Blog
Research and insights on AI-generated code.
Developer Productivity in 2026: The Metrics That Changed When AI Joined the Team
DORA velocity is up. Bugs per dev are up 54%. PR review time is up 441%. The 2026 data on what AI broke in productivity measurement and the four frameworks teams use instead.
AI Coding Tools Ranked by Real-World Output Quality — April 2026
Claude Code scores 80.8% on SWE-bench Verified. Cursor hit $2B ARR. Copilot has 4.7M subscribers. 7 tools ranked by benchmarks that actually predict production quality — plus security pass rates, hallucination risk, and per-tool attribution data.
Dario Amodei Said 90% of Code Would Be AI-Written. The Deadline Passed 6 Months Ago.
Amodei predicted 90% AI-written code by September 2025. GitHub reports 51% AI-assisted. GitIntel scans average 5.8%. The data behind the most debated prediction in software — and why the metric was unfalsifiable to begin with.
AI Tools Leave Invisible Fingerprints in Your Git History. Researchers Can ID Them at 97% Accuracy.
A January 2026 arXiv paper analyzed 33,580 PRs from 5 AI coding agents (Codex, Copilot, Devin, Cursor, Claude Code) and identified which tool wrote the code at 97.2% F1-score — even when developers didn't disclose it.
Shadow AI: 65% of Enterprise AI Tools Are Unapproved. Source Code Is the #1 Data They're Leaking.
Harmonic Security: source code is 30% of all sensitive data leaked via shadow AI — ahead of legal docs and M&A data. IBM: shadow AI breaches cost $670K more. 90% of enterprises use AI. 37% have governance policies. EU AI Act enforcement: August 2026.
AI Makes Developers 19% Slower. They Think They're 20% Faster.
METR's RCT: experienced developers are 19% slower with AI while believing they're 20% faster — a 39-point gap. BCG's March 2026 HBR study: 14% suffer 'AI brain fry,' error rates 39% higher. The productivity trap has peer-reviewed data.
AI Writes Your JSON. 1 in 4 Times, It's Wrong. ICLR 2026 Has the Data.
University of Waterloo's StructEval tested 12 AI models on 2,035 structured output tasks. GPT-4o tops out at 76%. Text→Mermaid collapses to 18.9%. The format-specific failure crisis is now peer-reviewed.
Your AI Coding Agent Gets Dumber the Longer It Works. Here's the Proof.
All 18 frontier models tested by Chroma Research degrade with context length. A January 2026 arXiv paper found models miss their advertised window by 99%+. Some start failing at 1,000 tokens. Here's what that means for AI-generated code.
AI Coded It. You Can't Debug It. Anthropic Has the Data.
Anthropic's Jan 2026 RCT: AI users scored 17 points lower on post-task skills tests than controls. 50% vs 67%. No measurable speed gain. The skill atrophy is documented and growing.
AI Writes Your Code. Nobody Understands It. Here's the Data.
88% of developers say AI worsens technical debt. Maintenance costs hit 4× by year two. Addy Osmani named it Comprehension Debt — the gap between how much code exists and how much your team actually understands.
AI Writes Your Tests. Coverage Goes Up 40%. Bugs Stay.
AI tools boost test coverage 40% and cut test-writing time 50%. But defect rates haven't moved. 1.75× more logic errors in AI code, AI-generated tests that validate the wrong behavior, and the 'coverage theater' pattern showing up in production postmortems.
AI Tools Hallucinate Package Names. Attackers Are Registering Them.
LLMs suggest non-existent npm/PyPI packages in ~5–20% of coding tasks. Attackers register those names with malicious payloads. Slopsquatting is the newest supply chain attack — and 51% AI-assisted commits means the attack surface is now enormous.
Your AI-Generated Code Can't Be Copyrighted. But It Can Infringe Someone Else's.
Supreme Court March 2026: pure AI code has no copyright protection under US law. But it can still infringe yours. 51% of GitHub commits are AI-assisted with zero documentation. EU AI Act enforcement hits August 2026.
Code Review Automation in 2026: What Teams Actually Ship With AI
Greptile catches 82% of bugs. Graphite catches 6%. CodeRabbit reviewed 13M PRs across 2M repos. Microsoft, Cloudflare, Asana, monday.com — what production AI review actually looks like.
Developers Are Opening 98% More PRs With AI. Their Reviews Got 91% Longer.
CodeRabbit reviewed 13M PRs: AI-coauthored PRs have 1.7× more issues. Incidents per PR up 23.5% YoY. PR size up 154%. The review bottleneck data nobody's talking about.
73% of Developers Use AI Every Day. 97% Don't Fully Trust It. Here's the Data.
15,000 developers surveyed. 73% use AI coding tools daily — up from 18% in 2024. Only 3% highly trust the output. METR measured the actual productivity gap. The verification problem nobody's solved.
The AI Coding Benchmark War Has a Legitimacy Crisis. Here's the Data.
Claude Opus 4.5 scores 80.9% on SWE-Bench Verified and 45.9% on SWE-Bench Pro. 35-point gap. OpenAI abandoned Verified. 59.4% of its hardest problems have flawed tests. The benchmark everyone cites is broken.
Cursor Hit $2B Revenue. Copilot Has 4.7M Users. Neither Leaves a Trace in Your Git History.
51% of GitHub commits are AI-assisted. Cursor has $2B ARR. GitHub Copilot has 4.7M subscribers. None of them leave attribution in your git history — and that's a compliance crisis.
Vibe-Coded Apps Are Shipping With Critical Security Holes. Here's the Data.
10.3% of Lovable-generated apps had critical RLS flaws. 21% of YC W'25 companies are 91%+ AI-generated. 63% of vibe coders are non-developers. The security bill is coming due.
AI Wrote Your Code. Now Your Churn Is Up 41%.
GitClear analyzed 153M lines of code and found AI tools have driven churn up 41%, duplication up 4×, and refactoring to a 10-year low. The quality crisis is measurable and growing.
AI Code Gets Smarter. It Doesn't Get Safer.
Veracode Spring 2026: syntax pass rates hit 95%, security pass rates flat at 45–55% since 2023. 35 CVEs from AI code in March alone. XSS fails 86% of the time. The divergence is now official.
AI PRs Are Flooding Open Source. Maintainers Are Drowning.
AI PRs generate 1.7x more issues than human ones (10.83 vs 6.45 findings). PRs up 20% YoY. cURL killed its bug bounty. LLVM added AI-disclosure rules. The review bottleneck is here.
The AI Agent Cost Crisis Nobody's Talking About
Developers are 19% slower with AI tools (METR study). 45% of AI code has OWASP vulnerabilities. Your team spends $15-25K/month on AI coding tools. Nobody's tracking the ROI.
What's Trending on GitHub: March 2026
OpenClaw 263K stars. Ollama 162K. n8n 150K. Dify 130K. GitHub has 4.3M AI repos (+178% YoY). Three patterns dominate: local-first, open-source infrastructure, and the agent plugin explosion.
The MCP Ecosystem Just Hit 8,600 Servers. Here's What That Means.
873% growth in nine months. 8,600+ MCP servers. Zero standardized security vetting. The protocol powering AI agents has a trust problem — and the EU AI Act deadline is four months away.
How last30days-skill Built a 14K-Star Research Engine in One File
14,449 stars. 10 platforms. One SKILL.md file. We break down the architecture of the most popular Claude Code skill on GitHub — query detection, source tiering, scoring, and 10 patterns you can steal.
5 Viral Repo Patterns Every AI Builder Should Know (March 2026)
We analyzed 5 repos that collectively gained 150K+ stars this month. The patterns they share are about memory, not features — subconscious hooks, temporal decay, research skills, peer messaging, and swarm orchestration.
The Official Claude Code Telegram Plugin Has 3 Tools. Here's What's Missing.
We mapped every Telegram Bot API capability against the official plugin. 13 features missing. We built them — inline keyboards, scheduler, document handling, and more.
We Scanned 13 Popular Open Source Repos for AI-Generated Code. Here's What We Found.
Deno is 41% AI-coded. Claude Code appeared in 99% of AI commits. Data from 6,500 commits across Next.js, Ruff, LangChain, Ollama, and more.