Most AI programmesnever get belowthe waterline.
Licences, pilots, a chatbot on the website. All of it visible. None of it load-bearing.
Three levels. Most companies stop at two.
The first two make a company faster. Only the third makes it different. It is also the only one that requires changing the organization, which is exactly why most stop just short of it.
Give people tools
Everyone gets a licence and finds a few hours back. Useful. Invisible on the income statement.
Speed up the process
The same workflow runs faster, or with fewer hands. Efficient. Still the same business.
Redesign the work
Rebuild the process around what AI can now do and what it needs to know. The only level that changes outcomes, and the only one that asks the organization to change.
A keel is not the part you show people.
It is the part that lets you carry sail without going over. Institutional knowledge, redesigned process, and clear ownership are the same kind of weight. Nobody sees them in a demo. Everything above them depends on them.
We place a company honestly on a five-stage maturity model, then move it. The hard part is never the technology.
The work that quietly eats the week
A status update nobody reads. The same report rebuilt every month. A summary written after every deploy because somebody has to. None of it is hard, and because it is never the most important thing, it is never the thing anyone fixes.
Stop being the trigger
Redesign the process around what a system can now do on its own, give it the context it needs, and put a gate wherever judgment belongs. The work does not get faster. It changes hands.
Judgment, which was the point
Deciding what matters, what good looks like, and when the answer is wrong. That is the part nobody can hand off. Everything else was overhead you had learned to live with.
This is Oliver. It does the work you keep redoing.
Oliver is the AI system we run our own company on. A Shadow is a workflow it owns end to end: it watches for the moment, gathers the real context, reasons about it, and delivers the result. You stop being the trigger and start being the editor.
Every run leaves a trace. Anything that ships to a human waits for approval. When there is nothing worth reporting, it says so rather than inventing something.
Get access- TriggerDeploy event received from GitHub
- GatherRead the diff, the commit history, and the release notes
- ReasonDraft a plain-English summary for the team
- GateHold for approval before anything is posted
- DeliverPost to the channel, and record the full run trace
How we think, in the open.
The Tools Are Fine. The Organization Isn’t.
AI tools can make work faster, but they do not automatically make a company smarter. This article explains why many AI initiatives stop at pilots, licenses, and small productivity gains, while the real value comes from redesigning work, capturing institutional knowledge, and clarifying who owns decisions. The companies that pull ahead will be the ones willing to change the organization, not just adopt the technology.
ReadFormer EY-Parthenon Leader Allan Kyle Joins Keel3
Allan Kyle spent three decades in it before AI became a buzzword. He studied neural networks in the 1990s, helped build machine learning tools for banking risk, and most recently advised private equity firms at EY-Parthenon on technology strategy for acquisitions. Now he's joining Keel3 to put that experience to work for mid-market companies ready to move from AI experimentation to measurable results.
ReadTOKENOMICS: You don't have a token problem. You have a staffing problem.
Companies are watching their AI bills climb and reaching for the wrong fix. The answer is not less AI or a cheaper model. It is the same thing every agency that survives eventually figures out about people.
ReadAsk one question about any process you run.
Why is a person still the trigger for this? If you cannot answer it well, that process is where your transformation actually starts. We are happy to work through the first one with you.