Tip-offs
•
Each week, we surface promising startups & founders through data signals.
This week, we're spotlighting startups that test AI agents →
AI agents demo well, deploy badly
Every company wants agents in production. Few trust them enough to get there.
The demo-to-production gap is the whole story in AI right now. While 57% of companies say they have agents in production, quality and reliability consistently rank as the top barrier preventing broader deployment.
When an agent books the wrong flight, misfiles a contract, or executes any high-stakes action incorrectly, "usually right" is unacceptable. Enterprises need provable reliability before granting autonomy.
This is exactly what a good evaluation layer provides. It helps build the confidence needed to move AI agents from testing into actual use.
AI Evals Startups
Startups building the eval, benchmarking, and monitoring layer for AI agents. Ranked by website-visit growth over the past 3 months.
Galtea | 959% growth in 3 months
An AI evaluation platform that stress-tests agents against auto-generated test cases and adversarial inputs, catching accuracy, safety, and compliance failures before launch. Raised a $3.2M Seed in March 2026.
Barcelona, Spain
Kashikoi | 741% growth in 3 months
A no-code simulation engine that generates AI users to interview and benchmark agents across multi-turn conversations. Raised a $500K Pre-Seed in June 2025.
San Francisco, US
Benchflow | 359% growth in 3 months
An AI evaluation platform with standardized, reproducible benchmarks and a sandboxed runtime for testing how agents and models perform before deployment. Raised a $1M Pre-Seed in February 2025.
San Francisco, US
Swept AI | 190% growth in 3 months
An AI agent supervision platform that adversarially evaluates and certifies agents before rollout, then monitors them in real time for drift, jailbreaks, and compliance failures. Raised a $1.4M Pre-Seed in August 2025.
Saginaw, US
ZioSec | 85% growth in 3 months
An offensive security platform that continuously red-teams LLM agents and agent-to-agent workflows, simulating prompt injection and other real-world attacks to surface vulnerabilities. Raised a $2.1M Seed in July 2025.
Boulder, US
AI Evals Founders
Recently funded founders building the eval, benchmarking, and monitoring layer for AI agents.
Ankur Goyal
Built databases at MemSQL, then founded Impira (acquired by Figma, where he led the ML Platform). Now founder & CEO of Braintrust, the eval and observability platform behind AI at Notion, Stripe, Vercel, and Cloudflare. Raised an $80M Series B in Feb 2026.
San Francisco, US
Rez Havaei
Ex-NVIDIA and Cohere AI researcher. Founding engineer at Vals AI. Now co-founder & CEO of General Analysis, running automated red-teaming & adversarial evaluation to harden agents. Raised a $10M Seed in Apr 2026.
San Francisco, US
Brooke Hopkins
Built autonomous-vehicle evaluation infrastructure at Waymo. Now founder & CEO of Coval, a simulation platform that stress-tests voice and chat agents. Raised a $28M Series A in Jul 2026.
San Francisco, US
Mahesh Sathiamoorthy
Ex-Google DeepMind (staff software engineer). Now co-founder & CEO of Bespoke Labs, building simulated enterprise environments that train and test reliable agents. Raised $40M Seed + Series A in Jul 2026.
Mountain View, US
Peter Elias
Ex-Optimizely principal engineer. Co-founder of Patch (App caching layer for data warehouses). Now founder & CEO of Probably, which wraps LLM output in a deterministic validator that rejects hallucinations before they reach the user. Raised a $9M Seed in Jun 2026.
San Francisco, US
Carried Interest
Articles of the week
