We Red-Teamed Our Own AI Agent and Found 14 Real Bugs
LangWatch red-teamed Langy, its own AI agent, across 15 attack categories and up to 20 turns each. Real transcripts, 14 bugs fixed, one issue still open.
LangWatch Blog8 min read
Developer findings
7 stories
The PR Hound: how we fixed PR review assignment with a daily agent
How LangWatch fixed PR review assignment with a daily agent: every PR gets one owner, one reviewer, and a Slack nudge until it moves. Human merges went from about 10 to 18.5 per workday.
LangWatch Blog6 min read
Background Agents on Slack: How we built our own Claude Tag before it was cool
Seven background agents run our own engineering, each scoped to one job in its own Slack channel, months before Anthropic shipped the pattern as Claude Tag.
LangWatch Blog9 min read
What happens when two engineering teams just... talk
Two AI engineering teams swap real stories on tool choice, LLM observability, BDD specs, and running parallel Claude coding agents without losing the codebase.
LangWatch Blog5 min read
Eat Sleep Append Repeat…
Re-architecting LangWatch on event sourcing and ClickHouse hit 6,000 events per second, real-time dashboards, and retroactive data improvements via replay.
LangWatch Blog6 min read
Four Refactors and a Funeral: Migrating a Live System to Event Sourcing
How LangWatch migrated a live platform to event sourcing on ClickHouse: four refactors, a fold-map-react pipeline, and a zero-downtime dual-write cutover.
LangWatch Blog20 min read
Internal Product vs Internalised Trauma: Supporting Event Sourced Systems
The tooling that keeps an event-sourced system supportable: Grafana dashboards, an ops console for blocked groups, projection replay, and time-travel debugging.
LangWatch Blog7 min read
Showing 7 of 7 stories
You're all caught up.
AiByts Weekly Brief
Get the top AI news and tools straight to your inbox every week.
Weekly · AI news
Join 15,000+ AI enthusiasts