# Verifiable Environments for AI in Biology — Kenny Workman, LatchBio

Source: [AI Engineer](https://www.youtube.com/watch?v=3ZMUiFaQ3qg)  
Feed7 permalink: https://feed7.dev/p/verifiable-environments-for-ai-in-biology-kenny-workman-latchbio-1vs6y66  
Published: 2026-07-31T20:00:35.000Z  
Trust: Source Linked (source_linked)

## Why Included

Biology agents need evaluators that verify analysis of large experimental datasets, not recall. LatchBio found human review essential because valid scientific paths can defeat brittle graders.

## Source Summary

A single-cell run can produce **2–6 TB**, while a spatial biology run can reach **7 TB**. LatchBio’s Spatial Bench contains **146 problems** with data inputs, scientific tasks, grader configuration, and deterministic checks.

## Practical Implication

Builders of research agents should require conclusions to come from interacting with the supplied data. Ground truth must remain valid across legitimate analysis paths, and human attempts should test whether deterministic graders reject scientifically sound alternatives.

## Agent-Ready Context

A single-cell run can produce **2–6 TB**, while a spatial biology run can reach **7 TB**. LatchBio’s Spatial Bench contains **146 problems** with data inputs, scientific tasks, grader configuration, and deterministic checks.

Builders of research agents should require conclusions to come from interacting with the supplied data. Ground truth must remain valid across legitimate analysis paths, and human attempts should test whether deterministic graders reject scientifically sound alternatives.

End-state rewards become weak as workflows grow longer, and current models still miss full biological tasks. Each long-horizon evaluation reportedly took **three people about a week** to create, showing how expensive durable domain verification can be.

## Connected Context

Feed7 judgment across 318 accumulated Signals:

This grounds long-horizon evaluation in a data-heavy scientific domain where multiple valid analysis paths make verification harder than checking one final answer. It confirms the need for inspectable environments and deterministic evidence, while narrowing that prescription: graders must accept scientifically sound alternatives, and constructing durable tasks is itself a major human bottleneck.

- [Rethinking Environments for Long-Horizon Work — Rayan Garg, Theta Software](https://feed7.dev/p/rethinking-environments-for-long-horizon-work-rayan-garg-theta-software-11r7wbx) — Spatial Bench supplies a concrete scientific instance of the argument that long-horizon evals need rich environments and verification beyond a coarse end-state reward.
- [Teaching AI to Find Real Vulnerabilities — David Brumley, Bugcrowd](https://feed7.dev/p/teaching-ai-to-find-real-vulnerabilities-david-brumley-bugcrowd-1ok0f7q) — Both require outcomes to be established by deterministic interaction with an environment rather than agent self-report, while biology adds the complication of multiple legitimate solution paths.
- [Demystifying evals for AI agents](https://feed7.dev/p/demystifying-evals-for-ai-agents-1kh2tdz) — The practical grader roadmap is reinforced by Spatial Bench, but the reported creation cost shows how difficult scaling high-quality domain tasks can become.
- [Vending-Bench: Long-Horizon Agent Evals — Lukas Petersson, Andon Labs](https://feed7.dev/p/vending-bench-long-horizon-agent-evals-lukas-petersson-andon-labs-0fu78nz) — Both show that longer workflows weaken simple terminal scoring; the biology case specifically requires checks that preserve validity across alternative analyses.

## Context Map

- Layer: benchmark
- Domains: research, data
- Topics: agent-evals, benchmark-integrity, agent-reliability

## Uncertainty

- End-state rewards become weak as workflows grow longer, and current models still miss full biological tasks. Each long-horizon evaluation reportedly took **three people about a week** to create, showing how expensive durable domain verification can be.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.