Agents Need Feature Flags

on YouTube ↗19m

" summary

Agents ship six concrete flag types: prompt variant flags, per-tool access flags, model-routing flags, memory-policy flags, autonomy-level flags (suggest vs auto-approve vs autoexecute), plus an agentwide kill switch. Demos show flipping a tool midconversation and stopping a runaway LangChain loop that burned $47,000; metric targets include kill-switch mitigation under 5 minutes, prompt rollback under 30, and a canary error-delta >2% at 5% rollout blocks promotion. Uses existing backends like launch darkly or unleash lift.

AI-written summary · about

" chapters (10)

1. 0:00Agent feature flags
2. 1:05Why agents need flags now
3. 4:37Behavior surfaces needing flags
4. 10:19Demo: tool access flag
5. 11:22Demo: kill switch for agents
6. 13:29Rollout playbook for flags
7. 15:44Common flag implementation failures
8. 16:25Business case and regulatory impact
9. 17:49Key takeaways for agent safety
10. 19:00Q&A

" related talks

• 06mFeature Flag Foul-upsHeidi Waterhouse · SpringOne Platform 2017
• 43mFeature Flags and GenAI parameterization and experimentationJosh Klein · Build Stuff 2024
• 27mAI Ships Fast and Breaks Things — Feature Flags Keep It ControlledJonathan Norris · OpenFeature Summit NA 2025
• 56mWhen Feature flags go badEdith Harbaugh · NDC Sydney 2017
• 11mFeature flag use cases you haven't heard about yetJoy Ebertz · LeadDev Berlin 2022
• 44mFlags of Flexibility - Unlocking Rapid IterationOussama Zaki · Agile meets Architecture 2025
• 40mWhen Feature flags go badEdith Harbaugh · NDC Oslo 2017
• 08mThe Art of Feature FlaggingDianing Yudono · DevOps.js Conf 2021