← back
Agents Need Feature Flags - Sachin Gupta
Takeaway
Treat agent behavior as deployable software with targeted rollouts and runtime controls that can promptly limit or stop actions.
Summary
- Agent releases should use familiar deployment controls such as canaries, segment targeting, rollback, and monitoring instead of exposing all users to every prompt or model change.
- The proposed flag taxonomy covers prompt variants, tool access, model routing, memory policy, autonomy levels, and kill switches, with child agents inheriting controls.
- Memory flags independently govern retention, scope, write access, and user visibility; autonomy flags separate suggestions, human-confirmed actions, and automatic execution.
- A real kill switch takes effect within seconds and is checked by in-flight work at its next decision point, while tool and model flags allow revocation or fallback without redeployment.
feature-flagsagent-operationscontrolled-rollouts
Original description
Most AI teams ship behavior changes to 100% of users on every deploy — no canary, no segment, no kill switch. Web teams stopped doing this in 2012. AI teams are about to learn why. Feature flags are table stakes in software engineering. In agent systems they're almost nonexistent — prompts, tool access, model selection, memory policy, and autonomy level all change globally the moment you ship. That's why your "small" prompt tweak just broke 12% of your users and you found out from a Discord screenshot. This talk walks the feature-flag patterns agents specifically need — beyond the standard boolean-toggle. We'll cover segment-targeted prompt variants, per-tool access flags, model-routing flags, autonomy-level flags (suggest vs. auto-approve vs. auto-execute), memory-policy flags, and the kill switch every agent system should have on day one but almost none do. Speakers: Sachin Gupta: Sachin Gupta is a Staff Software Engineer with 15+ years building backend platforms at internet scale, currently focused on the runtime trust boundaries that LLM coding agents blur and the creator of HeapLens, a Java heap analyzer extension used in 50+ countries. LinkedIn: / guptasachin1 GitHub: https://github.com/sachinkg12