Blog

Six multi-agent frameworks, measured across coding and reasoning benchmarks. Then the pattern both the loudest sceptic and the loudest proponent ended up agreeing on: many agents may think, one agent writes.

81% of engineering teams report spending more time in code review since adopting AI tools. The advice assumes checking is the cheap half. Measurement says that is where the hours went.

Our AI code reviewer found a SQL injection, described it accurately, and set the verdict to approve in the same response. Here is what we changed.

Copilot now takes part in more than one in five code reviews on GitHub. The research says AI reviewers earn their keep - and that the tuning which makes them usable suppressed about a fifth of real vulnerabilities on a standard benchmark.

We planted eight defects in our own CI. Three were caught. One gate was skipping 133,874 of 149,516 links and reporting green. Here is the command that proves it.

A change swapped a made-up number for a confident, wrong explanation of how Propshaft works. Spotting the difference is the thing you are actually hiring for.

A fake think tank published 100+ reports built to be repeated by chatbots, behind a $900,000 government contract. The same economics reached your blog. Four checks you can run.

RubyLLM retries POSTs that Faraday would not, and honors Retry-After with no upper bound. A 429 can park a Rails request for as long as the provider says.

RubyLLM is a model client with Rails generators. Langchainrb is retrieval plumbing. Pick by whether you query your own documents, not by provider count.