Prototyping is easy now. Finishing one thing is the hard part

Let anyone prototype, cheaply and often. Let only a few things through to a maintained tool, past a gate with four questions, and give each one an owner before it goes live. Most prototypes should die at the gate – that's what prototyping is for.

By Matthew Steele4 min read28 October 2026

A desk with a row of sticky notes under a sign reading Parked, and one printed procedure in a ring binder beside the keyboard
Illustrative scene, not a customer's premises.

Why building and testing never stops

A prototype takes an afternoon and gives you a demo. Finishing the same thing takes 6 weeks and gives you a procedure, a log, an owner and a list of things that went wrong. One of those is a lot more enjoyable, and in a business where nobody owns the gate between them, the enjoyable one is what people keep doing.

We don't think it's a discipline problem. It's a missing structure. Nobody's said where a prototype goes when it's done, so it either quietly becomes a live tool with no owner, or it quietly dies and whoever built it starts the next one.

Two lanes with different rules

Prototype lane Product lane
Who Anyone A named builder with a named owner
Time A week at most 6 to 10 weeks
Data Sample data or your own files only Live systems, through the connections IT controls
Standard "Does it show the idea?" "Does it do the job 40 times without a surprise?"
Ends with A 10-minute show-and-tell and a decision A procedure, a log, a stop point and a monthly check
Count Unlimited Capped by the maintenance budget

The prototype lane has no connection to live data on purpose – that's what makes it safe to let everyone in. Who may build live tools has the rules for the other lane.

The gate

Many prototypes, one gate, few productsA wide region on the left holds many prototypes: anyone, a week at most, no live data. In the middle, a gate with four questions: who uses it weekly, what number moves, who owns it, what does it touch. On the right, a narrow region of maintained tools with an owner, a log and a stop point, its width set by the maintenance budget. Below the gate, the parked list for prototypes that answered three of four.PrototypesAnyone. Any idea. A week at most.Sample data only, nothing live.Unlimited, and most should die here.The gateWho uses itevery week?What numbermoves?Who owns it?What doesit touch?Maintained toolsAn owner, a log, a stop point, a monthly check.The width is the maintenance budget: about an owner hour a week each.The parked listthree answers of four, and a note
Lots of prototypes, one gate, a few products. How wide the right-hand end is depends on the maintenance budget, not on how many good ideas you've got.

Four questions, asked by your AI lead with the department head, in a 15-minute conversation:

  1. Who'd use this every week, and have they said so?
  2. What number moves if it works, and what's that number today?
  3. Who owns it after it goes live, by name, and have they agreed?
  4. What does it need to read or write, and who grants that?

A prototype with four answers goes into the product lane. One with three goes on the parked list with a note on which answer's missing. The rest get a thank-you and get closed. In our experience something like 1 in 6 prototypes should pass – if everything passes, you haven't got a gate.

The maintenance budget

Every live tool costs its owner about an hour a week once it's settled: checking a sample, answering questions, changing something when the process moves. Count the hours your owners genuinely have and that's the number of tools your business can maintain. For a company of 50 people it's often 4 or 5. When the budget's full, a new tool means an old one gets retired, which is a healthy conversation to be pushed into.

Keeping ideas flowing while you finish things

Finishing doesn't have to kill off ideas, as long as the prototype lane stays open and cheap. The habits that keep both going, in our experience:

  • A monthly prototype hour where anyone shows anything. What a weekly show-and-tell looks like has the format.
  • A parked list people can see, so a good idea that failed the gate isn't lost and the person who had it can see why.
  • Credit, said out loud, for killing a prototype early. Someone who tried something for two days and reported that it doesn't work has done useful work.
  • One product in the product lane at a time per department. Two is a queue, not a programme.