I TRUST MY CODING AGENTS WITH PRODUCTION SECRETS NOW

Summary from the TLDR newsletter

Security is always a productivity-risk tradeoff. Not giving access to agents can slow things down significantly. Frontier models are getting better and better at distinguishing real from malicious instructions in the content they read. We may end up trusting agents with access more than humans as models improve.