work, research drafts and shopkeeping failures, with primary sources and practical limits.
WikiCrow writes cited drafts about human genes ↗
FutureHouse ran WikiCrow to draft scientific summaries of human genes using full-text papers. Its PaperQA agent searched, gathered evidence and returned to retrieval when terminology or evidence required it.
Write approval checkpoints that an agent can act on ↗
Define the action, evidence, approver and expiry for agent permissions. Includes concrete checkpoints for WAF edits, supplier quotes and code publication.
Assemble release notes from merged changes ↗
Read merged changes, investigate user impact, and assemble a release draft with references back to the work.
Build a creative concept from a working brief ↗
Develop a set of campaign directions, evaluate them against a brief, and prepare a coherent creative proposal.
Build a supplier shortlist against explicit criteria ↗
Investigate public supplier information, compare it against procurement criteria, and prepare an evidence-backed shortlist.
Choose a small first use case ↗
Find a bounded task with a visible output, a clear review step and a useful failure mode before expanding agent access.
Design the checkpoint before the use case ↗
A practical way to decide where an agent can proceed and where a person needs to review the work.
Evidence before confidence ↗
Build a research brief that makes its sources and uncertainties as easy to inspect as its conclusions.
EXAMPLE: A research checkpoint in motion ↗
An illustrative video entry showing how an agent workflow could be reviewed, with a deliberate human checkpoint.
EXAMPLE: Watch a workflow review ↗
An illustrative video entry showing how an agent workflow could be reviewed, with a deliberate human checkpoint.
Find and draft fixes for knowledge-base gaps ↗
Compare recurring support questions with existing documentation and draft improvements for missing explanations.
Investigate dependency updates before a merge ↗
Inspect proposed dependency updates, review upstream changes, and test compatibility in a disposable environment.
Keep a reading queue useful without daily sorting ↗
Review a saved reading list, follow relevant references, and prepare a concise digest tailored to stated interests.
Prepare a weekly plan from scattered commitments ↗
Reconcile tasks and calendar commitments, identify conflicts, and draft a realistic weekly plan.
Take a bug report all the way to a tested patch ↗
Inspect a repository, reproduce a reported bug, and prepare a tested patch for review.
Translate a story idea into a storyboard plan ↗
Expand a story brief into a scene sequence, check continuity, and create a production-ready storyboard outline.
Turn a research question into a sourced brief ↗
Investigate a defined question across multiple sources and produce a brief with traceable claims.
Turn a shared inbox into an action queue ↗
Review incoming messages, gather missing context, and turn requests into a prioritized queue of draft actions.
Watch a market and surface meaningful changes ↗
Monitor a bounded set of public information sources and explain changes against an established baseline.