Weekly commits to open-source crypto repositories have fallen from approximately 871,000 per week in 2024 to around 218,000 in early 2026. That is a 75% drop in the raw metric that has been the industry’s go-to proxy for developer health for a decade. The coverage has predictably been alarmed.
It is the wrong read.
Key Highlights
- Weekly crypto GitHub commits dropped 75%, from 871,000 in 2024 to 218,000 in early 2026 — but the metric is measuring the wrong thing
- AI coding tools (Copilot, Cursor, Claude, Devin) reduce commit frequency while maintaining or increasing actual output — one developer now ships what three previously did
- Electric Capital’s own data shows a 300% increase in developer activity within the AI-crypto sector over the past 12 months
- Ethereum unique monthly contributors remain above 2020 levels; Base and Arbitrum developer counts are growing, measured by deployed contract addresses
- The commit metric was always a proxy for effort, not output — AI tools have broken the correlation between the two
- New leading indicators are needed: deployed contracts per month, active builder wallets, protocol revenue per developer, and onchain volume per team member
What the Data Actually Measures
Electric Capital, which publishes the most comprehensive crypto developer reports, tracks public GitHub commits to crypto-related repositories. The metric counts the number of times developers push code changes to open-source projects.
In 2021 and 2022, when that metric hit its peaks, most developers at crypto protocols wrote code the old way: incrementally, manually, committed in small batches throughout the day. A productive developer might generate 15 to 30 commits per week. A team of 10 might generate 200.
In 2026, the same developer uses AI coding tools. Copilot, Cursor, Claude, and Devin handle significant portions of routine code generation, refactoring, testing, and documentation. Work that previously required 30 commits might be completed in five. Work that required a team of 10 developers producing 200 commits per week can now be produced by three developers producing 40.
The output does not change. The commit count does.
What Is Actually Happening to Crypto Development
Electric Capital’s own data shows a 300% increase in developer activity within the AI-crypto sector over the past 12 months, even as aggregate commit counts fall. The developers are not leaving. They are shifting toward AI integration work, which produces different artifacts: deployed agents, onchain contracts, protocol integrations.
The Ethereum developer count by unique monthly contributors has declined modestly from its 2022 peak but remains above 2020 levels. Layer 2 ecosystems, particularly Base and Arbitrum, have seen developer counts grow steadily through 2025 and into 2026, measured by deployed contract addresses rather than commits.
The commit metric was always a proxy. It counted effort, not output. AI tools have broken the correlation between effort and commit count. The proxy is now measuring something that does not matter.
The Deeper Problem With Developer Metrics
The crypto industry has historically used commit counts for three things: investor confidence signals, protocol health assessments, and competitive benchmarking between chains. All three uses are now suspect.
A protocol with 50 developers using AI tooling aggressively can outship a protocol with 200 developers working manually. A commit graph that slopes down is not evidence of a dying ecosystem. It might be evidence of a more efficient one.
The industry needs new leading indicators. Deployed contract addresses per month, unique active builder wallets, protocol revenue per developer, onchain transaction volume per team member: these measure output, not effort. Some chains are already tracking these internally. None have made them public reporting standards. Until they do, every developer activity drop headline will misread a tooling transition as an ecosystem collapse.
The most dramatic version of this shift is that AI agents are now executing onchain autonomously. What was once developer tooling is becoming infrastructure. The commit count measures neither.
L2 developer concentration tells the same story. Base and Arbitrum are winning not because more developers are committing more code, but because the developers who remain are more productive and building where liquidity exists.
The TCB View
The 75% commit drop is real. The interpretation being applied to it is wrong, and the wrongness is not subtle. It is the same category of error as measuring a factory’s output by counting how many times workers pick up tools rather than how many units ship.
AI coding tools are the most significant productivity shift in software development since version control systems. The idea that a developer-intensive industry would be immune to that shift, or that the shift would register as growth in a metric designed for the pre-AI era, does not hold together.
The protocols that understand this early will stop optimising for commit counts and start building teams designed around AI-augmented output. The protocols that keep reporting raw commits as a health signal will mislead their own investors and themselves. The developer activity story of 2026 is not a crisis. It is a measurement problem. Solve the measurement, and the picture looks considerably different.
