AnDaltan

Dude, amazing! The images are beautiful and it's 1000 times better when you know they're real and not CGI/AI.

If the same engagement dynamics that are supposedly fine for the public product have to be dialled back internally so your own workforce will stop scrolling and do their jobs, that's a pretty revealing commentary on your model and your goals.

The detail that makes this interesting is the two-layer mechanism. There’s a 20-hour free-running oscillator that doesn’t meet the usual definition of circadian because it’s temperature-sensitive and then on top of that a separate countdown triggered by sunrise that governs the spawning event. Two imprecise systems combining into precise, synchronised behaviour.

Also worth noting that the hydrozoan lineage lost the CLOCK/BMAL1/CRY genes associated with circadian rhythms in most other animals. So whatever this timing system is, it seems to have evolved independently. Rosato’s question in the commentary is a good one: how many other unconventional clocks are out there that people have missed because they were looking only for the usual genetic components? There’s something very neat about evolution backing into a precise clock this way because the reproductive timing pressure is doing so much work.

Would love to see this kickstart research into more unconventional time-keeping processes that might be out there.

The storage/ownership side of this makes sense. The part I’m more sceptical of is the agentic layer once it starts influencing detections and response.

The post is not just talking about natural-language search. It is talking about AI helping ingest data into OCSF, author new detections, modify existing rules, and support agentic investigation/response. That is where drift, approval, auditability and rollback matter more than the demo.

The Antimatter acquisition is probably the most interesting part of the announcement. If defensive agents are going to sit close to your security telemetry and workflows then their authz/authn model is pretty centrally important.

Yeah, I think that’s broadly right.

MCP has plenty of problems, but standardising on OAuth was one of the better calls. Expiry, scopes, rotation, delegated access, all much better than the usual CLI pattern of long-lived API keys. The CLI story there is still pretty rough.

And once the policy model is host/path matching, GraphQL and JSON-RPC become awkward immediately unless the proxy starts understanding payload semantics.