Capability As Feasibility Knowledge Plus Executable Proof

Issue 57 Edition 2026-02-26 6 min read

General

Sources: 1 • Confidence: High • Updated: 2026-04-13 03:58

Key takeaways

The document asserts that building software skill depends heavily on knowing what is possible and having a rough idea of how to accomplish it.
The document asserts that collected working examples can be used as inputs for coding agents by prompting the agent to build new things by combining two or more existing examples.
Willison publishes notes via blog/TIL posts and maintains over a thousand GitHub repositories, many of them small proof-of-concepts.
The document asserts that coding agents make the hoarding approach more powerful by fetching and reusing source code from the internet and local codebases as context for new tasks.
tools.simonwillison.net is described as Willison’s largest collection of LLM-assisted single-file HTML tools with embedded JavaScript and CSS that solve specific problems.

The document asserts that building software skill depends heavily on knowing what is possible and having a rough idea of how to accomplish it.
The document asserts that documenting a useful trick once with a working code example allows agents to consult it and solve similarly shaped projects in the future without re-deriving the technique.
The document asserts that knowing something is theoretically possible is not equivalent to having personally seen it working as running code, and that the latter is preferable.
The document asserts that accumulating answers to obscure or broad feasibility questions increases the likelihood of spotting problem-solving opportunities others miss.

The document asserts that collected working examples can be used as inputs for coding agents by prompting the agent to build new things by combining two or more existing examples.
Willison built a browser-based OCR tool by combining prior snippets using PDF.js to render PDF pages as images and Tesseract.js (WebAssembly) to extract text from those images in JavaScript.
A single prompt that included two example codebases produced a working proof-of-concept HTML page that converted dropped PDFs into per-page JPEGs and ran OCR, displaying results beneath each image.
The document asserts that iterating a few times with a model can quickly turn an LLM-generated proof-of-concept into a useful tool with lasting personal value.

Willison publishes notes via blog/TIL posts and maintains over a thousand GitHub repositories, many of them small proof-of-concepts.
tools.simonwillison.net is described as Willison’s largest collection of LLM-assisted single-file HTML tools with embedded JavaScript and CSS that solve specific problems.
The simonw/research repository is described as containing larger agent-driven examples where a coding agent researched a problem and returned working code plus a written report.

The document asserts that coding agents make the hoarding approach more powerful by fetching and reusing source code from the internet and local codebases as context for new tasks.
Willison often instructs agents to clone his public repositories to /tmp and mine them for examples to apply to new projects.
When using Claude Code, specifying curl can be necessary because the default WebFetch tool summarizes pages instead of returning raw HTML.

How often does the recombination prompting pattern produce correct, maintainable integrations across different problem types beyond the OCR example?
What is the measured time/effort saved (or additional time incurred) by maintaining the hoard of repositories, notes, and single-file tools?
What are the failure modes when agents fetch and reuse external or local code, including incorrect dependency assumptions, licensing constraints, or security risks?
What retrieval and context-delivery practices are sufficient to ensure fidelity (raw source, correct versions, complete dependencies) when using agent toolchains?
What organizational process best supports sharing, searching, and maintaining a growing library of executable examples across a team rather than an individual?

Rising demand for developer tooling that packages runnable examples into agent-ready context, treating proof-of-concepts as reusable capital.
Increased value of code retrieval and context-delivery infrastructure that provides raw, version-correct source and dependencies for agent workflows.
Growth opportunity for lightweight, single-file, self-contained tools that solve narrow tasks and can be rapidly recombined into integrated utilities by agents.

More documented workflows and tools focused on fetching raw source, pinning versions, and bundling complete context for coding agents.
Teams adopting shared libraries of executable examples and small repos as standard practice, with search and reuse as a core workflow.
Repeated examples of agent prompts successfully recombining multiple known-good snippets into working prototypes with fewer iterations.

Evidence that recombination prompting often yields brittle or unmaintainable integrations outside narrow demos, requiring heavy manual rework.
Persistent retrieval fidelity issues such as incomplete code, wrong versions, missing dependencies that prevent agents from producing runnable outputs reliably.
Material risks from reused code such as licensing conflicts or security issues that discourage organizations from agent-driven code reuse practices.