When I was sixteen, I started knocking on doors in the neighborhood with business cards I had printed at home. They were terrible. The font was wrong. The color was wrong. The card stock was the cheapest the office supply store sold. I was offering computer help to families who had just bought their first home computer and could not figure out the printer.
I did not have a business plan. I had a rectangle of paper with my name on it and a sentence that promised I would come figure out whatever was wrong.
The work was not always great. Some weeks I made twenty dollars. Some weeks nothing. But every job taught me something about what people wanted, what they would pay for, what they struggled with. By the time I started what would become Business Builders, I had run more experiments than most adults run in a decade.
I think about that kid a lot lately. The model he was using - knock on doors, see what people say, adjust the offer, knock on more doors - is the model every business in 2026 should be running. Most are not.
The old model is broken
The old playbook for new ideas looked like this. Form a working group. Produce a strategy document. Get it approved. Build a project plan. Allocate resources. Schedule a kickoff. Run the project for a quarter or two. Hold a launch. Measure results six months later.
That model worked when the environment moved slowly. Twelve to eighteen months from idea to learning was acceptable because the world stayed mostly stable across that window.
That window has collapsed. By the time you finish the project plan, the conditions you wrote it for have shifted. By the time you launch, the assumption underneath the launch is two versions out of date.
Big plans are actively expensive now, because every month they sit on the runway is a month somebody else is shipping a smaller, faster version of the same idea and learning what you should have been learning.
What replaces it
What replaces it is small experiments. Cheap. Fast. Specific. Time-boxed. Designed to learn one thing, not to launch a final thing.
Here is the version we run at the agency. When somebody on the team has an idea, the question is not "what would it take to build this." The question is "what is the smallest, ugliest version of this we could ship in two weeks that would tell us whether the bigger version is worth building."
Two weeks. Smallest. Ugliest. Ship it.
Sometimes the answer is a prototype. Sometimes a single landing page with a sign-up form. Sometimes a manual workflow that pretends to be automated, just to see if anyone wants the output. Sometimes a one-day workshop with five clients to test whether they would actually pay for the thing we think they want.
The point is not that the small thing replaces the big thing. The point is that the small thing tells you whether the big thing is worth building, before you spend a quarter discovering it was not.
This was always the better way to operate. AI has just made every individual experiment dramatically cheaper. The thing that used to take a developer two weeks to spike now takes one person an afternoon. The market test that used to need a budget now needs a few hours.
When the cost of an experiment falls by an order of magnitude, the right number of experiments rises by the same order. If you are running the same number of experiments you ran two years ago, you are dramatically under-investing in learning.
action
1. Pull one idea off the planning shelf that has been there more than a quarter. 2. Define the smallest, ugliest version you could ship in two weeks. 3. Time-box the window. Ship whatever exists at the end, even if to one client. 4. Write down what you learned in a single paragraph, not a deck. 5. Adjust and run it again, or commit, or kill it.
What is the small experiment you have been postponing because you were waiting for the plan to be ready?
