Don't get one-shotted: Use AI to test, review, merge, and deploy code — Tomas Reimers, Graphite

Original: Don’t get one-shotted: Use AI to test, review, merge, and deploy code — Tomas Reimers, Graphite

595 views · Jun 27, 2025 · 5:45 min · Watch on YouTube ↗

Takeaway

Build AI-native outer-loop tooling (review, CI, merge, deploy) because the inner-loop speedup is creating a review bottleneck and security risk.

Summary

Tomas Reimers (Graphite co-founder) frames software development as inner loop (dev) + outer loop (test/review/merge/deploy); AI is transforming the inner loop (~46% of GitHub code via Copilot) but the outer loop is the new bottleneck.
Higher AI-generated code volume causes review/test/merge overflow plus AI hallucination and security-vulnerability risk that human review must catch — small companies now face problems only big companies had.
Diamond (Graphite's AI reviewer): codebase + change-history-aware, summarizes/prioritizes/reviews each PR, integrates with CI to summarize errors and propose corrections.
Key metric: Diamond comments are dismissed at <4% and accepted (integrated into PR) at ~52% — beating typical human comments at 45-50% (March update).
Argues the entire toolchain (not just the IDE) must become AI-native — review, merge queues, deployment — to handle 10x developer output.

code-reviewgraphitedeveloper-tools

Original description

As AI tools like GitHub Copilot and ChatGPT help engineers generate code at an unprecedented rate, the “outer loop”—reviewing, testing, merging, and deploying—becomes more vital than ever. Studies have shown that up to half of AI-generated solutions contain bugs or vulnerabilities, underscoring the continued importance of thorough, human-in-the-loop reviews. In this talk we'll take a look at how next-gen developer tools can harness AI not just for generating code, but also reviewing it. By thoughtfully integrating AI into that fully understands your entire codebase, teams can accelerate velocity without sacrificing quality.

Attendees will learn real-world strategies and best practices for establishing an “outer loop” that safely and efficiently deploys high volumes of AI-assisted code, without compromising reliability. We’ll also discuss pitfalls to avoid when integrating AI into existing pipelines.

About Tomas Reimers
Tomas Reimers, Forbes 30u30 class of ‘23, is the CPO and Co-founder of Graphite, the NYC DevTools startup that’s revolutionizing how the fastest-moving engineers build and ship software. Prior to co-founding Graphite, Tomas was a dev-tools engineer at Meta, developing a framework that supported the work of 200+ product teams and more than 1000 developers. He is passionate about all things software and will enthusiastically debate best engineering and architecture practices with you for hours.

Recorded at the AI Engineer World's Fair in San Francisco. Stay up to date on our upcoming events and content by joining our newsletter here: https://www.ai.engineer/newsletter