Skip to content
Sign in
Artwork for Adventures in DevOps
Adventures in DevOps · February 20 · 32 min

DevOps trifecta: documentation, reliability, and feature flags

Share Episode We dive into the shifting landscape of developer relations and the new necessity of optimizing documentation for both humans and LLMs. Melinda Fekete joins from Unleash, and suggests transitioning to platform to help get this right by utilizing LLMs.txt files to cleanly expose content to AI models. The conversation then takes a look at the June GCP outage, which was triggered by a single IAM policy change. This illustrates that even with world-class CI/CD pipelines, deploying code using runtime controls such as feature flags is still risky. Feature flags can't even save GCP and other cloud providers, so what hope do the rest of us have. Finally, we discuss the practical implementation of these systems, advocating for "boring technology" like polling over streaming to ensure reliability, and conducting internal "breakathons" to test features before a full rollout. 💡 Notable Links: Diátaxis - Who is article this for? Fern - Docs Platform CloudFlare - Feature Flag causes outage AWS - Graceful degredation Building for 5 nines reliability Episode: Latency is always more important than freshness Episode: DORA 2025 Report 🎯 Picks: Warren - Show: Bosch - LA Detective procedural Melinda - Wavelength - Party Game

0:00-32:00
Play in
Transcript file

transcript

No transcript — this publisher did not publish one.

show notes

Share Episode         
         
We dive into the shifting landscape of developer relations and the new necessity of optimizing documentation for both humans and LLMs. Melinda Fekete joins from Unleash, and suggests transitioning to platform to help get this right by utilizing LLMs.txt files to cleanly expose content to AI models.

         

The conversation then takes a look at the June GCP outage, which was triggered by a single IAM policy change. This illustrates that even with world-class CI/CD pipelines, deploying code using runtime controls such as feature flags is still risky. Feature flags can't even save GCP and other cloud providers, so what hope do the rest of us have.

         

Finally, we discuss the practical implementation of these systems, advocating for "boring technology" like polling over streaming to ensure reliability, and conducting internal "breakathons" to test features before a full rollout.

         
💡 Notable Links:         
🎯 Picks:         
links11

more episodes

All episodes