AI Engineering Productivity is Anything But Normal

2026-07-21 · Source: Tomasz Tunguz · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering, Emerging Technologies & Innovation · Depth: Intermediate, quick

Summary

Recent data indicates a significant divergence in AI engineering productivity gains, categorizing companies into three distinct tranches. The first tranche, representing most companies, sees modest 20-30% productivity increases from AI IDEs, with some reports showing a 21% gain (Google) or 24% (GitHub), though Faros noted a 66% faster epic completion alongside a 54% rise in bugs. The second, "frontier" tranche achieves approximately 3x productivity by orchestrating AI agents across development tools. Examples include NVIDIA's 3x increase in committed code, Amplitude tripling weekly production commits, Anthropic's 2.5x code per engineer, and Replit tripling per-engineer output. The third tranche comprises "software factories," where AI mechanistically produces software. Nubank, for instance, achieved an 8x efficiency improvement and 20x cost reduction using Cognition's Devin for refactoring, while Goldman Sachs pilots Devin, estimating 3-4x the rate of prior tools.

Key takeaway

For Directors of AI/ML evaluating productivity tools, understand that simply deploying AI IDEs yields only 20-30% gains, often with increased bugs. To achieve 3x or greater engineering output, you must invest in building agentic "harnesses" that orchestrate AI across your development ecosystem. Consider piloting advanced AI software factories like Devin for large-scale refactoring to realize 8x efficiency improvements and substantial cost reductions.

Key insights

AI engineering productivity varies widely, from modest gains to 3x or more, depending on AI integration depth.

Principles

• Basic AI IDEs yield ~20-30% productivity gains.
• Orchestrated AI agents achieve 3x engineering output.
• AI software factories deliver 8x+ efficiency improvements.

In practice

• Integrate AI agents for context sharing across tools.
• Explore AI for large-scale code refactoring projects.
• Actively monitor bug rates with AI coding assistants.

Topics

• AI Engineering Productivity
• AI Code Generation
• AI Agents
• Software Factories
• Developer Productivity
• Cognition Devin

Best for: CTO, Investor, Entrepreneur, AI Engineer, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• 6 Ways to Enhance Developer Productivity with AI — Top teams achieve 100-150% productivity gains by restructuring around AI, not just adopting it, protecting human-centric tasks.
• ‘Tokenmaxxing’ is making developers less productive than they think — Increased AI code generation correlates with disproportionately higher code churn and revision rates, undermining productivity claims.
• Slow down to speed up: so much has changed in 6 months’ time — Rapid AI agent adoption boosts individual dev output but risks software quality and organizational integrity without human oversight.
• LAI #128: Developers Using AI Were 19% Slower. They Thought They Were 20% Faster. — The perception of AI's impact on developer productivity often misaligns with actual performance, especially on complex tasks.
• 😺 Anthropic: AI Is Building AI now — AI systems are increasingly handling execution tasks in their own development, fundamentally shifting human roles.
• What Happens When a Public Company Goes All In on AI — AI's rapid advancement fundamentally broke the "headcount equals output" law, enabling drastic productivity gains and organizational restructuring.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Tomasz Tunguz.