aaron8.
  1. Home
  2. AI development

AI does the typing. I do the knowing.

Most people paste a prompt in and hope. I run AI the way you would run a delivery team: a written specification, adversarial review, and proof against things the AI does not control. The result is software that matches what your business actually does, and you can inspect the evidence before you spend a dollar.

Describe the job See the finished thing See it live first

Live proof

Talk to it. Then read what it did not do.

A finished customer assistant for a small business. Pick the business, ask it something a customer would ask, and watch it answer from what it actually knows, book the next step, and hand anything sensitive to a human instead of guessing.

live demoA working AI assistant

This working build runs in your browser and needs scripts enabled. Everything on the page describes what it does; turn them on to use it.

That is the finished product, not a chatbot bolted onto a website: answers from your facts, actions it is allowed to take, and a clear line where a person takes over.

The work

What this unlocks.

Internal tools in weeks

The job tracker, quoting screen or portal that used to be a quarter's project is now a matter of weeks, because the specification and the build are the same brain and there is no handover for the meaning to go missing in.

Whole processes automated

Not one Zap: the intake, the rules, the system it lands in, the exceptions, and the report somebody actually reads. AI handles the volume; the analyst work decides what the rules really are, which is the part that was always hard.

Agents wired into your systems

Out of the box an AI can only talk. I build it hands: reading your CRM instead of guessing about it, booking the job instead of describing one, a customer chatbot that knows your stock and your policies instead of inventing them. Two of mine are live below; the same approach reaches accounting, field service, a warehouse, anything with an API.

Straight answers on where AI pays

Some of your workflow will repay automation tenfold; some of it is cheaper left alone. Eleven years of analysing businesses means you get that map honestly, not a pitch where the answer to everything is more AI.

Live, not a demo reel

Try the evidence yourself.

Two agents are on the internet right now, free, with public source: Doorknock researches a company before a salesperson rings and files it in HubSpot; Rain Check answers whether Thursday's concrete pour survives the weather. A third piece, a full platform build with its own test suite, is in a public repository you can compile.

2agents live with public source, wired to real systems
52automated checks on the platform build, runnable by you
6gaps proven in someone else's AI-built platform before customers found them
Doorknock Rain Check The platform build

The discipline

How it stays honest.

  1. A specification, not a wish. What it must do, what it must never do, the edge cases, and what finished looks like. The same document I would hand a human team, because it is the same problem.
  2. Proof, because "done" is a claim. Every build carries checks against things the AI does not control: the original data, an outside service, your own books. That habit has caught AI describing test pipelines that did not exist.
  3. Set against itself. One run builds; a separate run, knowing nothing of the first, tries to break it. That is how real defects get found in hours instead of by your customers.
  4. A human who knows your business signs off. AI does not know that your invoices must survive tax time or that Thursday's pour depends on the weather. I do, because finding those rules is the career.

Fair questions

Before you ask.

Can AI really build production software?
Yes, and it already builds mine: two tools live on the internet and a platform build with 52 automated checks, all public. What AI cannot do is know what your business needs, notice when it is confidently wrong, or prove its own work. That is the part you are hiring a person for.
Why do I need you if AI writes the code?
For the same reason a construction crew needs a builder who can read plans. AI is fast, tireless and literal, and it will build the wrong thing beautifully if nobody defines what right looks like. The value has moved from typing the code to specifying, checking and proving it, which is analyst work, and I have done it for eleven years.
What does it cost against an agency?
You still get a fixed quote in writing before anything starts. What has changed is how much scope fits inside a small business budget now: internal tools that used to be a quarter's project are now a matter of weeks. The quote shows you what that means for your job specifically.
We tried building it with AI ourselves and got stuck.
Now one of the most common jobs I see, and often very saveable. AI-built systems fail in patterns I know well, because I review them: security open beside one working login, tests that cannot fail, documentation for things that do not exist. Start at project rescues.