One Thursday morning, a reviewer sat down with the private test version of a client’s new website and left twenty-one notes in a little under three hours. Each one was pinned to the page she was looking at, with a screenshot attached.
Most were small and useful. Add this partner. Link to that report. This step needs clearer wording. Two were praise. And a few were bigger asks tucked inside a casual sentence. Could we add a loan calculator? Could real estate agents post their own listings? Shouldn’t we be aiming more at existing businesses?
Those are not work orders. They are a feature, a platform decision, and a change of strategy, and each deserves a conversation about scope and cost before anyone builds anything. My AI agent, the software that does the building, would have built all of it if I had handed it the pile. Cheerfully.
That is the problem with speed.
The question on my AI Mastermind call this week was how to speed up delivery with AI without creating operational chaos at machine speed. The speed is not the hard part. The hard part is what the speed lands on.
AI is an amplifier. It does not fix a process. It runs the process you already have, faster and at a volume you have never seen. If the process is sound, you get more of a good thing. If it is loose, you get more of the looseness before you have had time to notice.
I call this the Amplifier Rule. AI multiplies whatever is already there.
When building was slow, the slowness did some of your sorting for you. A person had to read the request, think about it, and do the work, and somewhere in that effort a little judgment leaked in. When building is fast, you have to sort on purpose.
So I never let the agent build straight from a note. In the first version of this feedback board, I told my agent to review every note marked New, ask me clarifying questions about each one, and only then write a ticket, which is just a written work order. The next session starts the same way. The agent reads the project board, the project files, and the product requirements document, the plain-English description of what we are building. Then it drafts the prompt for that day’s batch of work and asks what it is unsure about before anything begins.
The feedback tool is the front door, and the easier it is to use, the more you get. Tuesday night I soft launched that client’s new website at their board meeting. During the walk-through a member asked whether there was a final version of a report labeled draft. I captured it on the spot, in front of everyone, and told the room that this is exactly the kind of note I want from them.
Once the board worked, I turned it into a skill in Claude Code, a saved set of instructions I can call up on any project. When a build reaches its test version, I invoke it, and every project’s feedback shows up on my Feedback Hub. A system you can package is a system worth amplifying.
None of that works without everything behind the door. Opportunity evaluation before I say yes. Wireframes, which are rough sketches of each page, before anything gets built. A design system so the pieces match. Written requirements. Tickets on a board, with a rule that nothing gets built unless it is a ticket. None of it is exciting, and none of it is AI. All of it is what the AI is amplifying.
So what do I actually do about this. Slow down once, so you can speed up for good, and show your work.
Remember the teacher who asked you up to the board to show the class how you solved the algebra problem? The answer was not enough. You had to show your work. AI hands you the answer in seconds. Your job is to be able to show the work behind it: the steps, the rules, and the decisions.
Start with a sketch. I keep a yellow legal pad on my desk for exactly this. Break down the steps a request takes, from the moment someone asks for it to the moment it is done. A flow diagram, boxes and arrows, a simple list. It does not need to be pretty. That sketch is the work you show.
It will take a few hours, and it will feel like the wrong direction. You are trying to move faster and I am telling you to sit down with a legal pad. Do it anyway.
AI does not know when it is wrong. It works out the most likely answer and keeps moving, and every wrong guess becomes what the next one builds on. A mistake at human speed is an irritation. At machine speed it is a runaway freight train headed for town. By the time you notice, you are not fixing a mistake. You are cleaning up the wreckage.
AI will amplify whatever you give it. Make sure it is something worth amplifying.
Richard D. Stuart
The Weekly Prescription
P.S. If you suspect your process is looser than it looks, that is a good thing to find out before you add AI to it, not after. A free 20-minute working call, not a pitch: yourwebguy.org/become-a-client