LEETConsulting
← Insights

2 min readLeet Consulting

Don't rent your agents

Gartner predicts 70% of enterprises will abandon agentic AI built by vendor forward-deployed engineers by 2028. With costs falling 13x a year, owning the capability is the whole game.

  • Strategy
  • Production

Gartner published a prediction last week that every buyer of AI services should read. By 2028, it expects 70% of enterprises to abandon agentic AI built through vendor forward-deployed engineering, because costs rise sharply and they cannot evolve the solutions themselves. It also warns that through 2028 fewer than 20% of those engagements will turn recurring customer needs into core product capabilities.

Gartner's recommendations are blunt: an accountable executive sponsor, contract terms for knowledge transfer, IP and exit, forward-deployed engineers paired with internal experts, and actually executing the exit plan to prove independence.

Why lock-in hurts more with agents

The economics of AI keep shifting under every deployment. Epoch AI's new analysis, The plunging price of thought, estimates the cost of a given level of AI performance has fallen about 47% per quarter since 2023, roughly 13x per year. One example: o3 reached about 75% on GPQA Diamond for around 30 cents per question; just under 18 months later, a smaller model matched it for $0.0004, a 725-fold drop.

Model releases follow the same pace. Anthropic's release notes describe Claude Opus 5.5, released September 22, as close to the top model on most work at 40% lower run cost than Opus 5, two months after Opus 5 itself.

If you cannot change your own agent, you cannot capture any of that. You keep paying last year's price for last year's model.

How we structure engagements

We have always ended engagements with a handover, since our forecasting days. For agents, we make it contractual and technical:

  • You own the code, prompts, evaluation sets and connectors, in your repositories, from day one.
  • Your engineers pair with ours on the build, not after it.
  • Models are swappable behind an evaluation gate, so your team can adopt the next cheaper model without us.
  • Runbooks and training are deliverables with acceptance criteria.
  • An exit date is agreed up front. If you want us to keep operating the system, that is a separate, optional decision.

The best measure of an AI partner is how well you can do without them after the project. That is the one we want to be measured on.

Sources

  1. Gartner Predicts 70% of Enterprises Will Abandon Agentic AI Built by Vendor Forward-Deployed Engineering by 2028, Gartner, September 29, 2026.
  2. The plunging price of thought, Epoch AI, September 22, 2026.
  3. Claude release notes, Anthropic.