Radar
Practitioner perspective

Treat generated code as work that needs review

Simon Willison shares how he uses language models for coding, including managing context and accounting for mistakes.

Simon Willison - public profile image

Why it matters

Try a bounded task and review the result. This is personal practice from 2025, not a current comparison of models.

Who said it

Simon WillisonSimon Willison

Independent author perspective. Product mentions do not establish sponsorship or endorsement.

Evidence

First-person account · Read the original

Published Mar 11, 2025. Reviewed Sep 17, 2026. Review dates do not make older statements new.

Why this ranks here 63/100
Topic relevance 30% weight
50/100

General audience fit; choose a topic to personalize.

Evidence 25% weight
50/100

First-person account

Actionability 20% weight
100/100

Try a bounded task and review the result. This is personal practice from 2025, not a current comparison of models.

Originality 15% weight
100/100

Original contribution; repeated links count as one source.

Freshness 10% weight
0/100

Based on original publication, not our review date.

Editorial priority, not a probability or trust score. How ranking works

Put it to work

These are editorial topic connections, not endorsements or eligibility decisions.

01
Anthropic - publisher preview

Claude Code

Included with paid Claude plans; API billing is separate

Check your fitReview conditions before applying
Why this ranks here 50/100
Relevance 35% weight
50/100

General audience fit; eligibility is checked separately.

Practical value 25% weight
50/100

Included with paid Claude plans; API billing is separate

Timing 20% weight
50/100

Based on a usable confirmed window; rolling offers receive 50.

Action clarity 20% weight
50/100

Official action link. Individual conditions still need checking.

Editorial priority, not a probability or trust score. How ranking works

01
OpenAI

Codex for Open Source

6 months of ChatGPT Pro for selected maintainers

Details need confirmationReview conditions before applying
Why this ranks here 30/100
Relevance 35% weight
50/100

General audience fit; eligibility is checked separately.

Practical value 25% weight
50/100

6 months of ChatGPT Pro for selected maintainers

Timing 20% weight
0/100

Timing or availability needs confirmation.

Action clarity 20% weight
0/100

Official action link. Individual conditions still need checking.

Editorial priority, not a probability or trust score. How ranking works

01
GitHub - publisher preview

GitHub Copilot Free

Limited core Copilot access

Check your fitReview conditions before applying
Why this ranks here 50/100
Relevance 35% weight
50/100

General audience fit; eligibility is checked separately.

Practical value 25% weight
50/100

Limited core Copilot access

Timing 20% weight
50/100

Based on a usable confirmed window; rolling offers receive 50.

Action clarity 20% weight
50/100

Official action link. Individual conditions still need checking.

Editorial priority, not a probability or trust score. How ranking works