Autonomy needs a checkpoint. Anthropic reported that Claude agents, given high-level direction, spent ~21 hours mining DNA data and flagged a previously uncharacterised enzyme system with CRISPR-like repeats. Anthropic has not determined the system’s function; lab experiments remain human-run.
What happened
Anthropic reported that Claude agents, given high-level direction, spent ~21 hours mining DNA data and flagged a previously uncharacterised enzyme system with CRISPR-like repeats. Anthropic has not determined the system’s function; lab experiments remain human-run.
Why it matters for UK SMB / service operators
Useful pattern for operators: AI proposes and narrows candidates; humans decide what to test and what goes live. Same discipline applies to quotes, client emails, and workflow changes — draft fast, approve carefully.
Operator move
Write down where AI may draft and where a named human must decide before anything client-facing goes out. Treat “propose / decide” as the default workflow.
Read the full daily brief hub: www.aifusionautomations.com/news/