--- title: AI Projects Don’t Fail at the Technology. They Fail the Moment the Pilot Meets Monday Morning — Which Is Exactly Why the AI-Native ART Exists. url: https://agile-agilist.com/ai-projects-dont-fail-at-the-technology-they-fail-the-moment-the-pilot-meets-monday-morning-which-is-exactly-why-the-ai-native-art-exists/ description: The model works in the demo. It breaks when it meets people, processes and the operating environment. What BCG, McKinsey, MIT and DORA say about where AI really fails — and why the AI-Native ART is built for that exact point. --- # AI Projects Don’t Fail at the Technology. They Fail the Moment the Pilot Meets Monday Morning — Which Is Exactly Why the AI-Native ART Exists. The demo works. It always works. The model summarizes the contracts, drafts the code, triages the tickets, and the room is impressed. Then it meets Monday morning. Someone has to own the output. Someone has to check it before it reaches a customer. The workflow it slots into was designed for a world where a human did that step at human speed. The data it needs lives in three systems with three owners. And nobody’s job description changed. That is where AI projects fail. Not in the model. In the meeting between the model and the organization. ### The research says the same thing four different ways **BCG** allocates AI effort as 10% algorithms, 20% technology and data, and 70% people and processes. Seven tenths of the work sits outside the technology entirely. **McKinsey’s** 2025 State of AI survey tested 25 attributes against EBIT impact from generative AI. The one with the largest effect was workflow redesign. Yet only 21% of organizations using gen AI had fundamentally redesigned even some of their workflows — and more than 80% reported no tangible enterprise-level EBIT impact yet. **MIT’s** research into enterprise AI found that of the systems organizations seriously evaluated, about 5% reached production. The core barrier was not infrastructure, regulation or talent. It was learning: systems that do not retain feedback, adapt to context or improve over time. Mid-market firms moved from pilot to production in roughly 90 days; large enterprises took nine months or more. Same models. Different organizations. **Google’s DORA** research put it most bluntly: AI is a magnifier. Teams with healthy systems get better. Teams with weak processes ship low-quality work faster. Four methods, one finding. The binding constraint on AI value is organizational. ### Three places the pilot breaks **People.** Who verifies the output? Who owns the decision it informs? Generation is now elastic; human judgement is not. When verification is implied rather than assigned, responsibility quietly leaks out of the system. **Process.** The AI gets dropped into a workflow built for human-speed steps. The step gets faster. The approvals, handoffs and queues around it do not. Net flow barely moves — which is precisely what McKinsey’s workflow-redesign finding is measuring. **Operating environment.** Data scattered across owners. Governance written for annual review. Funding approved once a year for a cost that moves every week. The pilot ran in a sandbox where none of that existed. Production is where all of it exists at once. None of these are solved by a better model. ### Why the answer is an ART — not a team, not a tool A single team can make AI work locally. It cannot redesign a workflow that crosses five teams. It cannot fix data owned somewhere else. It cannot set the guardrails for a whole product. A tool can do none of those things. This is the reason AI-Native SAFe puts the AI-Native ART at the centre of the model. An AI-Native ART is a team of teams organized around one or more products. It connects the work of its teams to measurable outcomes, lets learning travel between teams, and coordinates the shared technology and AI safeguards they all depend on. Put simply, it is the smallest unit of the organization that owns the whole contact surface between AI and operational reality — the people, the cross-team workflow, and the environment. It is where local AI wins become system-level results. Most organizations are not structured this way. They are organized by function or region, so AI arrives as a dozen separate pilots in a dozen separate teams — and nobody owns the workflow between them. That is the structural version of the MIT finding: plenty of pilots, very little production. ### The spend angle, in one line If 70% of AI value lives in people and processes, a budget that puts nearly all of its AI money into tools is aimed at 30% of the problem. ### Three questions for your leadership team **1. For our biggest AI pilot, which workflow did we redesign around it — or did we drop it into the old one?** **2. Who is named as the verifier of AI output in that workflow, and how much review capacity do they actually have?** **3. Which organizational unit owns the outcome that pilot is supposed to move — and does it span every team the workflow crosses?** AI transformation is not a technology rollout. It is an operating model decision. The AI-Native ART is where that decision becomes real. The world has changed. The leaders who notice will be the ones the next decade is built around. I write on Thursdays at agile-agilist.com. #### Sources - BCG, 10-20-70 principle, as described by Matthew Kropp (VentureBeat) - McKinsey, The State of AI in 2025: How organizations are rewiring to capture value - MIT NANDA, The GenAI Divide: State of AI in Business 2025 - Google DORA, 2025 State of AI-assisted Software Development - Scaled Agile, AI-Native ART (framework guidance)