OpenAI Announces It Would Like Your CRM Money

OpenAI shipped AgentKit — a toolkit for building agentic workflows — and the framing is very much "enterprise, automated, your-business-processes-but-AI," which is to say it is aimed directly at the throat of every company currently charging nine figures to put a chatbot in front of your CRM data.

The toolkit will look familiar to anyone who has spent time in this space. It should. The ideas have been circulating for two years — handoffs, tool use, memory, orchestration, agents that can spawn other agents. The difference is that when OpenAI ships it, it stops being a research artifact and becomes a product line.

The companies that should be updating their résumés: Salesforce, HubSpot, Docusign, ZoomInfo. The entire category of software whose core value proposition is "we store your business data and help you act on it." OpenAI just announced it would like to help you act on it. Directly. Without the middleman.

Salesforce has had a peculiar relationship with this moment. They spent 2024 and 2025 building Agentforce — their big agentic play, the thing Marc Benioff has been doing keynotes about — on top of, among other things, OpenAI's models. They were building the house while renting the land from someone who just announced they're going into the homebuilding business.

The Salesforce community is already doing the math. Salesforce Ben — a site that is essentially the trade press for Salesforce administrators, people whose entire professional identity is wrapped around knowing where every checkbox is in the platform — published a piece asking whether Salesforce should be worried. The answer implied by the question is yes. The answer implied by the fact that they had to ask is also yes, but sadder.

What's actually interesting here isn't the competitive threat in isolation. It's the shape of the move. OpenAI has spent two years being the picks-and-shovels play — sell the intelligence, let everyone else build the product. AgentKit is the moment they decided the picks-and-shovels metaphor was leaving too much money on the table. Why sell the shovel when you can dig the mine.

Every SaaS company that built on top of foundation models is now in the same position: they either differentiated on something other than "we have AI" — actual proprietary data, actual workflow depth, actual switching costs — or they are, as of this week, a slightly more expensive version of the thing that just ate them.

Some of them differentiated. Most of them did not.