Build What Your Work Needs, with an AI Assistant

The path from an idea to a tool you use every week — what to decide, how to set up, how to ask, how to check what it built, and how to keep it running — for people who know their work but have never written code

Build the app or tool your work needs by directing an AI assistant, without learning to program first: the five steps (decide, set up, direct, check, ship), where to start for your situation, and what Collegica has built this way.

You know exactly which tool would make your work easier: the spreadsheet that should be an app, the follow-up list nobody keeps, the report you rebuild by hand every month. Until recently, getting it built meant learning to program or paying someone who had. Now an AI assistant can write most of the code.

What it can’t do is decide what’s worth building, notice when it got something confidently wrong, or keep the thing running next month. That part is yours, and it’s what Collegica teaches. You don’t need to learn a programming language first. You need to learn to direct the assistant and check its work.

NoteThe path in one minute
  1. Decide what to build, and how small the first version is.
  2. Set up a place for the work that the assistant can read.
  3. Direct the assistant in pieces it can finish.
  4. Check what it built, without being a programmer.
  5. Ship it, and keep it working.

Each step below says what it means, what to read, and what is still being written.

Where to start, by what you need

If you are… Start with Then
Running a small business Train It Before You Use It: AI for a small business, one phase at a time Let It Open the Folder: let it work on your own files
Writing a thesis or research Move Your Thesis from Word to Plain Text Python for Reproducible Research
Managing your own money A Year of Spending, from Your Own Statements and the OWL Planner app Planning to Age Well, with AI
Changing what you do after 45 Finding Your Next Job, with AI Planning to Age Well, with AI
Already a developer One Repository, Three Agents: Claude Code, Codex and Antigravity on one repository An Agent on Your Own Machine
Worried about privacy Let It Open the Folder, especially “A folder is not a model” An Agent on Your Own Machine

Step 1: Decide

Most failed projects fail here, before any code. Deciding means writing down three things: what problem the tool solves and for whom, how small the first version is (one screen, one task), and what “done” looks like so you can tell when to stop. That page of notes is also what an assistant needs to build from.

  • Start small on purpose. Train It Before You Use It shows the method in a business: automate one step, watch it, let the next problem tell you what to build next.
  • Use AI to decide, too. Ask it to interview you about the problem before it proposes anything.
  • Coming: a guide to writing a spec an assistant can build from.

Step 2: Set up

An assistant does its best work in a folder it can read, with your rules written down where it will find them. Setting up means: a folder (and, as soon as you’re ready, a git repository, so every change can be undone), an instruction file the assistant reads every time (AGENTS.md or CLAUDE.md), and tools that install the same way on every computer.

Step 3: Direct

Directing means asking for work in pieces the assistant can finish and you can check: one feature, one fix, one page at a time. It also means choosing the assistant, and knowing when a free or local model is enough.

Step 4: Check

An assistant will sometimes be confidently wrong. Checking means running the thing, testing it against cases where you know the answer, reading what changed (a “diff”) well enough to spot surprises, and keeping a person in the loop for anything that reaches a customer.

Step 5: Ship

Shipping means the tool runs somewhere you and others can use it, and still works next month. For most people that is a web page or an app that runs in the browser, published for free.

  • The Apps page: two apps that run entirely in your browser, each in its own public repository.
  • Coming: a web app built and deployed with an assistant, start to finish, and the same for a phone app.

Built this way

Collegica is its own case study. OWL Sessions, the exercise app, was built by directing Claude: all eight commits that created it are co-authored by the assistant. The 3D campus, the narrated videos and most of this site were built the same way: 126 of the 207 changes made to the site so far carry an AI co-author. The source is public, so you can read how.

Get the next step by email

New guides on this path, and the course being built alongside them, go out on the Mechatronics3D newsletter first.

Subscribe on Substack