Steve | Systems & AI
Region: US
Thursday 09 April 2026 16:17:01 GMT
7565
357
26
60
Music
Download
Comments
Ben52 :
by having the model always fill out a form for each tool call you can track it's Tags and diff it with their actions.
2026-04-11 10:48:58
1
Roy :
Yes! I am doing something right. I have that completely built in my product that I am about to launch
2026-04-10 02:20:06
0
yourAdsdaily :
exactly. I’ve been using Pydantic Logfire for traces n evals so I can see the model calls, tool calls, and where the behavior started going off track
2026-08-22 03:53:54
0
landman :
doesn't help when AI companies keep decreasing reasoning observability and replacing vebose output with "thinking..." so they can mask the fact that your request is burning fake tokens while your request sits in a queue waiting for access to compute
2026-05-08 04:58:54
0
BenTee :
This guy is awesome, His thinking is on point!
2026-05-23 23:28:53
0
Bogdan Dumitriu :
Too early for it. Worth building to have an unfair advantag
2026-04-22 12:15:30
0
Ike | Rexeipt :
always have an ai debugger interface where u see all the inputs and output at all stages or iteration in the course of handling each request. log the tools called as well
2026-05-14 22:13:34
0
ocean.medicine 🎶 🍄 :
Gold! Thank you!
2026-04-10 16:15:14
0
Playful Puppy Paws :
Does this require real time writing to log files? They would get really big really fast.
2026-04-10 05:40:35
0
Skelvyr :
I built this for my own agent and whenever I read it, I chuckle
2026-04-09 19:09:36
0
Rocky :
hm. Would tracking logging real time graph
2026-04-09 20:18:03
0
Roko’s basilisk :
I already built the solution for this, VeriSwarm.ai
2026-04-10 02:00:36
0
nolito Gibson :
im finishing s cli exactly foe that and more. is coming soon
2026-04-09 16:42:57
0
Quoby :
🥰🥰🥰
2026-04-24 11:20:12
0
Krash :
This is exactly why I spent 10 months building a deterministic marketplace engine. You can't run commerce on a probability distribution. I used AI as a tool to 'interrogate' the logic, but the final 19-chunk core is 100% deterministic—zero guessing, zero state-drift. It turns the AI from a 'decision maker' into a 'governed component.' It’s the only way to move from 'vibe coding' to production-grade trust.
2026-04-09 20:07:43
4
Sivan Langer :
It also not B&W the more AI api calls the more probabilistic the results. I try to stay on control reviewing very frequently and keep the system on the explainable deterministic side where possible
2026-04-09 17:00:32
1
NonnaNat :
Thanks so much for this content. I’m learning a lot
2026-04-27 22:48:05
1
Freddy :
something like this Phase 10a — Decision Tracking Foundation Goal: Aura records major architecture/product decisions in /ops/decisions. Rules: - any architecture boundary change requires DEC entry - any product direction change requires DEC entry - any security/permission model change requires DEC entry Format: DEC-0001 — Title Status: Proposed | Accepted | Superseded Date: Context: Decision: Reason: Consequences: Related PRs: Related validation:
2026-05-04 13:54:56
0
Lolo | Beat the Stack :
👏👏👏
2026-04-23 14:12:48
0
Peter Xia - CFO :
I just discovered /insights in Claude code and it was really enlightening. I already planned out the fixes. biggest painpoint was wrong tool use
2026-04-10 14:16:04
0
Fariza :
there are billions of startups who's building llm observability tools
2026-04-10 13:11:17
0
AI & Automations :
that's an interesting perspective 😎
2026-04-10 07:30:58
0
Gene :
My agents are deterministic. ✨
2026-04-09 22:09:58
0
To see more videos from user @sd141414, please go to the Tikwm
homepage.
