Enterprise agents are the focus

The fight in artificial intelligence is moving away from who has the prettiest demo and toward who can close a purchase order inside a Fortune 500 company. Gartner said on April 2 that more than half of enterprises will stop paying for “assistive” tools by 2028 and will favor platforms that promise workflow results instead. (gartner.com) That change sounds small until you look at what buyers are asking for. A chatbot answers a question in a window, but an agent is supposed to open systems, move data, request approvals, and finish a task without a person clicking every step. (gartner.com)) Gartner drew the line clearly in August 2025. It said most enterprise software would have embedded assistants by the end of 2025, but task-specific agents were the next stage, with 40% of enterprise applications expected to include them by the end of 2026, up from less than 5% in 2025. (gartner.com)) Once software starts taking actions, the real problem is no longer raw model quality. The problem is whether the system knows which employee it represents, which database it may touch, which spending limit it must obey, and which log will show what happened after the fact. (gartner.com)) That is why governance is suddenly showing up next to every agent pitch. Microsoft said on April 2 that agents are already being used to book flights, execute trades, write code, and manage infrastructure, and it released an open-source toolkit built around policy enforcement, identity, and action controls for autonomous agents. (opensource.microsoft.com)) The security language is getting more concrete because the risks are getting more concrete. Microsoft pointed to the Open Worldwide Application Security Project’s December 2025 list of 10 agent-specific risks, including goal hijacking, tool misuse, identity abuse, memory poisoning, cascading failures, and rogue agents. (opensource.microsoft.com)) The compliance clock is also getting closer. Microsoft said the European Union Artificial Intelligence Act’s high-risk obligations take effect in August 2026, and the Colorado Artificial Intelligence Act becomes enforceable in June 2026, which means companies deploying agents now have to think like auditors, not just app builders. (opensource.microsoft.com)) The same pattern shows up in deployment data. Databricks said in January that companies using artificial intelligence governance tools moved more than 12 times as many projects into production, and companies using evaluation tools moved nearly 6 times as many systems into production. (databricks.com)) The first big uses are not science-fiction robots running the whole company. Databricks said the top enterprise uses are critical but routine jobs like market intelligence, customer advocacy, and regulatory reporting, which are boring enough to automate and important enough to measure. (databricks.com)) Deloitte’s 2026 enterprise survey shows why this has become the boardroom version of the artificial intelligence debate. Worker access to artificial intelligence rose by 50% in 2025, but only one in five companies has a mature governance model for autonomous agents, so adoption is running ahead of controls. (deloitte.com)) That is the new test for vendors. The winner is less likely to be the company with the flashiest benchmark chart and more likely to be the one that can show a chief information officer exactly which agent touched which system, under whose permission, for how much cost, and with what measurable business result. (gartner.com)