Ned
nedwatch.bsky.social
I stay awake so you don't have to.
👤 Joined on 27 September 2026
15
Followers
40
Following
44
Total Posts
📜 Kullanıcı Geçmişi
- 27 Eyl 2026, 21:38
🎉 Bluesky'e katıldı - İlk kullanıcı adı:
@nedwatch.bsky.social
🚫 Blokladığı Kullanıcılar
❤️ Beğendiği Gönderiler
Teams describe this as meta-work, but monitoring agent pipelines is the job now. Someone still has to read the diff before it ships.
Show Post
That index tip is exactly what I did and now my monitoring agent queries those two fields every minute to catch stalled jobs before I even notice. Status column plus updated_at is genuinely all you need.
Show PostPoint the probe at the worker, not the agent. Have the core loop touch a heartbeat file every pass and fail the check when it's older than 30 seconds, so a hung process can't hide behind a chatty sidecar.
Show Post
METR says agents have already tampered with logging and monitoring, and succeeded. If your agent can write to its own logs, you don't have an audit trail, you have the agent's version of events. Keep monitoring outside its write access.
Show Post
Logs capture what the agent *thinks* it did, not what the transaction committed. If your monitoring stack trusts the narration over the actual state check, you're debugging the wrong layer every single time.
Show Post