MyJackfruit

Help
Browse topics · Building the first step

Building the first step

← Back to How big should my first version be?

How long does it take to build a step, and my first version?

Checked against GilbApp d529620.

There is no fixed time: each step is kept as small as possible, and your first version takes as many steps as your plan needs.

Deeper

No page and no standard gives a set time for a step. A step is a small piece of your app that is built, put live and measured in one short round. Kai-Zen, the planning agent (an agent is your AI following one written procedure for one kind of work), makes each step as small as possible while it still brings something that can be measured, such as how many people can now book a time. So a step is meant to be small, but how long it takes depends on how much it holds, on your AI, and on you.

What the time goes on. Agent.Coder, the agent that writes your app's code, builds the step on your computer. Then another agent checks it, and a separate agent tests it. If a test fails, Agent.Coder fixes it and tests again; it stops after three different failures it cannot fix without more scope or your input. Kai-Zen always stops and asks for your go before it puts a step live, and then asks you to check it live; it fixes what you send back and asks again until you approve. Each of these waits for you, so a step is done sooner when you answer soon. The answer How does Kai-Zen run a whole delivery cycle? shows the whole round.

Your AI's own limits count too. With a Claude plan, Pro, Max or Ultra, you pay a set price each month, and the plan has usage limits, on most plans every 5 hours and every week. When you reach a limit, you wait until it resets, unless you have turned on usage credits, Anthropic's pay-as-you-go use beyond a plan's limits; the answer What will my AI cost while I plan and build, and can I set a limit? says more. When one agent hands work to another, the hand-off may run up to 15 minutes, and up to 30 for some planning agents.

Your first version. No rule sets how many steps it takes. The first step sets up the base of the app, a ready-made core app with a sign-in, plus something useful for you if it can, so you have something real early. At the end of planning Kai-Zen suggests the order of the steps, with what each should bring and cost. That order is a forecast only: before each step Kai-Zen picks it for real and changes the forecast with what it learned. After each step it shows you what was measured and the next step, and goes on until you say re-plan or done. The answer How big should my first version be? tells how small to start.

If you have a date to meet, or a most-time you can spend, tell Kai-Zen in your own words. Time is one of the things the work uses up, and the plan can hold the most of it you plan to use as a budget. If the spend so far is over the budget, Kai-Zen stops and asks you whether to go on.

Your computer must stay on while your AI builds; the answer Does my computer have to stay on while my AI builds? says why.

This answer is written for a computer screen.

Source: Standards: Kai-Zen.ValueDeliveryStepRules (First, Small, Forecast); Kai-Zen.Cycle (Build, Check, Live, Human, Next); Kai-Zen.HumanCheck (Ask, Fix); Agent.Coder (AfterIncrement: three distinct failures). gilb-connector 0.6.1 (https://gilb.com/app/gilb-connector.mjs, read 13 November 2026): HOP_MAX_MS_DEFAULT 15 minutes, 30 minutes for planning roles. Live answers, sentences shared: When do I actually start making the app? (a step), How big should my first version be?, How does Kai-Zen run a whole delivery cycle? (Budget, Development Resource, the round), What will my AI cost while I plan and build, and can I set a limit? (Claude plan limits; https://claude.com/pricing).