The AI Implementation Heresy

Jul 8th, 2026

In “AI Is Working, Your Strategy Is Not” we showed adoption running ahead of strategy. Then in “Everyone Chats with AI, Almost No One Can Build With It” we mapped the gap between asking AI questions and building with it. Today we follow the builders to the morning after their agent goes live. That is when a quiet countdown begins.

💸 A $9 Billion Answer to the Wrong Question

The money is chasing real pain. Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027. MIT researchers found that roughly 95% of generative AI pilots delivered no measurable financial return. Boards see those numbers, panic, and reach for the oldest fix in enterprise software: hire an army.

We watched this movie before. For fifty years, the tech world has sent brilliant experts into struggling communities with the best technology money could buy, and left little lasting capability behind.

A case study is the nonprofit world: billions spent on technology, hackathons delivering quick fixes, and most of it quietly failing. Not because the technology was bad, but because we dropped it into places that were not ready to run it, and walked away without building capacity.

Dr. Kentaro Toyama spent a decade building technology for nonprofits, schools, and communities, watched well-funded project after project fail, and named the pattern the Law of Amplification: technology amplifies what an institution already is. It rescues nothing. His book, Geek Heresy, lends this issue its title, an open homage.

The parachute is back. This time, it carries AI agents and thousands of experts wearing someone else’s badge. Here is where the $9 billion landed, in about 60 days:

Note who is investing: the consulting firms are buying into the companies built to disrupt them.

Even the investors say so. Goldman Sachs’ Marc Nachmann, on the Anthropic venture: “Having the model alone doesn’t change your workflows or how you operate. You need people who can combine the technology with what’s actually happening in the business.”

🌱 Agents Are Living Systems

An agent sits closer to an employee than to software. It holds knowledge that goes stale, follows policies that change, and serves people whose needs drift. Across the 10,000+ agents we track, the pattern is consistent:

• 9 of 10 need weekly retuning: knowledge base updates, policy changes, prompt rewrites.
• Unmanaged for 90 days, usage falls 40%+ and satisfaction falls 50%+. Users don’t file complaints. They quietly stop showing up.
• IBM sees the same wall: 94% of enterprises report agent sprawl, and only 1 in 5 has governance that covers agents.
• Industry cost analyses point in the same direction (directional estimates, not our data): ongoing agent maintenance runs 15% to 30% of the initial build cost per year, and most budgets understate the true total cost by 40% to 60%.

Microsoft’s Judson Althoff, announcing the largest implementation bet in the company’s history, acknowledged the work continues long after the contract is signed: “Enterprise AI engineering expertise with deep industry knowledge is required to build a system that acts as a continuous loop of improvement.”

When the company with the most to gain from just selling licenses tells you the work never ends, believe them. And they are reinventing a commercialization channel that served them well for 50 years but will not serve them well over the next five.

🧭 The Managers You Need Already Work for You

In February, we published “The Two Percent”: in every company we study, roughly 2% of employees have quietly become the AI leaders nobody appointed. Today, our tracker reads 2% to 5% and climbing.

Those people are your agent managers. Subject matter experts, not technicians. Give every agent that matters a named owner and real hours every week, and pay these managers like people managers, because that is what they are. Their direct reports just happen to be digital. Underpay them, and they walk, leaving orphan agents drifting on stale knowledge until another expert adopts them.

The two percent are already doing the work, without the title or the hours. Find them, and give them both.

⏱️ How the Countdown Stops

The industry committed $9 billion in 60 days. An unmanaged agent starts to drift from delivering value in 90. Three moves stop the countdown:

📣 Feel free to use this data in your communications, citing "AI4SP" and linking to AI4SP.org.

📬 If this email was forwarded to you and you'd like to receive our bi-weekly AI insights directly, click here to subscribe: https://ai4sp.org/60​