Flask creator Armin Ronacher (@mitsuhiko) posted on November 1 that he is convinced an Opus 5.5 agent could now re-implement all of CPython in about two weeks against just the tests, with no human input, possibly with ABI compatibility, no GIL, a JIT, and better numbers on every measurable metric. The post had about 1,637 likes, 98K views and 162 replies. He later added that a rewrite does not strictly need an existing test suite, and that having the agent build its own tests is often better for porting. antirez agreed, while Mario Zechner and Meta engineer Kevin Newton pushed back. This is a personal prediction; no such rewrite has been done.

Key Takeaways

  • ✓Original post: @mitsuhiko, Nov 1, 11:52 UTC, about 1,637 likes, 269 bookmarks, 162 replies and 98K views at fetch time
  • ✓His core claim: given just the tests and no human input, an agent could rewrite CPython in about two weeks, "probably" with full compatibility including ABI, no GIL and a JIT, which he called "completely crazy"; it is a prediction, not an experiment
  • ✓Follow-up post (14:07 UTC, about 212 likes): an existing test suite is not required, and many pre-existing tests are bad for hill climbing; he says an agent he ran a few weeks ago re-implemented the Excel formula engine and built its own test suite from scratch
  • ✓antirez (creator of Redis) agreed that you can ask an agent to build a large test suite before starting a port; Mario Zechner disagreed, saying his problem with agents is trust and pre-existing test and conformance suites increase it
  • ✓Meta engineer Kevin Newton pushed back that very few Python JITs have succeeded because of complexity and compatibility, and that a good one without human input is unrealistic; others noted that under PEP 703 the free-threaded build is not ABI-compatible with the standard one
⚡

Key Decision Metrics at a Glance

TL;DR Verdict
🏆 Benchmark RankRank
See how it ranks on the leaderboard
Full Leaderboard→
💰 Cost & PlansCompare
Claude Opus 5.5 · in: $4 / out: $20
Compare Plans↗
🛠️ Ecosystem & IDERules
AI coding agents · CPython · test suites
Setup IDE Rules↗
🧭

Turn your technical choice into a development budget

Compare 40 dev plans & simulate token costs vs $20/mo subscriptions

⚔️ Compare Model→💰 Compare plans & limits↗
🔗

Project Links & Resources

Direct Access
Direct access to official project resources and documentation
📖
PEP 703: Making the Global Interpreter Lock Optional in CPython
peps.python.org/pep-0703/
↗
🐙
CPython regression test suite (Lib/test)
github.com/python/cpython/tree/main/Lib/test
↗
🔬

In-Depth Technical Analysis

Background

Armin Ronacher created the Python web framework Flask and regularly shares how he works with coding agents. On November 1 he posted that an Opus 5.5 agent could now rewrite all of CPython in about two weeks, using only CPython's test suite and no human input. The post drew about 1,637 likes and 98K views, and the debate quickly moved from whether this is possible to how much a test suite is really worth.

How it would work

His idea is to treat the test suite as the acceptance criteria and let the agent keep running tests and fixing code until everything passes. He added that the result could "probably" keep ABI compatibility, drop the GIL (global interpreter lock), add a JIT compiler and beat the original on every measurable metric. In a follow-up he said existing tests are often bad for this kind of iterative optimization, and that it is better to have the agent build a test suite specifically for the port; when an agent re-implemented the Excel formula engine for him a few weeks ago, it wrote its own tests from scratch.

Evidence and pushback

There are no benchmarks or actual rewrite results here; the two-week figure and the across-the-board speedups are his prediction. The pushback is serious too. Meta engineer Kevin Newton noted that very few Python JITs have ever succeeded, because of complexity and compatibility. One reply cited PEP 703: CPython's own free-threaded build is not ABI-compatible with the standard build, so "ABI-compatible and no GIL" is itself in tension.

What developers can use

Whichever side you take, the thread offers a practical pattern: before asking an agent to rewrite or port old code, have it write tests that pass on the old implementation, add coverage checks, and use the old code as the answer key. Mario Zechner's point is worth keeping too: existing conformance suites increase your trust in what the agent produces. Treat this as a widely discussed practitioner view, not a verified result.

Action HubReady to adopt this in production?

Benchmark side-by-side against alternatives, or calculate monthly token cost vs subscription break-even.

⚔️ Launch VS Compare→💳 View Plans↗
Up NextX discussion: Shopify CEO Tobi Lütke backs the monorepo as a Shopify engineer calls it a boon for coding agents, while critics still prefer split repos
Read Now→