Built an outage investigation agent with Jev: ~5 s per incident, vs 32 and 119 s for general-purpose LLMs.
Right service 57/60 times, but often without checking what failed.
How do you check that an AI agent actually did the work, not just got lucky? https://t.co/iu0tPbwoaV
came to X to share serious engineering stuff.
40+ posts, 8 followers. thought maybe reach was the problem, so i boosted a post ๐
paid: 2,701 impressions, 2 profile visits
organic: 91 impressions, 2 profile visits
same destination, took the toll road ๐ฅฒ
@TypeLLM this is interesting. iโm building TypeRCA around incident investigation using decision models and currently it supports Jev, and I think it could be a good real-world place to test TypeLLM too. happy to try it
iโm Minglei. staff engineer at Uber, dad of two girls.
iโve spent years building and debugging production systems, and now AI agents. happy to share what iโve learned, give feedback on what youโre building, and learn from you too.
letโs connect and follow each other ๐