The number I keep rereading isn’t the enzyme. It’s the setup: roughly 950 agents, 21 hours, 210 million tokens, and a harness that runs many Claude sessions in parallel. That’s not a lab-only trick. It’s the same shape as any agent swarm I’d point at a codebase or a log archive, just aimed at DNA.
The part worth stealing is the triage. Claude writes a short human-readable report per candidate, then critically evaluates it, and most get eliminated. The team studies which proposals they’d actually test and feeds that back into the instructions. If I ran a swarm like this, the generator would be the easy half. I’d spend my effort on the filter, and on writing down my own taste.
The story — Anthropic launched a life sciences research group with its own Bay Area lab. From an initial prompt, Claude agents gathered over 200,000 reverse transcriptases, flagged 3,500 candidate systems and narrowed them to 20. One agent spotted a CRISPR-like repeat array beside an odd RT in jumbo phage DNA. Further analysis and lab testing identified a previously uncharacterized phage system, array-associated reverse transcriptases (ART). Its function is unknown; a pre-print is out. (Source)