Gartner predicts 70% of enterprises will abandon vendor-built agentic AI by 2028
Gartner predicts 70% of enterprises will abandon agentic AI built by vendors' forward-deployed engineers by 2028, as cited in a video.
A video posted on the Agentic AI Foundation site on November 10, 2026 cites a Gartner prediction that 70% of enterprises will abandon agentic AI built by vendors' forward-deployed engineers by 2028. Erica Hughberg, a community advocate at Tetrate, uses the figure to argue that companies should build and own their own agents. The video is a conversation with Demetrios Brinkmann. The Gartner figure comes only from Hughberg's account. The video does not name the Gartner report or say how the forecast was made, so the number stands as her citation rather than as something the video checked.
Hughberg's main claim is that the obstacle is not model quality. In her words, "It's not the models that's holding us back. It's the systems around the models." She says the push toward agents comes from boardrooms, where leaders ask what their AI strategy is, rather than from teams trying to solve a specific problem. She also gave a practical example. She had to reintroduce the same design system to a Claude project repeatedly, which she took as a sign that the surrounding context, not the model, was the weak point.
Hughberg is open about the limits of her evidence. She said she has a feeling that few companies run agents at scale in production, and then added that she does not have proper data to back it up. Her definition of production is narrow: someone has built their own agents rather than using another company's. Under that definition, she said, Netlify's built-in agents for its own users and Bloomberg's agentic features for its clients do not count as the company running its own agents. Brinkmann pushed back. He said many companies already run agents in production through Claude Code and Codex, which are vendor-built tools. He added that companies rarely publicize building their own because it is hard, and admitting difficulty earns little credit.
Hughberg also set out a distinction she is still testing. Her gut feeling is that a tool should be deterministic, giving the same output for the same input, while an agent should improvise. She noted that some tools may themselves be agentic and therefore not deterministic, so the distinction is not clean. She sees agent-to-agent protocols, known as A2A, mattering most when her agent can talk to an agent on a remote system she did not build. She also suggested that delegation is a useful way to judge how autonomous an agent is, comparing tasks, problems, outcomes and responsibilities. She said she has not yet been able to delegate responsibility to an agent.
The video's case rests mostly on Hughberg's own experience and her reading of how companies talk about their AI work. The Gartner figure is the video's only cited forecast, and the video gives no further detail on it.
Why it matters
The forecast, if accurate, suggests buyers of agentic AI should plan to own and maintain agent systems rather than depend on vendor engineers.
What you can do
Assess whether your team can build, run and maintain agents in production without vendor engineers.
Original source
We have no link to the original source for this story yet.
Further reporting
Where we saw it
Demetrios from the AAIF
AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works