Enterprise AI: Slow, Steady Rollout

OpenAI is widening access to tools that let ChatGPT call outside software, but its biggest workplace rollout is still ahead. (help.openai.com) OpenAI’s help documentation says GPTs can be configured with “actions,” which connect a custom GPT to an external application programming interface, or software endpoint, using authentication and an OpenAPI schema. The same documentation says building and editing GPTs is limited to the web experience. (help.openai.com 1) (help.openai.com 2) The release notes and plan documentation show OpenAI shipping these controls first through paid web products such as Plus, Pro, and Team-era workspaces, while Enterprise and Education controls are governed through separate workspace settings and admin policies. OpenAI renamed ChatGPT Team to ChatGPT Business on August 29, 2025, and its current Business documentation still describes the product as a self-serve plan for organizations with two or more users. (help.openai.com 1) (help.openai.com 2) OpenAI’s Enterprise and Education guides put the emphasis on governance: workspace owners can manage GPT access, sharing, action-domain allowlists, and whether directory-synced groups appear in sharing menus. OpenAI’s SharePoint guidance also splits deployments into centrally managed Enterprise and Education rollouts or user-led pilots, with the admin-managed option reserved for those larger plans. (help.openai.com 1) (help.openai.com 2) (help.openai.com 3) That makes the current phase look less like a mass launch than a staged handoff from individual users to managed workplaces. OpenAI’s Codex pricing pages now separate Plus and Pro from Business, Enterprise, and Education billing, and say Enterprise and Education usage runs through workspace credits and seat controls rather than a flat consumer-style subscription. (help.openai.com) (help.openai.com) Anthropic is moving through a different channel: large consulting and information-technology services firms. Horses for Sources reported on April 13 that Accenture has 30,000 Claude-trained practitioners, Deloitte rolled out Claude to 470,000 employees, and Cognizant deployed it to 350,000 more. (horsesforsources.com) The same report said Infosys signed a new Anthropic deal covering regulated industries, where procurement, security review, and workflow integration usually slow adoption. That points to enterprise artificial intelligence spreading through existing service partners and internal process redesign, not only through direct software subscriptions. (horsesforsources.com) OpenAI and Anthropic are both chasing the same budget line: workplace software that can reach internal documents, approved apps, and company-specific tasks without breaking security rules. The difference is that OpenAI’s latest public signals are coming from product notes and admin controls, while Anthropic’s are showing up in deployment counts tied to consultancies. (help.openai.com) (horsesforsources.com) The next marker is not another demo. It is whether Enterprise and Education rollouts arrive with the same pattern already visible in the docs: tighter admin settings, narrower permissions, and more of the work happening inside existing corporate systems. (help.openai.com) (help.openai.com)