Last year, in my first months of working with AI, I built about four brand and client websites and never charged more than 3,000 for any of them. Before AI I would have quoted that kind of job at 10,000 to 25,000.
Nobody asked me to charge less. I told every client I was using AI, because it worked in their favor, and I explained carefully what they got for the money and what they did not. Not one of them questioned a price. The discount was entirely mine.
How long a site took depended on its size, on whether there was a CMS, and on whether the design was unique or more generic. I should still have asked five to eight thousand for the paid ones. I did not, because it had felt too easy. It felt like cheating, and I was embarrassed to put a number next to something that came together that fast. I am still amazed every day at what you can build when you give an AI the right instructions.
The old price was a team
The 10,000 to 25,000 was never the price of a website. It was the price of a small team: a designer, a developer and a project manager, working together on a small business site. The 3,000 was the price of one person. In my head I had done a simple subtraction. The machine does the typing now, so most of the team is gone, so the price should fall to match.
That subtraction was wrong, and it took me a while to see why. The roles did not go away. Someone still had to decide what the site was for, design it, specify it precisely enough to build from, and check every page of what came back. A team used to do that between them. Now it was me, alone, doing all of it at once.
AI removed the typing. It removed none of the thinking, and the thinking is a large share of what a client pays for.
This is the efficiency penalty in its plainest form. Get faster at the job, charge by the time it took, and you have given yourself a pay cut for improving. I did it to myself without anyone asking.
What the client's question misses
A client once asked me whether using AI means everything can be done much faster. It is a fair question, and the honest answer is yes for some of it. Production work does go faster. Some things take as long as they ever did, and some take a long time for the AI to get right.
I gave that client the answer I would give anyone. Building with AI means a lot of thinking goes into a project. Where a team once shared the work, you now carry all of it. AI is not perfect, so you have to be prepared for that, check everything, and have your checks and balances in place. Quality comes from a workflow built to support it, because AI does not build something great by itself. I have a lot of engineering experience, which means I know which questions to ask and which rules to set, and a good part of the value sits there.
The clearest example is the build I did for CULT Collective. It connected many data points, and getting it right was slow, because everything depended on details. The AI could run for a long stretch, but when it reached something that needed a decision, it stopped, and everything stalled until I came back to the machine. The waiting was not free time. Some steps could not start before the one in front of them finished, and the whole time I was watching, checking and steering.
An hour the machine spends working is still an hour I am responsible for.
What I charge for now
I price fixed scope, based on the deliverables. The client knows what they get, what they do not get, and what it costs before anything starts. That also ends the argument about hours, because hours were never the thing being sold. What I sell is the outcome, the judgment about what to build and what to leave out, and my name on the result when something breaks.
None of those appear on a timesheet, and none of them got cheaper when the typing did.
There is a second reason the price does not simply fall with the hours. The time a project takes is no longer the whole cost of it. Part of why I can work fast is what I built before any client asked: the checks, the conventions, and tools like Listspace, a project board that my AI agents keep up to date. That took a lot of time and effort, and none of it shows up in the hours of any single job.
I have not used the board with a client yet. It is being tested now, and I will use it the next time I build for a client. The idea is that status reporting, a large part of what a project manager does, stops being a task, because the client can see the real state of the project without asking me. The judgment part of the job stays: what comes first, what waits, and what the client has to decide.
What I cannot tell you is what those websites should have cost. Five to eight thousand is my best guess today, and I notice it comes from the same place the 3,000 did: my own feeling about how hard the work was. The client never asked what it cost me in effort. They were asking what it was worth to them, and I never did the work of finding out. The embarrassment was information about me, and I used it as information about the price.