A room of AI operators gets together and talks about how to help small and mid-market business owners adopt AI. The framing is benign. The implementation is honest. The operators want the business owners to thrive.
It does not matter. The narrative ends in the same place.
The Capture
A business owner has twenty years of industry knowledge in their head. Most of it is tribal, undocumented, and scale-limited: it cannot be taught to a new hire fast enough to matter, and the owner cannot personally serve more customers than they already do.
An AI services operator walks in, installs a Personal Agentic OS, and starts capturing the owner's decisions. Every workflow gets written down. Every judgment call gets externalized into a reusable skill. Over six to twelve months, the harness knows how the owner thinks.
That harness is now an asset. The owner owns a copy. The operator owns the template. One of those two parties is about to use the template to serve the owner's competitors. The question is which.
Intentions Do Not Matter
The business owner does not have a way to verify, after the fact, that the operator did not keep a copy of what they captured. They have to trust on intent, on reputation, on the contract. Most operators are trustworthy. Some are not. And the ones who are not destroy the reputation of the ones who are, because the narrative writes itself: these consultants extracted my edge, automated me out of my own industry, and took my people with them.
Whether any given operator did that or not is beside the point once the story is running. Every AI services firm gets accused. Some deserve it. The category gets painted with the same brush.
This is the default trajectory. Humans love scapegoating, and AI services firms are a perfect scapegoat: legibly tied to a visible technology, visibly profiting from the client's dependency, and structurally positioned to benefit from the client's knowledge. The set of people talking about this right now is small. In two years it will be mainstream.
The Practitioner Version
The pattern shows up one layer down too. Agentic work captures what the worker does. The worker's workflow is the agent's training data. Some portion of workers will wake up and realize that the organization asked them to use the harness as a productivity tool, and what the harness did was externalize their brain into organizational IP. The individual version of the revolt is quieter and runs on the same logic.
The Move
If you run an AI services company, the practical posture is as follows.
- Name the dynamic in the first meeting. Tell the owner what the harness will capture and what it will not. Put it in the contract.
- Give the owner sovereign ownership of their own context. Plain files on their machine. See your brain shouldn't be SaaS for the posture and why a vendor-hosted workspace fails this test.
- Do not reuse an individual client's captured skills against their competitors, even in sanitized form. If you need a vertical template, pay a different client to build one with you on the books.
- Publish your internal playbooks before you sell them. The operator who is open about methods cannot be accused of operating in the dark.
If you are a business owner considering hiring an AI services firm, ask the operator how they handle the captured context after the engagement ends. If the answer is fuzzy, the answer is no.
The Larger Implication
The category of "AI consulting" as it is being practiced in 2026 is structurally fragile. Practitioners who care about staying in business for a decade should think about whether their offer is to extract the dinosaur's edge or to build something new alongside the client instead. The first lane has a backlash on its calendar. The second lane does not.
The consulting model that captures and reuses tribal knowledge looks like a cash machine until the day a critical mass of clients realizes what was captured. Plan for the day before it arrives.