Yegor Denisov-Blanch

Yegor Denisov-Blanch is a Stanford Artificial Intelligence Laboratory research scientist and co-founder of P10Y who studies whether AI coding tools genuinely improve software-engineering productivity. His research uses real codebases and expert-informed evaluation to distinguish useful engineering output from misleading activity, technical debt, and AI-generated rework.

From entrepreneurship to engineering research

Denisov-Blanch left school in eighth grade, taught himself to code, and built a business-to-business e-commerce company in Spain. He subsequently studied operations research at Indiana University’s Kelley School of Business, graduating in 2016. At DHL, he worked on strategy and digital transformation and became chief of staff to the company’s chief executive for Europe, the Middle East, and Africa. He later earned an MBA from Stanford, conferred in January 2025.

At Stanford, Denisov-Blanch applied operations-research methods to software development and co-founded P10Y to commercialize related measurement techniques. His 2024 research on expert code-review evaluations, co-authored with Igor Ciobanu, Simon Obstbaum, and Michal Kosinski, developed models that approximate expert judgments of implementation time and code complexity. His subsequent research analyzed Git histories from more than 100,000 engineers across over 600 companies.

• Measure delivered work, not visible activity. Denisov-Blanch evaluates functionality, maintainability, complexity, refactoring, and rework instead of treating commits, pull requests, or lines of code as productivity. His controversial analysis of exceptionally low-output developers estimated that approximately 9.5% of engineers in one dataset were “ghost engineers,” a finding specific to his methodology and sample.

• AI productivity depends on engineering context. His large-scale developer-productivity research indicates that simpler greenfield work benefits more from AI assistance than complex tasks in established codebases. Unfamiliar programming languages, extensive dependencies, limited model context, bugs, and subsequent cleanup can diminish or reverse apparent gains.

• Codebase health shapes AI returns. His experimental Environment Cleanliness Index combines tests, typing, documentation, modularity, and code quality to examine why some teams benefit more from coding assistants. His engineering AI ROI framework distinguishes access from meaningful usage and pairs engineering output with guardrail metrics for rework, technical debt, risk, and team health. His AI engineering practices benchmark tracks progression from individual experimentation to shared workflows, autonomous tasks, and agentic orchestration.

• Verification must include downstream costs. A 2026 enterprise study he co-authored found that pull-request throughput more than doubled after an AI-driven productivity mandate while reviewer workload also increased; the observational design limits causal conclusions. His research on model consensus and truthfulness similarly argues that agreement among generated answers cannot substitute for independent verification.

Denisov-Blanch has also written about AI’s English-language bias, arguing that weaker Spanish-language performance creates economic disadvantages and warrants stronger multilingual research, data, and evaluation.