AI-assisted development

Anyone can ask an AI for a website now. Getting one that is fast, secure and still running in five years takes an engineer who knows what to ask for and what to throw away. We build with AI tools every day. We steer them, test what they produce, and ship only code we understand.

Start a project

What is included

How we go about it

  1. Write the rules down first

    Before any code, we write what the system must do, what it must never do, and how we will know it works. The AI works from that, and so do we.

  2. Steer, not just check

    We choose the tools, the libraries and the approach, and we use only systems we know well. The AI works inside those choices. It does not get to pick its own.

  3. Every change is a pull request

    Changes arrive in pieces small enough for a person to read properly. Each one is reviewed by an engineer and tested by tools for speed and accessibility before it is accepted. This site is held to those tests.

  4. Kept in GitHub, deployed from it

    Everything is committed to GitHub with its full history. The site or app is built and deployed from there automatically, so any version can be rebuilt or rolled back.

  5. A person signs it

    Someone with a name is responsible for everything that ships.

A good fit when

Work that used it

AI-assisted development questions we get

Software development where an AI model drafts the code, tests and documentation, and an engineer steers the work, tests it and decides what is kept. It is faster than writing every line by hand. It is not the same as letting the AI decide.

No. We steer it. We decide the architecture, the tools and the features, and we use only code and systems we know and can review. Every change goes through a pull request, is committed to GitHub, and is built and deployed by an automated pipeline, so there is a record of what changed and a way back.

Yes. We start by reading what is there and finding out what it actually does. Then we fix the faults, close the security holes and put tests around it so it stays fixed.

Only as safe as the review behind it. A model can produce code that looks right and still leaks a password or skips an access check. We review and test everything, the same as we would for code written by a person.

Often faster, and sometimes cheaper. The typing gets quicker. The thinking, the testing and the responsibility take as long as they always did.

Yes, and it shows where each side is strong. The brand is our original from 2008, rebuilt by hand by our founder, Trout, and our art director, Steven Hanlen, before any AI was involved. Our founder, Trout, set the direction and the vision and reviewed the work as it went. An AI agent, working inside those rules, wrote much of the code. Together we built in days a site that would normally take weeks, if not months. Each change still has to pass automated speed and accessibility tests before it is published.

Not well, in our experience. Art direction and branding are where AI tools are weakest, and it is hard to beat human creativity and talent. Our brands are drawn by human art directors. AI comes in afterward, to build quickly and accurately inside the system they set.

Often paired with

Need help with AI-assisted development?

Tell us what you have and what you need. A sentence or two is enough to start.

Start a project