LinkedIn publication text
Will more than 40% of agentic AI projects be canceled by the end of 2027?
That is what Gartner predicted on June 25, 2025. I built a 43-slide infographic to separate what holds from what does not.
What the primary source says
- It is an analyst forecast, not a measured result; the method is not disclosed in the press release.
- The poll of 3,412 attendees cited in the text measures investment intent and is not the basis of the 40%.
- Gartner itself projects that 33% of enterprise software will include agentic AI by 2028.
What the evidence shows
- RAND (2024): organizational and data causes weigh heavily; the “80% fail” figure is a third-party estimate.
- MIT NANDA (2025): the “95%” refers to generative AI pilots and comes from a preliminary report.
- McKinsey (August 2026): 40% of large companies now scale agents (27% in 2025), but only 37% see some EBIT.
- Gartner (April 2026): only 17% of organizations have deployed agents.
- MAST study: even after tuning, a multi-agent system reached 40.6% accuracy (it started at 25%).
Conclusion: the direction is plausible; the number is not verifiable. 40%, 80% and 95% measure different things and should not be added together.
How to change the trajectory: clear value before the pilot, the right tool for each task, measure reliability (pass^k), human validation, data and infrastructure, cost under control, and proof of autonomy from vendors.
Inside the infographic: a 2022–2026 timeline of “autonomous AI” and “agentic AI”, glossary, tables, importance and positive and negative implications on every slide, and a checked bibliography.
© Aridio Silva @aridiosilva · October 2026