On 18 June I spent the day at the Deel Summit, watching the HR industry quietly walk back its own hype. Enterprises are pouring billions into AI and counting the spend as proof it works. One of the speakers put it plainly: the spend is activity, not impact, and barely anyone in the room could show the impact.

The contradiction ran through the day. Self-driving cars, an app that tracks your pizza to the door, and the comparison of a new starter in the building still waiting for a laptop no one could provision on time.

The market has named the overclaim. AI washing is the regulator’s term for selling AI you do not have, and agent washing is the analyst’s term for calling a chatbot an autonomous agent. Underneath both sits one gap: the distance between what AI is being sold as and what it does without a human directing it.

MIT’s NANDA initiative measured that distance. Of the $30 to $40 billion poured into enterprise generative AI, 95% of organisations are seeing no measurable return, and only around 5% of AI pilots ever reach production with real impact at all [1]. MIT is blunt about the cause, and it is not the models. It’s simply the approach, the learning gap, and the failure to wire it into the actual work [1].

That is the inversion the hype was built to hide. What you buy is not an intelligence that chases your goal, it is an instrument that does nothing toward your outcome until a person points it, sets its limits and contains it.

On a recent call I said that if left to run on its own, agentic AI goes off and does its own thing, like an ADHD Spaniel. Capability with no human above it does not wait to be useful. It runs loose. People keep expecting otherwise because they treat it as a colleague. AI and LLMs are not human, and the whole overclaim, in my opinion, depends on people forgetting that.

This is not just my opinion as an operator. The same conclusion is arriving from three directions at once.

The regulator got there first, in the language of fraud. In March 2024 the SEC charged two investment advisers for claiming they used AI when they did not, and named the practice on the record: such AI washing hurts investors [2].

The analyst got there in the language of the market. Gartner called the agent-side version agent washing, ordinary chatbots and automation rebranded as autonomous, and reckoned only about 130 of the thousands of self-described agentic vendors were the real thing [3]. It expects more than 40% of agentic projects to be scrapped by 2027 on cost and thin value [3]. A year later it went further: a separate forecast that 40% of enterprises will pull their agents by 2027 over governance gaps found only after something breaks, because they run agents as binary, locked down or fully trusted, and that is the root of the failure [4]. Gartner’s fix is graduated control by autonomy level, every agent logged against its access, its data, its owner and its rollback: direction within limits, not prohibition and not blind trust [4].

The vendors got there in their own marketing, which is the language that counts, because nobody undersells until the truth has caught up. The HR platform says it on its own pages: the agents do not replace human decisions, the team keeps control of decisions and approvals, you build the agent to your own rules and thresholds [5].

The law is trickling it in too, from 2 August 2026 the EU AI Act requires high-risk systems to be built so a human can actually oversee them, with a stop and an override [6], and requires that people be told when they are dealing with an AI and not a human [7]. I have spent the last six months saying it is not human intelligence and people conflate the two. Brussels is about to make telling them the difference a legal duty.

So the overclaim sold autonomous intelligence, and the regulators, the analysts and the vendors are all, separately, walking it back to the same place. The intelligence was meant to be the human, and the human has to stay above the loop. That is people-side AI governance, the methodology kind, not the certification kind that checks whether you filled in the form.

It has a name, because I named it contAIn™ is the methodology layer that directs AI and puts humans above the AI loop, so every tool and agent you run works toward the outcome you have defined instead of one of its own. I created it from production failures, until the constraints that fixed them became a method. That is the evidence, and I have written a research paper about it [8].

The market sold you the car and forgot to tell you you need driving lessons. It is now admitting, from three different directions, that the driver was the point. The only question remaining is whether you are steering your AI, or still waiting, like the new starter, for the laptop to turn up.

References

[1] MIT NANDA (A. Challapally, C. Pease, R. Raskar, P. Chari), The GenAI Divide: State of AI in Business 2025, July 2025. Report PDF (hosted copy): https://www.artificialintelligence-news.com/wp-content/uploads/2025/08/ai_report_2025.pdf

[2] U.S. Securities and Exchange Commission, press release 2024-36, SEC Charges Two Investment Advisers with Making False and Misleading Statements About Their Use of Artificial Intelligence, 18 March 2024. https://www.sec.gov/newsroom/press-releases/2024-36

[3] Gartner, press release, Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027, 25 June 2025. https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027

[4] Gartner, press release, Gartner Says Applying Uniform Governance Across AI Agents Will Lead to Enterprise AI Agent Failure, 26 May 2026. https://www.gartner.com/en/newsroom/press-releases/2026-05-26-gartner-says-applying-uniform-governance-across-ai-agents-will-lead-to-enterprise-ai-agent-failure

[5] Deel, AI Workforce product page (FAQ: agents do not replace human decision-making; build an agent to your rules, thresholds and approvals), accessed 20 June 2026. https://www.deel.com/solutions/ai-workforce/

[6] EU Artificial Intelligence Act, Article 14 (Human Oversight), in force 2 August 2026. https://artificialintelligenceact.eu/article/14/

[7] EU Artificial Intelligence Act, Article 50 (Transparency Obligations), applicable from 2 August 2026. https://artificialintelligenceact.eu/article/50/

[8] Samantha Maeer, The Feed Loop: How AI-Generated Governance Documents Amplify the Patterns They Were Designed to Suppress, ResearchHub (peer-review sharing link). https://www.researchhub.com/paper/11303835/the-feed-loop-how-ai-generated-governance-documents-amplify-the-patterns-they-were-designed-to-suppress/reviews