Simon Willison: AI measurably boosted my productivity — data from Datasette GitHub code-frequency chart

Core argument

Willison examined the GitHub code-frequency chart of his long-standing open-source project Datasette and identified an explicit peak: 37,022 lines added and 9,528 deleted in one week in 2026 — historically the most significant spike in the project's entire existence. He directly attributes this to the impact of frontier AI models on his coding productivity.

Context

Willison is one of the most careful documenters of working with AI tools in real development. This time he presents not an anecdote, but a measurable historical data point — actual code activity visible in git history. The Datasette project has run for over 8 years, with the previous high being 15,998 lines added in 2018.

Why it matters

For developers, this is one of the few publicly available, verifiable pieces of evidence about the measured impact of frontier models on individual productivity — not a survey, but an actual git history spike. Implication: switching from mid-tier to frontier models for coding is measurably and dramatically different.

Details / arguments

• Largest spike: 37,022 lines added, −9,528 deleted in one week of 2026
• Second largest: 14,638 added in late 2025 (correlates with Opus 4.5)
• Previous record from 2018: 15,998 lines — surpassed by 2.3×
• Correlation points: Opus 4.8, GPT-5.5, Fable 5, GPT-5.6 Sol
• Datasette: Python library for exploring SQLite databases, over 8 years of development