AI code generation
AI coding tools can change how quickly developers work and which work they attempt. METR's 2026 follow-up found serious selection problems in estimating the gain. We would measure accepted software, review effort and defects in a defined workflow. Adoption figures, lines generated and an old study's headline number are not enough on their own.
Research snapshot · updated Nov 12, 2026
State of the art (2026)
By mid-2026 AI code generation is incumbent infrastructure, not a bet. GitHub Copilot crossed 4.7 million paid subscribers in January; Cursor (Anysphere) hit roughly $2bn ARR and drew a $60bn SpaceX acquisition offer in June, having earlier raised at a $29.3bn valuation. Anthropic Claude Code, GA since May 2025, runs in production at Stripe and Ramp, while Cognition opened Devin to all engineering teams in February after folding in Windsurf. The live contest is no longer adoption but autonomy: how far agentic, multi-step retrieve-reason-edit-verify workflows displace human-authored pull requests, and whether incumbents (GitHub, Anthropic, OpenAI) absorb the standalone IDE layer.
Core thesis
The measured effect of AI coding tools depends heavily on the task and on who is being studied. In a randomised study by METR, 16 experienced open-source developers worked on 246 tasks, and tasks where early-2025 AI tools were allowed took 19% longer. METR's February 2026 update says newer tools are hard to measure, because developers choose which work to do with AI. Other studies have reported large gains on specific routine tasks, and those results do not carry over to all professional work. The useful measure for a team is its own accepted code, review time and defects.
Explore the deeper analysis in CanaryIQ
Explore what is covered below. These previews show the structure; log in to read the full analysis.
Signal stack
Evidence stacked leading → lagging
10 signals
Signal stack
Evidence stacked leading → lagging
Technology-native KPIs
Metrics that predict trajectory, tracked over time
6 tracked
Technology-native KPIs
Metrics that predict trajectory, tracked over time
Landscape map
Who builds what, and who depends on whom
93 players · 6 layers
Landscape map
Who builds what, and who depends on whom
Catalyst calendar
Dated events that will move the position
6 ahead
Catalyst calendar
Dated events that will move the position
Technology roadmap
Milestones on the path to maturity
8 milestones
Technology roadmap
Milestones on the path to maturity
Watchlists
Companies, people and papers, each with a condition for removing it
20 · 20
Watchlists
Companies, people and papers, each with a condition for removing it
Decision frameworks
What our view means for your job
Locked
Decision frameworks
What our view means for your job
Thesis changelog
When our view changed, and why
9 updates
Thesis changelog
When our view changed, and why
Change our mind
3 disconfirming conditions
You've read the verdict. The file is much deeper.
The full signal stack, technology-native KPIs tracked over time, the landscape of who depends on whom, the dated catalyst calendar, decision frameworks for every desk, live watchlists and a record of every time our view on AI code generation has changed. All of it is in CanaryIQ.