Thierry Donneau-Golencer

We ran AI review on 28 merged PRs that humans had already approved. Every one had critical findings. Here’s what "LGTM" missed and how we changed our code review process.

Learn how Faros’s Time Machine helps companies cut AI coding costs, maintain quality, and choose the right model for every task.

Restrictions on AI-generated code are spreading across open source projects as review queues overflow. Our Speed Trap report shows what that costs teams and what to do instead.

A work restart is a task sent back to development after review, QA, or deployment. Learn what restarts cost in developer time and AI tokens, and how to reduce them.

Optimize AI coding by routing each task to the right model and harness, using benchmarks from your own merged code to reduce cost per verified outcome.

Why do coding agents fail? We analyzed 4,000 errors across 6 models and discovered the real culprits.

LLM judges give higher scores to certain coding styles, independent of whether the code works. We measured the bias, tested causes, and calibrated for it. See how we did it.

The same flaw is hiding in your own backlog, at a higher rate. Here's why graded scoring catches what pass/fail testing misses.

The AI code evaluation framework behind our open vs. frontier model test: rubric-based scoring, a blinded LLM judge, and validation on real SWE-bench data.

Faros introduces Token Intelligence. Trace every AI token to the work it produced, classify spend by efficiency, and decide which tools and models to keep, scope, or cut. The first step from tokenmaxxing to outcome maxxing.

AI coding assistant pricing changes are reshaping engineering budgets. Build a defensible ROI calculation by tool, team, and model before your next renewal.

Track Claude Code usage, adoption, and engineering impact using Faros's observability platform.

There's a better way to measure AI productivity than counting lines of code. Focus on outcome metrics that prove business value: cycle times, quality, and delivery velocity. Learn why lines of code fails as an AI productivity metric, what outcome-based alternatives actually work, and when tracking AI code volume matters for governance and risk management.

Understand Claude Code's context window and usage limits, what really drives token costs, and how to manage AI coding spend by tying usage to engineering ROI.

Learn what context engineering is, how it differs from prompt engineering, and 5 best practices for building more reliable AI coding agents.

Get comprehensive metrics to measure productivity gains from AI coding tools. The Faros Iwatani Release helps engineering leaders determine which AI coding assistant offers the highest ROI through usage analytics, cost tracking, and productivity measurement frameworks.

The Bain Technology Report 2025 reveals why AI coding tools deliver only 10-15% productivity gains. Learn why companies aren't seeing ROI and how to fix it with lifecycle-wide transformation.

Discover the 5th DORA metric: Rework rate. Learn what it is, why it matters in the AI era, and how to start tracking it today. Get industry benchmarks, see what good looks like, and find practical tips to reduce wasted engineering effort and boost performance.

Most AI tools don’t improve delivery. The GAINS framework helps engineering leaders measure real productivity impact across 10 transformation dimensions—from throughput to organizational efficiency.

Uncover the limits of DevEx surveys and learn how to ensure insights are valid, actionable, and grounded in operational reality.

Finally, automation is optimizing software development capitalization, a vital but laborious accounting practice.

For many organizations, acting on employee surveys is challenging due to problems in the survey itself and the partial picture it paints. A novel approach is blending survey and systems data to create a more holistic understanding.

A simple checklist can help engineering managers achieve net positive gains in team productivity, lead time, and quality.

GitHub Copilot is one of the fastest adopted tools in the history of software development. One year after its release, over 1 million developers and 20,000 organizations are using the tool. But how to measure its impact on your engineering operations? Read on..