Cross-posted from The Autonomous Edge, a free weekly briefing on AI agents in the enterprise. Get it by email every Monday.
OpenAI cut the price of agentic computer-use in half this week, while IBM made the case that governance needs to catch up with deployment — and a blockchain-analytics firm found out the hard way why that matters. Here's what moved in agentic AI and enterprise automation over the past seven days.
This week's developments
OpenAI shipped GPT-6.1 Terra, a mid-tier model built for computer-use and workflow automation at half the price of its predecessor. The model targets tasks like drafting proposals, interpreting technical drawings, and filling out forms, and OpenAI says its internal error rate on a stress-test benchmark fell from 53.6% to 16.7% at low reasoning effort, and to 4.3% at maximum effort. It rolled out October 26 in ChatGPT Work, Codex, and the API, priced 50% below the GPT-5.6 Terra model it replaces. Source
IBM introduced identity and access controls for AI agents at TechXchange, citing a Gartner estimate that the average Fortune 500 company will run over 150,000 agents by 2028 with only 13% confident they can govern them. The lineup spans HashiCorp Vault 2.2 with Agent Identity (now generally available), a new Secure Hub in preview, and integrations with watsonx Orchestrate, Amazon Bedrock, Microsoft Foundry, and Google Gemini; named Vault customers include ABN AMRO and Target. The pitch: persistent credentials and unclear agent ownership are becoming the governance gap enterprises need to close before scaling further. Source
Chainalysis disclosed a breach it attributed to "offensive agentic techniques," exposing customer names, emails, user IDs, and organizational role labels. The blockchain-analytics firm said in an October 30 post that it found no evidence investigation content, credentials, or session tokens were accessed, but hasn't yet determined exactly which records were touched for each affected organization. The company gave no technical detail on how AI featured in the intrusion. Source
This week's adoption data
UBS said it now has more than 12,700 personal AI agents in use across its workforce, with 608 use cases live in production — up 81% from a year ago — and over 1,090 more in development, per its Q3 2026 results. Source
Read across
The pattern this week: capability and governance are moving in opposite directions at the same time. OpenAI just made agentic computer-use cheaper, and UBS's numbers show what happens when that cost curve bends down — use cases up 81% year over year inside one bank. IBM's TechXchange push, backed by Gartner's 150,000-agents-by-2028 estimate, is a bet that governance tooling needs to scale at least as fast as deployment does. And Chainalysis's breach is the cautionary note underneath all of it: "agentic techniques" are now showing up on the attacker's side of the ledger too, not just the defender's. For more of our back catalog on this shift, see the archive.
Subscribe to The Autonomous Edge at https://buttondown.com/TheAutonomousEdge to get this every Monday.
Next issue lands next Monday.
Top comments (0)