Your Prompt Is Not Enough. Add a Gate.
A single prompt instruction loses to the rest of the prompt's bias. Programmatic quality gates enforce what prompts suggest.
Blog Posts
Ask a File, Don't Read the Whole Thing
How routing file reads through a cheaper model saves input tokens when your AI agent only needs a quick answer, not the full source code.
How We Cut CI git merge Fetch from 137s to 1s
Shallow clones break git merge. We replaced --unshallow with the GitHub Compare API and exponential --deepen, cutting fetch time from 137s to under 1s.
Let Your AI Agent Forget on Purpose
Why we built a tool that lets an AI agent explicitly drop file contents from its own context window, and the token economics behind it.
Runtime Downloads Are Hidden CI Costs
MongoMemoryServer downloads a 100MB binary on every CI run. Pre-caching the correct platform-specific archive to S3 eliminates that hidden latency.
One Web Fetch Ate 28% of Our PR Cost
One web fetch compounded across 22 agent turns and ate 28% of the entire PR's cost. We added Haiku as a summarization filter to stop the bleed.
Why PR Bodies Should Tell the Story
AI agents create PRs but leave reviewers guessing what happened. We added Claude-generated sections summarizing work done, bugs found, and trade-offs.
Why Retargeting a PR Explodes the Diff
GitHub's change-base-branch API only updates metadata. When sibling release branches are involved, the PR diff explodes with unrelated files.
Our Agent Had the Checklist and Ignored It
An AI agent had a 41-check quality checklist but kept making cosmetic edits instead of addressing failures. The fix was application-layer forcing.
Zero Changes Passed Our Quality Gate
Our verification gate let zero-change PRs pass because there were no files to check. The quality checks we built were never enforced.
Tests Can Go Stale at 100% Coverage
When source code changes but tests don't, the quality evaluation is outdated. We detect this automatically using content hashes from data we already have.
How We Finally Solved Test Discovery
A 5-tier scoring system validated against 17 real repos replaced flat scoring that broke with real data. Content-aware matching fills the gaps.
What 100% Test Coverage Can't Measure
100% line coverage means every line was executed, not that meaningful scenarios were tested. Quality checks fill the gap after 100%.
Test File Discovery Is Still Unsolved
Given an implementation file, which test files are relevant? We tried stem matching, content grepping, and hybrid discovery. Each fails differently.
39 Duplicate Jest Errors Cost Us $300
A single PR cost $300 because 39 identical Jest TypeErrors survived our log cleaning pipeline. Here's the root cause, the fix, and how we prevent it.
Vanilla Claude vs GitAuto Test Generation
We tested a calculator with vanilla Claude and GitAuto. Claude wrote 19 happy-path tests. GitAuto wrote 41 including adversarial edge cases.
Can You Guess What Tests a Calculator Needs?
A 40-line Python calculator. How many tests would you write? Most say 10-15. GitAuto generated 41. Here's what you missed.
What Are Adversarial Tests and Why Run Them
Adversarial tests push code with unexpected inputs - infinity, NaN, type mismatches, duck typing. Learn why they catch bugs happy-path tests miss.
Requesting Changes to GitAuto's Pull Requests
A real-world example of how GitAuto handles review comments and iterates on its pull requests based on user feedback.
How We Reached 92% Coverage with GitAuto
We dogfooded GitAuto on our own codebase and reached 92% line coverage, 96% function coverage, and 85% branch coverage.
9 Best Unit Test Agents 2025 Compared
Compare 9 leading AI-powered unit test agents and automated testing tools in 2025. Find the right solution for your development team.
Why Unit Tests Feel Like a Waste of Time
A developer's controversial take on unit testing sparked debate. Here's what the community really thinks and how AI changes the equation.
Unit vs End-to-End Testing: QA Focus Guide
Explore the key differences between unit and E2E testing, when each approach delivers maximum value, and how to build a comprehensive testing strategy.
Why You Need Both Manual and Automated Tests
Manual testing costs are spiraling and teams are drowning in repetitive tasks. Here's how to break the bottleneck without losing coverage.
What are DORA Metrics and Why They Matter
DORA (DevOps Research and Assessment) metrics are a key benchmark for measuring software delivery performance.