Every founder and executive is feeling it: the pressure to implement Artificial Intelligence (AI) now. Not next quarter. Not after the next planning cycle. Now. The competitive anxiety is real, and so is the noise, a constant stream of case studies, product launches, and commentary suggesting that organizations not yet running AI-powered workflows are already falling behind. By early 2024, 65 percent of organizations reported regularly using generative AI, nearly double the figure from just ten months prior ^[1]. The problem is not that leaders are paying attention to AI. The problem is that urgency is substituting for judgment, and in that substitution, the human “I” (the intentionality, ownership, and critical thinking that make implementation succeed) goes missing entirely.

The most common failure pattern in AI adoption is not a technical one. It is conceptual. Organizations rush to select tools before they have defined the problem those tools are meant to solve. The result is what strategists sometimes call solutions in search of problems: a new capability deployed against a vague objective, generating activity but not outcomes. Research from the RAND Corporation, based on interviews with 65 experienced data scientists and engineers, identified misunderstandings about project purpose as the single most common cause of AI project failure, ahead of data gaps, infrastructure limits, and technical complexity ^[2]. When asked why they are automating a particular process, too many teams default to “because we can” or “because competitors are.” Neither is a business case. Both are symptoms of trend-following dressed up as strategy.

| “Urgency is substituting for judgment, and in that substitution, the human ‘I’ goes missing entirely.”

The deeper issue is that AI deployment forces a tradeoff that most organizations have not made explicit. Efficiency, risk, and quality do not move in the same direction when you automate. AI can compress the time and cost required to produce output. It can also introduce new categories of error, reduce the human judgment applied to edge cases, and erode the quality dimensions that are hardest to measure but most visible to customers. McKinsey’s 2024 global AI survey found that 44 percent of organizations had already experienced negative consequences from generative AI use, with inaccuracy identified as the most prevalent risk; yet only a minority had implemented comprehensive mitigation strategies ^[1]. Efficiency alone is not sufficient justification for implementation. It is one variable in a more complex equation, and treating it as the only variable is how organizations create problems they did not anticipate.

Consider a mid-sized professional services firm that begins using AI to draft client communications. In the first month, response times improve and the team handles a higher volume of inquiries with the same headcount. The efficiency case looks strong. Six months later, a long-standing client mentions, almost in passing, that the correspondence has begun to feel impersonal: formulaic, stripped of the specific language and attentiveness that characterized the relationship. No single email was wrong. But the cumulative effect was a detectable shift in tone that the client registered before the firm did. The “I,” meaning the practitioner’s voice, judgment, and relational awareness, had been automated out of the loop. This pattern has a structural analog in the broader data: the S&P Global Market Intelligence 2025 survey of over 1,000 enterprises found that 42 percent of companies abandoned most of their AI initiatives, a dramatic increase from 17 percent the previous year, and that organizations with higher project failure rates reported greater concern about reputational damage with both customers and employees ^[3].

What founders and decision-makers need is not a new tool. They need a disciplined set of questions to ask before implementation begins. Is the problem clearly defined, or is the team solving for something vague? Is the underlying process stable enough that automating it will produce consistent results? Will this create meaningful value for the customer, or only for internal metrics? What level of risk is acceptable if the system produces an error, and who (which human “I”) is responsible for catching it? RAND’s research is pointed on this: successful AI projects are laser-focused on the problem to be solved, not the technology used to solve it, and leaders should commit each product team to a specific, well-scoped problem for at least a year before expecting durable results ^[2].

Central to this framing is the question of human oversight. Not every AI-assisted process requires the same level of review. Some outputs, such as internal summaries, first drafts, and data formatting, carry low stakes and warrant light oversight. Others, including customer-facing communications, financial analysis, and compliance documentation, demand substantive human judgment before anything is published or acted upon. The “I” must remain in the loop wherever the consequence of error is high. McKinsey’s 2025 findings reinforce this directly: high-performing organizations treat AI as a catalyst for redesigning workflows rather than replacing judgment, and they are significantly more likely to implement risk governance structures across the full Artificial Intelligence lifecycle ^[4]. The right level of oversight is calibrated to the consequence of error, and that calibration is a leadership decision, not a default setting.

| “The organizations that get durable value from AI are not the ones that move fastest. They are the ones that never let the ‘I’ leave the room.”

There is a broader strategic point worth making directly: moving fast is not always better. The organizations that will get the most durable value from AI are not necessarily the ones that implement first. They are the ones that never let the “I” leave the room. McKinsey’s 2025 survey found that meaningful enterprise-wide financial impact from AI remains rare, concentrated among a small cohort of high performers who invest in governance, redesign workflows before selecting technology, and scale with discipline rather than speed ^[4]. Delaying a poorly scoped implementation is not falling behind. It is avoiding a mistake that would cost more to unwind than it ever would have saved.

AI is not the problem. The technology works. What fails, consistently and predictably, is the absence of intentional human judgment before, during, and after deployment. When AI has no “I,” meaning no ownership, no defined problem, and no calibrated oversight, that is the problem. Founders who want to use AI effectively should start not with tools, but with clarity: about the problem they are solving, the tradeoffs they are accepting, and who remains accountable when the system falls short.

References

[1]A. Singla, A. Sukharevsky, L. Yee, and M. Chui, “The state of AI in early 2024: Gen AI adoption spikes and starts to generate value,” McKinsey & Company, May 2024. [Online]. Available: https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-2024

[2]J. Ryseff, B. F. De Bruhl, and S. J. Newberry, “The root causes of failure for artificial intelligence projects and how they can succeed: Avoiding the anti-patterns of AI,” RAND Corporation, Rep. RR-A2680-1, 2024. [Online]. Available: https://www.rand.org/pubs/research_reports/RRA2680-1.html

[3]S. Sakoui, “AI project failure rates are on the rise: report,” CIO Dive, Mar. 2025. [Online]. Available: https://www.ciodive.com/news/AI-project-fail-data-SPGlobal/742590/ (citing S&P Global Market Intelligence, 2025 survey of 1,000+ enterprises.)

[4]McKinsey & Company, “The state of AI in 2025: Agents, innovation, and transformation,” McKinsey & Company, Nov. 2025. [Online]. Available: https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai