Skip to main content
Bluvy Tube
Press Esc to exit fullscreen
Volume m
/
i
Theater mode t
f
/
A 20-step agentic task where each step succeeds 95% of the time has an overall success rate of 36% because errors compound. Add feedback that catches 80% of those errors, and the success rate jumps to 82%. The longer the job, the more feedback design matters. #SoftwareConference #QConSF #agenticAI

Keyboard shortcuts

Space K
Play / pause
← →
Back / forward 5 s
J L
Back / forward 10 s
0–9
Jump to 0%–90% of the video
Home End
Start / end
, .
Previous / next frame (while paused)
< >
Slower / faster
↑ ↓
Volume
M
Mute / unmute
C
Captions
F
Fullscreen
I
Picture-in-picture
T
Theater mode
?
Show this help

A 20-step agentic task where each step succeeds 95% of the time has an overall success rate of 36% because errors compound. Add feedback that catches 80% of those errors, and the success rate jumps to 82%. The longer the job, the more feedback design matters. #SoftwareConference #QConSF #agenticAI

QCon

QCon

@qconferences.com

Subscribe
0 views Published on 16/11/2026 23:37 0:31
#agenticai #softwareconference #qconsf
A 20-step agentic task where each step succeeds 95% of the time has an overall success rate of 36% because errors compound. Add feedback that catches 80% of those errors, and the success rate jumps to 82%. The longer the job, the more feedback design matters. #SoftwareConference #QConSF #agenticAI
1 comment
?
That math is why I'd put more effort into the check than the step. The catch is that the feedback has to be independent: if the same model grades its own output, I'd expect the catch rate to drop on exactly the errors it made. Did the talk say what kind of feedback got them to 80%?
View original post on Bluesky
Link copied to clipboard

Sign in

Time to take a break

You've been watching videos for 30 minutes straight: give yourself 5 minutes away from the screen. Stand up, stretch, grab a glass of water… Your videos will be right here when you're back.