This week's AI agent headlines split cleanly into two columns: eye-popping enterprise deployment numbers, and a growing chorus of warnings that oversight hasn't kept pace. Here's what's worth your attention.
This week's developments
Cognition AI raises $2B at a $48B valuation. Devin's revenue run-rate has nearly doubled to roughly $900M as the coding agent embeds deeper into engineering orgs at Nvidia, Citi, GE Aerospace, and Mercedes-Benz — a clear signal that enterprise coding-agent spend is scaling fast, not plateauing. Source
Salesforce closes its acquisition of Fin (formerly Intercom) and separately shipped six new GA "job-ready" agentic AI agents for Agentforce. The company says it has now delivered 7 billion cumulative "Agentic Work Units" (3.2B in Q2 alone), and cites Fin resolving 79% of Anthropic's own support conversations autonomously — on top of Fin's existing 76% resolution rate across 30,000+ customers. It's one of the clearest public ROI data points in the market right now. Fin acquisition | GA agents
OpenAI launched its Agents API in public beta, putting the managed Codex harness — sessions, context compaction, sub-agent coordination, nine hosted-sandbox partners — behind a single API call with no extra fee beyond compute. That materially lowers the bar for enterprises building production agents in-house rather than buying a packaged product. Source
Google's Threat Intelligence Group reports attackers ran a full agentic credential-harvesting campaign in under six hours, exposing more than 23,800 secrets — concrete evidence that adversaries have moved from single-prompt AI misuse to autonomous, multi-agent attack chains that compress the detection window defenders have to work with. Source
Anthropic CEO Dario Amodei warned in an open letter that AI agent "swarms" could be capable of assembling a persistent botnet and taking over parts of the internet within 6-12 months, potentially causing damage in the hundreds of billions of dollars. It's notable less for the specific timeline than for who's saying it: a leading agent vendor's own CEO flagging catastrophic-scale multi-agent risk in public. Source
The adoption-confidence gap, in numbers
A Harness survey of 700 tech professionals at companies with 1,000+ employees found 77% claim a full AI agent inventory, but only 44% run active discovery tooling to verify it. 74% say they trust their agent testing, yet just 19% have automated release-blocking gates in place. Source
Read across
Put these together and a pattern from earlier issues keeps repeating: the deployment numbers (7 billion Agentic Work Units, a $48B coding-agent valuation, a public agent-builder API) are compounding faster than the governance numbers (44% real discovery, 19% real release gates). The Google threat report and Amodei's warning aren't abstract risks — they're describing the exact gap the Harness survey just measured. If your organization is scaling agent usage this quarter, the discovery and gating numbers are a better budget line to fix first than the next capability upgrade.
If this was useful, forwarding it to one colleague who'd care is the best way to help it grow. Subscribe at https://buttondown.com/TheAutonomousEdge for weekly briefings on AI agents and enterprise automation.
Next issue lands next Monday.
Top comments (0)