Should a smaller business build AI agents in-house instead of hiring outside help?

Not necessarily. Most small and mid-sized businesses do not have an R&D team, so outside builders are often the fastest route to a working agent. The point of Gartner's warning is not to avoid outside help, but to structure it so the knowledge stays with you. A good engagement leaves you with a running agent, written documentation, a named internal owner and the freedom to change providers without starting over.

Who owns your AI agent? Gartner's warning

Q: What is forward-deployed engineering (FDE)?

Forward-deployed engineering is a delivery model in which a vendor's engineers work directly inside a customer's business to build and deploy an AI solution, often an agent tied to the customer's own systems and data. It is fast because the people who know the technology sit next to the people who know the process. The risk, according to Gartner, is that the customer never builds the ability to run, change or fix the system once the vendor's engineers leave.

Q: How can a small business avoid being locked in to whoever built its AI agent?

Decide ownership before the work starts. Put in writing that the agent runs in your environment under your accounts and keys, that you own the prompts, configuration and integration code, and that the builder delivers documentation and a handover session. Keep the first project narrow, so a person on your team can understand what the agent does, and agree on an exit plan from day one rather than at renewal time.

AI agent ownership is a company's ability to run, change and fix an AI agent on its own, after the people who built it have left. It sounds like a contract detail, but it is quickly becoming the difference between an agent that keeps paying off and one that gets switched off. On September 29, 2026, Gartner predicted that by 2028, 70 percent of enterprises will abandon agentic AI built by vendor engineers embedded in their business, trapped by soaring costs and unable to evolve it on their own. That warning was written for large enterprises. However, it lands even harder on a small or traditional business without an R&D department, because almost every AI agent such a company runs will be built by someone from the outside. The question is not whether to get help. It is whether you will own what that help leaves behind.

Key takeaways. In a September 29, 2026 press release, Gartner predicted that 70 percent of enterprises will abandon agentic AI built through vendor forward-deployed engineering by 2028, and that fewer than 20 percent of those engagements will turn custom requirements into standard product features. The failure is structural, not technical: no ownership, no knowledge transfer, no exit plan. Meanwhile, small businesses are adopting AI fast and leaning on outside builders to do it. The practical lesson: whoever builds your agent, settle in writing that it runs in your environment, under your keys, with documentation and a named internal owner, before the first line of work starts.

What Gartner actually predicted

Forward-deployed engineering, or FDE, is the model in which a vendor sends its own engineers to sit inside a customer's business and build an AI solution on the customer's systems. It has become the default way large vendors deliver agents, because it is fast. Yet Gartner's prediction is blunt: by 2028, 70 percent of enterprises will abandon what those engineers built. In addition, Gartner expects fewer than 20 percent of FDE engagements to turn a client's custom needs into standard vendor features through 2028. In other words, most of what gets built stays a one-off that only the builder fully understands. "FDE success starts with getting the engagement structure right, from scope and incentives to governance, ownership, and exit," said Mukul Saha, Sr Director Analyst at Gartner. Notably, none of those words is about the model or the code.

Why agents get abandoned after the builders leave

The pattern is familiar to anyone who has inherited a spreadsheet macro nobody can explain. An outside team builds something that works on the day of the demo. Then the team leaves, a supplier changes its invoice format, the ERP gets an update, and the agent quietly starts making mistakes. Nobody inside the business knows where the instructions live, which account the agent logs in with, or who is allowed to change it. Consequently, every small fix becomes a paid ticket back to the original builder, the cost climbs, and eventually someone decides it is cheaper to turn it off. That is exactly the trap Gartner describes: soaring costs plus no internal ability to evolve the system. It is also one of the recurring reasons we covered in the AI project graveyard. Projects rarely die because the AI stopped working. They die because nobody owned them.

Why this matters more for a smaller business

A large enterprise at least has an IT department that can, in theory, take over a handed-off system. A 40-person manufacturer or a regional accounting firm usually does not. At the same time, these businesses are moving quickly. According to CIO.com's analysis of the small-business AI market, Bluevine's 2026 report found 74 percent of small business owners using or actively testing AI, and the article argues the real bottleneck is now implementation support rather than software licenses. Therefore, more small firms will hire outside help to build agents in the next year, not fewer. That is healthy, and it is often the right call, as we explained in build vs buy just flipped. The risk is signing for a result while leaving the ownership question for later, because later usually arrives as an invoice.

What owning your agent looks like in practice

Ownership is not about writing code yourself. Instead, it comes down to four plain conditions you can check without a technical background. First, the agent runs in your environment, under accounts and API keys that belong to your business, not to the builder. Second, you hold the working parts: the instructions the agent follows, its configuration and the integrations to your systems, stored where you can see them. Third, there is a short written document that explains what the agent does, what it can touch and how to pause it, plus a named person on your team who has been walked through it. Fourth, there is an agreed exit: if you change providers, the agent keeps running and the next builder can pick it up. Our guide to running AI inside your own environment covers the first condition in detail, and it also happens to be the one that makes the other three possible.

What to do before you sign

Name an internal owner first. Pick one person accountable for the business outcome, not the budget. Gartner makes the same recommendation for enterprises: an executive sponsor measured on results.
Write ownership into the scope. Your environment, your accounts and keys, your copy of the instructions and integration code, and a handover session with documentation as a named deliverable.
Start with one narrow process. A single, well-bounded agent is something a non-engineer can understand and supervise. Our implementation guide shows how to pick it.
Agree the exit on day one. Decide what happens if you stop working with the builder, so leaving is a choice and not a rebuild. Gartner frames success as the customer's ability to run the system independently.
Measure it yourself. Track hours saved and errors caught against a baseline you own, using benchmarks like our payback numbers, so renewal is a decision based on your data.

Frequently asked questions

What is forward-deployed engineering (FDE)?

It is a delivery model in which a vendor's engineers work inside a customer's business to build and deploy an AI solution on the customer's systems. It is fast, but the risk is that the customer never builds the ability to run or change the system once the engineers leave.

Why does Gartner expect so many vendor-built AI agents to be abandoned?

Gartner predicts 70 percent of enterprises will abandon agentic AI built through vendor FDE by 2028, trapped by soaring costs and unable to evolve it on their own. Fewer than 20 percent of engagements are expected to turn custom requirements into standard product features.

How can a small business avoid being locked in to whoever built its AI agent?

Decide ownership before the work starts: your environment and keys, your copy of the instructions and integration code, documentation and a handover, a named internal owner, and an exit plan agreed on day one.

Should a smaller business build AI agents in-house instead?

Not necessarily. Outside builders are often the fastest route for a company without an R&D team. The goal is to structure the engagement so the knowledge stays with you, and you can change providers without starting over. For the platform side of that choice, see our take on managed AI agent platforms.

Who owns your AI agent? What Gartner's vendor-built warning means for your business