Blockworks shipped a feature this week that's less about AI and more about labeling — which, if you're building anything that touches money, is the part worth studying.
Agentic Detection, live now inside Messari Monitoring (the asset-monitoring product Blockworks acquired for $10M+ in June 2024, down from Messari's $300M 2022 valuation), lets exchange listings desks, custodians, and compliance teams get a compliance-relevant development the moment an AI agent flags it — before a human analyst has confirmed anything. Every alert carries a tag saying whether an Agent or an Analyst produced it. That tag is the whole story.
What actually shipped
Messari Monitoring already tracked events across X, GitHub, governance forums, Discord, Telegram, project blogs, and 1,000+ news sources spanning 1,000+ crypto assets, classifying developments into roughly 10 categories and 55 subcategories that map to Messari's published taxonomy. Alerts route out over Slack, email, mobile push, webhooks, API, PagerDuty, Jira, and Telegram — the standard enterprise alerting stack, not a chat window.
What changed is speed and who gets to answer first. Analyst verification used to take about five minutes per development. Blockworks' proprietary model now produces an alert in seconds. For a listings desk deciding whether to halt deposits on a token after a governance exploit, or a custodian watching for a chain fork, five minutes is a long time to be behind.
The trust design, not the AI
Here's the part that should be boring and isn't: Blockworks didn't quietly swap analyst alerts for agent alerts and call it an upgrade. It kept both, tagged separately, with analyst-verified as the default view — customers opt into the Agentic Detection layer view-by-view. As co-founder Jason Yanowitz put it: "Our customers told us they now need alerts in near real-time. Agentic Detection allows them to hear from our agents first and get the analyst's verification second."
First and second, not first and only. That ordering is the actual product decision, and it's the same distinction we push on every listing here: an agent's own claim is not the same evidentiary weight as a checked one. Blockworks isn't calling the agent's output verified — it's calling it fast, and shipping the label that lets a customer tell the difference in a Slack channel at 2am.
Why this matters beyond one product
Compliance and listings decisions are exactly the kind of high-stakes, time-sensitive call where "an agent said so" and "we confirmed it" need to stay visibly separate — the same reason a Sato Score component distinguishes self-reported claims from reproduced or reviewed ones instead of blending them into one number. Agentic Detection is a live, shipping instance of that pattern applied to market-surveillance infrastructure that tier-one exchanges actually route decisions through, not a demo.
It's also a signal about where AI-agent output is landing first in crypto infrastructure: not trading execution, not custody itself, but the surveillance and alerting layer sitting in front of both — the part where being fast and being wrong have asymmetric costs, and where a wrong call gets caught by a human downstream instead of executed automatically.
What to watch
Whether the Agent/Analyst tag holds up under pressure — the moment a customer quietly defaults to acting on agent alerts without waiting for verification is the moment the label stops doing its job. Also worth watching: whether other monitoring and surveillance vendors adopt the same two-tier tagging convention, or whether "agentic" becomes another word vendors slap on unlabeled model output. The taxonomy is the feature; watch whether it survives contact with a busy compliance desk.
Sources
- ▸[Blockworks Lets Exchanges Take Compliance Alerts Straight From AI Agents](https://thedefiant.io/news/infrastructure/blockworks-agentic-detection-ai-agents-compliance-alerts-exchanges) — The Defiant