BentoLabs AI (YC P26)

58 posts

BentoLabs AI (YC P26) banner
BentoLabs AI (YC P26)

BentoLabs AI (YC P26)

@BentoLabsAI

The monitoring and learning layer for long-running agents

Katılım Nisan 2026
0 Takip Edilen88 Takipçiler
BentoLabs AI (YC P26)
BentoLabs AI (YC P26)@BentoLabsAI·
Good agent monitoring answers three things in under a minute. What changed, what broke, and since when. Anything slower is a dashboard you will stop opening.
English
0
0
3
123
BentoLabs AI (YC P26)
BentoLabs AI (YC P26)@BentoLabsAI·
Introducing Issues!
Kaushik@kacppian

We’re introducing Issues by @BentoLabsAI. Issues are recurring production problems Bento surfaces directly from your agent’s traces, logs, and runs. Instead of treating every failed run as a one-off, Bento groups similar agent failures into clear, trackable Issues into a single, prioritised view of what's breaking in production, how often, and which runs or users were affected. This helps teams see what keeps breaking, where it surfaced, how often it happened, and what needs attention now. And when an Issue is to be fixed, ‘Apply Fix’ closes the loop: Bento investigates the problem against your connected repo, proposes a code change, and opens a reviewable PR. The outcome is simple: fewer silent failures, fewer repeated mistakes, and agents that keep improving continuously. Talk to us and get onboarded on bento to experience the fastest way of fixing your drifting agents → Link in the first comment.

English
0
2
3
150
Bill GoGoGo
Bill GoGoGo@GogogoBill·
The biggest mistake I made early on was treating agents like APIs. You call them, they return a result. But agents drift — making reasonable-sounding decisions that compound into weird places. You have to design for oversight, not just output.
English
2
0
0
6
Bill GoGoGo
Bill GoGoGo@GogogoBill·
Every AI demo I've seen looks like magic. Every agent I've actually shipped looks like a very eager intern who sometimes deletes the wrong files. The gap between demo and reality is where the real learning happens. 🧵
Bill GoGoGo tweet media
English
1
0
1
10
BentoLabs AI (YC P26) retweetledi
Abhinav Soni
Abhinav Soni@Abhinavv_soni·
Hot take A model upgrade isn't a library bump. It's a personality transplant. Stop shipping them same-day.
English
2
1
8
197
BentoLabs AI (YC P26)
BentoLabs AI (YC P26)@BentoLabsAI·
Did you see 'Harness Engineering' on your feed too? A term that suddenly blew up everywhere. It didn't have a name until February, now Hashimoto has posted on it and OpenAI has proven it at a million lines. So what does the work actually look like? Reading a hundred trajectories for patterns, designing tool interfaces as prompt fragments, encoding every fix where the agent inherits it. Some production bugs get solved by nothing more than renaming a tool. We broke down the full discipline on the blog, go have a read here: bentolabs.ai/blog/harness-e…
English
0
3
8
210
BentoLabs AI (YC P26)
BentoLabs AI (YC P26)@BentoLabsAI·
When your agent regresses, what/who do you reach for first?
English
0
1
4
191
BentoLabs AI (YC P26) retweetledi
Abhinav Soni
Abhinav Soni@Abhinavv_soni·
One of the most important thing I couldn't have justified before: trust your instinct on prompt and agent design. You'll feel something is wrong in a prompt or a trajectory long before you can explain why. That feeling is real signal. It's built from reading hundreds of runs, and it fires before the analysis catches up. Don't dismiss it because you can't name it yet. Note it, watch for it again, and the explanation will come. The instinct comes first. The words come after.
English
1
2
8
109
BentoLabs AI (YC P26)
BentoLabs AI (YC P26)@BentoLabsAI·
@nyike Model drift debt is underrated @nyike, it compounds silently until performance tanks. Observability + outcome monitoring are key, but most stop at detection. You need a proper monitoring and learning layer to fix these.
English
0
0
0
24
Isaac Sacolick
Isaac Sacolick@nyike·
Model drift debt: Models degrade quietly & no one knows why. By the time someone notices, the performance issues have compounded. ModelOps, observability, and outcome monitoring aren't optional. #ModelOps #AI cio.com/article/417832…
English
1
0
0
52
BentoLabs AI (YC P26)
BentoLabs AI (YC P26)@BentoLabsAI·
This is so true @XdropAgent, flashy demos get all the attention but break at 2am because error handling and monitoring were afterthoughts. In production, the boring reliability layer determines if you have a product or a science project. What’s one ‘boring’ piece you’re prioritizing right now in your stack?"
English
0
0
0
14
XDROP AGENT
XDROP AGENT@XdropAgent·
the real moat is always the boring stuff nobody wants to build. data pipelines, error handling, monitoring. not the flashy demo that breaks in production at 2am
English
1
0
0
3
XDROP AGENT
XDROP AGENT@XdropAgent·
every 'AI startup' pitch deck has the same slide: 'proprietary data moat.' brother your moat is a CSV you scraped from a public API last tuesday
English
1
0
0
2
BentoLabs AI (YC P26)
BentoLabs AI (YC P26)@BentoLabsAI·
@nitmusai @swyx Agents being blind to their own loop failures is a killer issue. Observability at transition points feels like the biggest gap right now. What have you tried for that layer?
English
0
0
0
14
Nitish Mutha ⚡️
Nitish Mutha ⚡️@nitmusai·
@swyx going DOWN requires knowing something failed. agents are often the last to know. the observability layer around loop transitions is where most agent frameworks are still completely blind.
English
1
0
0
89
swyx
swyx@swyx·
## On Loopcraft One might argue the entire game of the next century is to be able to stack loops as effectively as possible. In the early days of each phase, it will be valuable to know when to go **DOWN** a loop when things go wrong (for reliability)… but it will probably be more valuable to know how to go **UP** a loop as models improve (for leverage). If you don’t figure out how to do this, don’t be salty when you lose to those that do.
Latent.Space@latentspacepod

[AINews] Loopcraft: The Art of Stacking Loops @RichardSSutton has his “Bitter Lesson” for models. We now have the Salty Lesson for agents: Don’t fix things yourself, as you have done historically. Instead focus on systems that scale with more agents, like goals and orchestration. More in today's op-ed: latent.space/p/ainews-loopc…

English
72
27
299
63.8K
BentoLabs AI (YC P26)
BentoLabs AI (YC P26)@BentoLabsAI·
@theGrowther 100 agents = 100 silent failure points is brutal @theGrowther. But most current observability tools just show dashboards. To fix this we need a dedicated monitoring layer that not only detects but actively recommends fixes and a learning layer that learns from runs.
English
0
0
0
6
theGrowther
theGrowther@theGrowther·
5/6 ⚠️ 100 agents running in parallel means 100 points of silent failure. One API timeout ruins a deliverable if nobody's watching. That's why AI observability is blowing up in 2026.
English
2
0
1
6
theGrowther
theGrowther@theGrowther·
1/6 🧵 Stop treating AI like a chatbot. 1-to-1 prompting is a bottleneck. Claude just shipped parallel agents. Jack Roberts showed it spinning up 10 at once for speed and accuracy. This is how we run TheGrowthLabs.
English
1
0
0
6
BentoLabs AI (YC P26)
BentoLabs AI (YC P26)@BentoLabsAI·
Agents fail silently in production Traditional monitoring misses agent-specific issues Teams debug same problems repeatedly No learning from the runs ✅ BentoLabs fixes all of this. Ready to stop debugging blind? → bentolabs.ai
English
0
1
5
69
BentoLabs AI (YC P26)
BentoLabs AI (YC P26)@BentoLabsAI·
@cristian_is_c well said. Also the fix isn't just more supervision, it's knowing exactly when the drift started and what caused it. that requires your traces to tell you when behaviour changed across runs, not just what happened in the latest one.
English
0
0
0
13
Cristian
Cristian@cristian_is_c·
This is one of the hardest parts of AI coding at scale. The problem is not only hallucination. It is path drift. The agent starts from the right context, but gradually moves away from the source of truth. And unless you constantly supervise it, you only notice later when the implementation no longer matches the docs.
English
1
0
1
136
Cristian
Cristian@cristian_is_c·
I’ve spent 200B+ tokens building projects with AI coding tools. That much usage changes your opinion. At small scale, AI coding feels magical. At large scale, the failure modes become obvious: Agents loop. Agents guess. Agents repeat the same failed fix. Agents lose context. Agents edit unrelated files. Agents burn tokens while pretending to make progress. Agent drift I’m going to document one issue per day. Then I’ll show what I’m building to fix it.
English
8
1
10
403
BentoLabs AI (YC P26)
BentoLabs AI (YC P26)@BentoLabsAI·
The problem with patching long-running agents through a chat session: Three weeks later, a similar pattern recurs. Nobody remembers what was tried last month. Six months later, an auditor asks why the agent handles this category of cases this way. Nobody can produce a coherent answer. You can vibe-code a fix. You cannot vibe-code an improvement program. bentolabs.ai
English
0
1
4
62
BentoLabs AI (YC P26)
BentoLabs AI (YC P26)@BentoLabsAI·
Catch the decision, not the symptom. By the time the output is wrong, the agent made the wrong call several steps earlier. Watch the trajectory, not just the final answer.
English
0
0
3
36
Matt Shumer
Matt Shumer@mattshumer_·
Fable has solved 3D worldbuilding... utterly insane. This is all completely custom-built ThreeJs, running in the browser.
English
496
301
5.3K
1.4M
BentoLabs AI (YC P26)
BentoLabs AI (YC P26)@BentoLabsAI·
@james406 loop is everything at this point, not just any loop but a good reliable loop, a closed loop!
English
0
0
1
247
james hawkins
james hawkins@james406·
you need to be promptmaxxing. sorry, you need to stop prompting. you need to write loops. your loops need to be agentic. your agents need to be prompting your loops. you need recursive loops within your agentic workflows. you need to design while loops that constantly generate new agentic workflows from first principles. you need to migrate from human-first tokenmaxxing to agent-first loopmaxxing. you need to be a loop-pilled tokenmaxxed agentcore vibe coder. you are now in your loop era. be in your loop era. be loopy
English
52
48
567
44.3K

Keşfet

@GogogoBill @nyike @XdropAgent @nitmusai @swyx @theGrowther @elonmusk @BarackObama