Why vendor-built AI agents fail UK SMEs

Gartner expects 70% of enterprises that have agentic AI built by a vendor's own engineers to abandon it by 2028. That is not a fringe prediction. That is a mainstream industry forecast, and it carries a warning that UK Managing Directors, Operations Managers, and IT Managers at smaller businesses cannot afford to ignore.

The dynamic at the heart of this forecast is simple: external developers build something complex, bill for it, and leave. When they go, the capability goes with them.

The forecast UK businesses need to understand

Analyst firm Gartner forecasts that 70% of enterprises will abandon agentic AI built through vendor forward-deployed engineering by 2028. The driver is a practice known as "forward-deployed engineering", where a vendor's engineers embed in a business, build bespoke AI agents tailored to that organisation, and then exit the engagement.

The problem is not that the agents are poorly built. Many are sophisticated. The problem is that the knowledge required to run, adjust, and maintain them lives entirely with the people who built them, not with the business that paid for them.

Once the consultants leave, the business faces a choice: pay ongoing fees to keep the original team engaged, or watch the system degrade until it is cheaper to abandon it than maintain it. Gartner's forecast suggests many will choose the bin.

Why the "build, bill, bin" cycle is an SME problem too

Enterprise case studies can feel remote from the reality of a 60-person fire protection firm in the Midlands or a professional services business in Manchester. They should not.

The same dynamic plays out at smaller scale. An Operations Manager hires an AI agency to automate a key workflow. The agency delivers a working system, the contract ends, and six months later nobody in the business can explain how it works, let alone update it when a process changes.

For an enterprise with a large IT department, absorbing that cost is painful but survivable. For an SME running on tight margins, it can wipe out every efficiency gain the project was supposed to deliver.

The forecast should be read as a structural warning, not a story about big company mistakes.

What is actually going wrong beneath the surface

AI adoption is 80% people and culture, and 20% technology. That ratio explains everything about this forecast.

Vendor-built agent projects tend to invert that ratio. They optimise for technical sophistication. They produce impressive demonstrations. They generate genuine short-term efficiency gains. But they rarely build the internal understanding that a team needs to own what has been created.

When the developers leave, three things are typically missing. The first is documentation: clear, plain-language records of what each agent does, how it is triggered, and what happens when something goes wrong. The second is training: structured time with the internal team to understand the logic of the system, not just how to use it on a good day. The third is a realistic handover plan: a roadmap that assumes the consultants are not coming back.

Without those three things, even a well-built agent becomes a liability. An IT Manager who cannot diagnose a failure is dependent on external support. An Operations Manager who cannot adapt a workflow when a supplier changes their process is stuck. A Managing Director who signed off on the project is left explaining why the efficiency gains have evaporated.

This is not about technical complexity being inherently bad. It is about who owns the knowledge.

Three things to insist on before signing any AI agent contract

1. Demand a training plan, not a handover document. A handover document tells your team what the system does. A training plan builds the capability to understand why it does it and how to change it. Ask to see the training plan before the contract is signed, not after the build is complete.

2. Ask for plain-language documentation of every agent's logic. This means a description of each trigger, each decision point, and each output that a non-developer can read and act on. If the vendor cannot produce this, the system will be unmaintainable the moment they leave.

3. Insist on a realistic internal ownership roadmap. This is a written commitment to what your team will be able to do independently by the end of the engagement. It should name specific capabilities, not vague goals. "Your IT Manager will be able to edit workflow triggers without external support" is a roadmap. "We will leave you set up for success" is not.

None of these asks is unreasonable. Any vendor unwilling to commit to them is signalling that their model depends on your ongoing reliance.

The honest caveat about AI agents

AI agents are genuinely powerful. End-to-end automated workflows, with quality gates and structured handoffs, can deliver meaningful time savings and cost reductions for SMEs that get them right.

But they are also the most complex category of AI deployment available to smaller businesses right now. That complexity is not a reason to avoid them. It is a reason to be precise about who owns the knowledge when the project ends.

Off-the-shelf AI software, configured by your own team with proper training, will outperform a bespoke agent that nobody in your business understands. A system your Operations Manager can maintain beats a sophisticated one they cannot touch.

The forecast should recalibrate expectations, not eliminate ambition. The question to ask is not "can we build something impressive?" It is "will we still own this in two years?"

Build internal capability before you build agents

The real lesson from the 70% abandonment rate is that SME teams were never trained to own their AI systems. When external consultants disappear, so does all capability. Building internal AI literacy first, through proper structured training, is the only defence against vendor lock-in.

gecco's Training and consultancy programme is built around exactly this principle. The GRAFT methodology and SCALE adoption model mean the logic behind every agent is documented and transferable. The goal is a capable internal team that owns its AI, not an ongoing reliance on outside expertise to keep the lights on.

If your Managing Director, Operations Manager, or IT Manager is weighing up an AI agent project and wondering what to demand from any vendor, the AI Readiness survey is a practical starting point. Taking the survey gives you access to 65+ free resources and a custom AI Readiness report, followed by a free 30-minute showcase to walk through the results with a member of the gecco team.

Website · LinkedIn · Case Studies · Newsletter