# Anthropic says ~950 Claude agents spent 21 hours on an enzyme lead. What counts as discovery? — retained record

> Retained reddit record from reddit:AI_Agents.

Canonical: https://highsignal.app/data/records/ba2ecdb7550a9aa2

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:AI_Agents)  published 2026-09-24 17:37 ingested 2026-09-25 02:36

[original source ↗](https://www.reddit.com/r/AI_Agents/comments/1wp86dy/anthropic_says_950_claude_agents_spent_21_hours/)

## retained text



```
Anthropic reported yesterday that a search using roughly 950 Claude agents over 21 hours identified an unusual DNA-repeat array and an accessory protein beside a previously known reverse transcriptase. Human scientists then tested the lead in their lab. Anthropic says the biological function is still unknown. This is an early vendor report and preprint, not an independently replicated result.

The agent-system question seems more useful than the headline: what denominator makes this workflow convincing? I'd want to know how many candidates were discarded, how many survived human review, total compute plus scientist time, and whether another group can reproduce the finding from the same data and procedure. A plausible lead is valuable, but it is a different outcome from a characterized biological mechanism.

The same evaluation problem shows up in coding or research agents: a large pile of interesting suggestions can hide the cost of validating each one. Which metric would you use to compare an agent swarm with a smaller human-led search?

Source and technical report in a comment, per this subreddit's link rule.

*AI-assisted discussion; no independent experiment is claimed.*
```



document fetched 2026-09-25 00:19 · canonical https://reddit.com/r/AI_Agents/comments/1wp86dy/anthropic_says_950_claude_agents_spent_21_hours
