By 2028, 70% of enterprises are expected to abandon some agentic AI solutions created through vendor forward-deployed engineering, according to Gartner. The problem is not agentic AI itself, but an implementation model that can leave companies facing rising costs, weak knowledge transfer and long-term dependence on external suppliers.
Agentic AI differs from a conventional chatbot because it can plan and execute multi-step tasks, use tools and adapt its actions to a goal. This makes implementation more complex and often pushes companies to rely on specialised vendor engineers working directly with the customer.
A project can work technically and still fail economically
Gartner warns that a technically successful deployment is not enough if the customer cannot later operate, optimise and extend the system without the original supplier. A company may achieve a fast proof of concept but discover that every meaningful change requires expensive external support.
This creates vendor lock-in at the level of architecture, knowledge and operating processes. The long-term cost can become much higher than the initial project budget.
External engineers should not replace the internal team
Gartner argues that forward-deployed engineers should work alongside internal developers, domain experts and end users rather than functioning as a substitute for them. The customer also needs a clearly identified manager responsible for the business outcome, not only for technical delivery.
Knowledge transfer should be part of the project from the start. The organisation needs to understand how the agent is designed, what data and tools it uses, how execution is controlled and how the system can be changed or replaced later.
An exit plan should exist before deployment
Agentic systems are not finished at launch. They require continuous monitoring, optimisation and governance as models, costs and business processes change. Gartner therefore recommends that companies plan from the beginning for independent operation and, if necessary, migration away from the vendor.
The firm also warns about so-called “FDE washing”, in which conventional consulting or professional services are rebranded as forward-deployed engineering without creating reusable product capabilities. Gartner expects that by 2028 fewer than 20% of FDE projects will translate recurring customer needs into core product features.
Agentic AI is still expected to spread
The forecast does not mean that companies will abandon agentic AI as a category. Gartner has analysed more than 100 agentic deployments and expects domain-specific agents to generate measurable returns in many organisations.
The main barriers include data quality, architecture, proliferation of too many agents, token costs, model reliability and organisational change. Gartner also expects that by 2029 at least 70% of organisations using agentic AI in production infrastructure and operations may experience a material incident linked to insufficient execution controls.
The strategic question is therefore shifting. It is no longer only whether an agent can be deployed quickly, but whether the company will still be able to control, maintain and economically justify it two years later.
