The news
Enterprise AI ROI faces a new test. On September 29, 2026, Gartner predicted that by 2028, 70% of enterprises will abandon agentic AI built for them by vendors' forward-deployed engineers, citing rising costs and customers' inability to maintain the systems themselves.
Forward-deployed engineering (FDE) is a model in which a vendor embeds its own engineers inside a customer to build custom software. Gartner said customers often fail to gain the knowledge and control needed to maintain and develop these systems once the vendor's engineers leave, according to The Register. Gartner also predicted that through 2028, fewer than 20% of FDE engagements will turn recurring customer requirements into features of the vendor's core product.
Mukul Saha, a senior director analyst at Gartner, said success depends on getting the structure of the engagement right, from scope and incentives to governance, ownership and exit, as reported by The Register and Techstrong.ai. The Register also reported Saha saying some providers use the forward-deployed label for ordinary implementation work while charging premium fees. Gartner recommends using FDE only for problems that need deep product expertise, fast adaptation or close integration with the customer's environment.
A second data point comes from software teams. In its 2026 Global Technology Report, based on a survey of nearly 300 senior technology leaders, Bain & Company found developers complete 21% more tasks with AI coding tools, but code review time has risen 91% and developers run 47% more workstreams at once, CIO Dive reported. Leaders expect a 95% boost in developer productivity and 148% faster release cycles within one to two years, against current gains of 20% to 27%.
"The bottleneck has moved from writing code to trusting it," said Purna Doddapaneni, a Bain partner, according to CIO Dive. Bain recommends giving AI tools better context on codebases, automating testing, security checks and policy controls, and measuring end-to-end delivery rather than coding speed alone.
Finance teams report steadier results. KPMG's Q3 AI Finance Report, a survey of more than 1,000 senior finance leaders in 20 countries, found 75% use AI in 2026, up from less than one-third in 2024, CIO Dive reported. At least 71% said AI at least met their ROI expectations, but only 23% said it exceeded them. Organizations with stronger governance reported outcomes three to six times better than peers.
The numbers
- Enterprises Gartner expects to abandon vendor FDE-built agentic AI by 2028
- 70%
- FDE engagements expected to become core product features through 2028
- Fewer than 20%
- Rise in code review time with AI coding tools (Bain)
- 91%
- More tasks completed by developers using AI coding tools (Bain)
- 21%
- Finance leaders saying AI exceeded ROI expectations (KPMG)
- 23%
- Ascerta Series A
- $18 million
Why CEOs should care
For buyers and CIOs, Gartner's warning is about ownership. Before signing a forward-deployed engagement, ask who owns the code, the prompts and the intellectual property; how knowledge transfers to your staff; what it costs to run the system after the vendor's engineers leave; and what the exit plan is. If the vendor cannot answer in writing, assume you are buying dependence.
For CFOs, the Bain numbers show why productivity claims and returns can diverge. A 21% gain in tasks completed means little if review time nearly doubles and releases do not speed up. Ask engineering leaders for end-to-end measures, such as time from idea to production and defect rates, rather than lines of code or tasks closed. The KPMG survey suggests the same discipline pays off in finance: firms with stronger governance reported results three to six times better.
For boards, the gap between expectation and outcome is the risk to watch. Leaders in Bain's survey expect a 95% productivity boost within one to two years against 20% to 27% gains now. Ask management which AI projects have a measured baseline, who signs off on the results, and which vendor contracts would be hard to unwind.
The bigger picture
A market is forming around proving AI's value. On September 30, Ascerta, a startup that measures AI adoption, return on investment and compute use across tools such as Microsoft Copilot, Anthropic's Claude, AWS Bedrock and Salesforce Agentforce, said it raised an $18 million Series A led by Dell Technologies Capital, SiliconANGLE reported. The company said its customers average a 47% improvement in ROI; that figure is the company's own claim.
Together, the reports point to a shift from adopting AI to accounting for it. The cost now sits less in the model and more in review, maintenance and the people who keep systems running after launch.
What’s next
Expect procurement and audit teams to push for exit clauses, knowledge-transfer terms and outcome metrics in AI contracts through 2027. Watch whether vendors with large forward-deployed teams start turning custom work into standard product features, the conversion Gartner expects fewer than 20% of engagements to achieve.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
