Simon Willison Showcases Claude Code as a Rapid Prototyping Agent · history
Version 2
2026-05-11 19:40 UTC · 8 items
What
Simon Willison published two posts on May 4, 2026, using Claude Code as an agentic rapid prototyping tool to explore emerging technical developments. In the first, he built a browser-based interactive playground running a WASM-compiled subset of Redis to demonstrate Salvatore Sanfilippo's proposed native array data type — a feature in an open PR that adds over 18 new commands[1]. In the second, he used Claude Code to build an experimental Python binding to the TRE regex library via ctypes and tested it against pathological ReDoS inputs, finding TRE significantly more robust than Python's standard re module due to its lack of backtracking[2].
Why it matters
Willison's posts illustrate a concrete use pattern for agentic coding tools: rapidly building just enough — a WASM playground, a ctypes binding — to make an unshipped or underexplored technology immediately tangible and shareable. The TRE finding also has practical security implications, pointing toward a production-usable alternative for Python applications that process untrusted regex input.
Open questions
Will Salvatore Sanfilippo's Redis native array PR be merged in its current form, and will ARGREP's TRE-backed regex behavior survive into a stable Redis release?[1]
How much of TRE's regex feature set is missing relative to Python's re module — and is the ReDoS resistance worth the trade-off for real workloads?[2]
Can Claude Code-generated experimental artifacts like the WASM playground and ctypes binding be hardened into production tools, or are they best understood as disposable demos?[2][1]
Narrative
Simon Willison published two posts on May 4, 2026, each using Claude Code as an agentic rapid prototyping tool to make an emerging or underexplored technology immediately accessible.
The first post announced a browser-based interactive playground for a significant Redis feature still in development: a native array data type proposed by Salvatore Sanfilippo (antirez) in an open PR that adds more than 18 new commands[1]. Willison had Claude Code build the playground using a WASM-compiled subset of Redis, allowing users to experiment with the new array commands directly in the browser without any local install. He highlighted ARGREP — a command that runs server-side regex grep over array values using the TRE regex library recently vendored into Redis by antirez — as the most interesting of the new additions[1].
The second post followed from that same discovery. Curious about TRE's security properties after seeing it included in Redis, Willison used Claude Code to build an experimental Python binding to TRE via ctypes and tested it against pathological ReDoS inputs[2]. His finding: TRE handles those inputs significantly better than Python's standard library re module, primarily because TRE does not support backtracking — the mechanism most commonly exploited in regex denial-of-service attacks[2]. The ctypes binding and tests are explicitly experimental, intended to probe and demonstrate TRE's behavior rather than to ship as a library.
Across both posts, the framing is consistent: Claude Code functions as an agentic collaborator for rapid prototyping, building just enough — a WASM playground, a ctypes binding — to make nascent or obscure technology tangible and shareable. Neither artifact is presented as production-ready; their value lies in the exploration they enable and the findings they make concrete for an audience. Several Wikipedia articles on adjacent topics (Llama, 3D printing, Memex, regex engine comparisons, TRE, Jesse Vincent) have been surfaced as reactive background context by the pipeline but carry no substantive claims of their own.
Timeline
- 2026-05-04: Willison publishes Redis Array Playground post, announcing a Claude Code-built WASM in-browser demo of Salvatore Sanfilippo's proposed Redis native array data type with 18+ new commands [1]
- 2026-05-04: Willison publishes TRE Python binding post, demonstrating ReDoS robustness via a Claude Code-built ctypes binding and comparing TRE to Python's standard re module [2]
Perspectives
Simon Willison
Advocates for Claude Code as an agentic rapid prototyping tool that lowers the cost of exploration — building experimental playgrounds and language bindings to make emerging technologies immediately tangible and shareable.
Evolution: consistent
Tensions
- Whether TRE's lack of backtracking is an acceptable trade-off for real-world regex use: it provides strong ReDoS resistance but may not support the full feature set Python developers expect from the standard re module. [2]
- The status and trajectory of Sanfilippo's Redis array PR remains unresolved — the playground exists precisely because the feature is not yet in a stable Redis release, leaving its eventual shape uncertain. [1]
- Whether Claude Code-generated experimental artifacts (WASM playgrounds, ctypes bindings) can or should be hardened into production tools, or whether they are best understood as disposable demos that illustrate a capability rather than deliver one. [2][1]
Sources
- [1] Redis Array Playground — Simon Willison (2026-05-04)
- [2] TRE Python binding — ReDoS robustness demo — Simon Willison (2026-05-04)