- Recent Funding
- Founders
- Henry Ehrenberg
About Henry
Henry Ehrenberg is a co-founder of Snorkel AI, a company that emerged from the Stanford AI Lab in 2019. Snorkel AI focuses on developing data and environments for high-performing frontier and agentic AI. Ehrenberg is involved in the engineering aspects of the company and actively shares insights on AI model performance and benchmarking.
Strengths: founder archetype
one sourced signal · hover a cell for it, click for the source
Why technical builder
Henry Ehrenberg co-founded Snorkel AI out of the Stanford AI Lab, indicating a strong research background. His X posts consistently discuss technical details of AI model performance, benchmarks, and engineering, demonstrating a deep involvement in the technical development and evaluation of AI systems.
The signals behind it 2
Henry Ehrenberg | Stanford AI Lab to Snorkel AI
Every tag links its source.
Network
Background
1
- Role Co-founder + EngineeringX
Timeline, line by line 7
S
BIn their words
8
- “Claude Fable 5.1 now tops the Senior SWE-Bench leaderboard, edging out Fable 5 on the pass^3 tie-breaker (more consistently shipping quality code).”Discussing AI model performance on a benchmark. · Sep 1, 2026X
- “At higher effort tiers, we noticed that Fable 5.1 tends to over-engineer on Senior SWE-Bench tasks. For example on turborepo-feat-add-circular-package, Medium correctly uses petgraph's Tarjan SCC method whereas High hand-rolls its own 366-line implementation.”Analyzing AI model behavior and efficiency on coding tasks. · Sep 1, 2026X
- “We expect agents to act like senior engineers, but most benchmarks still evaluate them like interns.”Introducing a new benchmark for AI agents. · Jul 1, 2026X
- “We expect agents to build real features going on just a quick Slack message, nothing like the super technical instructions most benchmarks provide. Senior SWE-Bench fixes that.”Explaining the purpose and design of Senior SWE-Bench. · Jul 1, 2026X
- “We expect agents to build real features going on just a quick Slack message, nothing like the super technical instructions most benchmarks provide. Senior SWE-Bench fixes that.”Explaining the purpose and design of Senior SWE-Bench. · Jul 1, 2026X
- “One last thought: we hate slop tasks. They burn compute and don’t actually measure anything. Every Senior SWE-Bench task has gone through multi-stage design, refinement, and review.”Emphasizing the quality and rigor of Senior SWE-Bench tasks. · Jul 1, 2026X
- “There is asymmetric power in defining how we measure progress. You can move the field of AI ($Ts) with a small team of smart people and a concentrated investment ($Ks). ImageNet, MMLU, and SWE-Bench all did it. Let's close the evaluation gap.”Discussing the impact of benchmarks on AI progress. · Feb 11, 2026X
- “Our customers who are training with the scaled dataset are seeing larger perf lifts and transfer than I was expecting.”Discussing the effectiveness of a dataset for training AI models. · Jan 9, 2026X
Interviews and coverage
2
- Snorkel AI — Unlocking AI for EnterprisesLightspeed Venture Partners · Apr 7, 2021
L - Stanford Spin-Out Snorkel AI Solves a Major Data ProblemStanford HAI · Aug 18, 2020
H
How this is researched
Researched from public pages: press, interviews, the company's site and the founder's own posts. Each source logo links the page that states the row; anything we couldn't source is left out. The archetype is our reading of that record, from a fixed set so founders can be compared. Updated Oct 12, 2026.
The week's top funded startups and market insights, in your inbox every Monday.
Unsubscribe anytime · privacy