BentoLabs AI

The monitoring and learning layer for long-running agents

Y CombinatorSpring 2026ActiveB2BArtificial IntelligenceDeveloper ToolsB2BMonitoring

What they do

Problem

Teams deploying long-running AI agents struggle to detect silent failures and goal/tool drift and to debug and improve agent reliability in production.

Solution

A monitoring and learning layer that sits in the production loop to detect agent failures/drift, surface impacted users and root causes, and recommend prompt/skill/harness fixes.

Who it's for

Engineering and ML teams building and operating production AI agent systems at startups and enterprises.

Founders

Abhinav Soni

Founder

Previously- Hire #1 at Emergent (YC S24). Abhinav led the Agents team, helping scale from $0 to $100M ARR in just 8 months and hit #1 on SWE-Bench, twice. Unofficially he was called ‘the agent whisperer’. He built BentoLabs after realizing that for Agents whatever he couldn't see it wouldn't get fixed. And the current monitoring tools were not delivering the value they promised. So he built the layer that actually finds the silent failure, fixes them and closes the loop.

LinkedIn · X

Kaushik ASP

Founder

Kaushik was hire #2 at Emergent (YC S24), where he led full-stack engineering, built the initial infrastructure for long-running agents, and shipped the mobile-dev agent that converted ~50% better than any other. Previously Decathlon, Pazcare, and Synup.

LinkedIn · X