Agents, Trust & Licensing
Enterprise leaders are publicly debating whether autonomous AI agents can be trusted and how they should be licensed. Salesforce’s CEO raised trust concerns about powerful agent tech, while a Microsoft executive suggested agents might need software licenses like human employees—and platform vendors are already changing AI pricing. Those conversations point to procurement demanding clearer governance and seat/licensing models for agent deployments. (timesofindia.indiatimes.com) (businessinsider.com) (techtarget.com)
A fight that used to be about how smart artificial intelligence agents are is turning into a fight about whether companies can trust them enough to give them logins, budgets, and a line item in procurement. Marc Benioff said OpenClaw was powerful after he tested it, but “not enterprise-great” because it lacked trust, security, reliability, and availability. (timesofindia.indiatimes.com) That is a bigger warning than it sounds, because OpenClaw is not a chatbot that drafts email. The project was pitched as software that can book flights, manage calendars, and act across an entire computer on its own. (timesofindia.indiatimes.com) The trust problem is concrete, not philosophical. Cisco said OpenClaw could run shell commands, leak application programming interface keys, and execute scripts with minimal safeguards, and Bitsight said it found more than 30,000 exposed instances on the public internet soon after the tool went viral. (timesofindia.indiatimes.com) That is why enterprise buyers are separating “can it do the task” from “can it do the task inside a regulated company.” Benioff’s answer was to talk about local agents, customer agents, and employee agents tied into Slack and Salesforce’s own stack instead of dropping a consumer-style autonomous tool straight into the business. (timesofindia.indiatimes.com) At the same time, Microsoft executives are pushing a second argument: if agents start doing work that employees used to do, software vendors may charge for them the way they charge for human seats. Business Insider reported that a Microsoft executive described agents as a new paid “seat,” not just a free feature attached to one worker’s license. (businessinsider.com) That idea matters because most enterprise software has been sold for years on a simple count: how many humans need access. If one manager can supervise five agents, a vendor that keeps charging only for the manager loses revenue, so the pricing model starts to move from per person to per person plus per agent. (businessinsider.com) (deloitte.com) Consultants and analysts have been signaling this shift for months. Deloitte wrote in November 2025 that software-as-a-service subscriptions and seat-based licensing could give way to hybrid pricing that mixes usage-based and outcome-based charges as autonomous agents spread through enterprise software. (deloitte.com) Vendors are already adjusting in public. TechTarget reported on April 10, 2026 that ServiceNow is embedding artificial intelligence features throughout its platform as part of a pricing change aimed at making adoption easier for customers that have struggled to prove return on investment. (techtarget.com) Put those pieces together and the real debate inside large companies stops being “should we try agents.” It becomes “which agents get identities, which systems they can touch, who signs off on the risk, and whether procurement is buying a tool, a coworker, or a metered service.” (businessinsider.com) (timesofindia.indiatimes.com) (deloitte.com) That is why the loudest argument in enterprise artificial intelligence right now is not about model benchmarks. It is about governance paperwork, security controls, and billing logic, because those are the parts that decide whether an agent stays in a demo or gets deployed across a Fortune 500 company. (timesofindia.indiatimes.com) (businessinsider.com) (techtarget.com)
