IDITS DADUseful things for whatever you're working on.

Course

Building AI agents that actually work

Six agents, built three ways — the concept, the no-code version, and the code. Written from the ones I run inside my own business, including the parts that break.

The problem

Most agent advice stops at the demo

The demo always works. You give it a goal, it thinks out loud, it produces something impressive. Then you point it at your actual work and it wanders, invents a number, or stops halfway and calls it done.

That gap is not a model problem. It is a missing done-condition, too many tools, and instructions written like a prompt instead of a job description. This course is about closing that gap, six times, on jobs a one-person business actually has.

Three tracks

Pick your depth, or read all three

Track A — Concept

What an agent is, the four parts, the stopping rule, and how to tell when a prompt or an automation is the better answer.

Track B — No-code

Every one of the six agents built in a no-code assistant builder: instructions, tools, done-conditions, and the failures to expect.

Track C — Code

The same agents in TypeScript: tool calling, loop control, structured output, step caps, logging and cost control.

The builds

Six agents

  • Research agent — turns an open question into a decision brief with prices, trade-offs and sources.
  • Triage agent — classifies incoming messages and drafts the obvious replies without ever sending.
  • Content line — idea to outline to draft to edit, each step with its own standard so it keeps your voice.
  • Extraction agent — mixed documents in, one clean spreadsheet row out, with a count check so nothing is skipped.
  • Follow-up agent — finds the follow-ups that quietly never happen and writes the specific message for each.
  • Operator agent — the multi-step one: several tools, chosen order, structured result, capped and logged.

What you get

Inside

  • Six agent builds, each covered at all three depths.
  • The instruction blocks I use, in full, ready to paste.
  • A done-condition worksheet so every agent you build is checkable.
  • The failure catalogue: what breaks in week two and the fix for each.
  • Working code for the operator agent, including tool definitions and step limits.
  • A cost-control section: where the money actually goes in an agent loop.

Best for

Solo operators and small teams who want two or three agents running on real work — not a tour of every framework.

Not for

Anyone looking for an autonomous business. Everything here keeps a human on anything that sends, spends or deletes.

Straight answer

I am one person building this business in public. This course teaches the agent setups I run here, with their limits stated. It will not make you money on its own and I make no earnings claims — it will get working agents into your week.

Questions

Before you buy

Do I need to be able to code?
No. Tracks A and B require none. Track C is there for when you outgrow the no-code version, and it is written so you can read it before you can write it.
Which tools does it lock me into?
None. The no-code track is written against the three things any assistant builder must have — persistent instructions, tools, and a destination — so it survives you switching platforms.
Is this theory or builds?
Builds. Every section ends with something running. The concept track exists because skipping it is why most people's first agent does not work.
Can I see some of it first?
Yes. Three lessons are published free and in full: what an agent is, one complete no-code build, and the six patterns. They are linked below.
What is the refund policy?
Covered by the standard refund policy on this site. If it is not what I described, ask and you get your money back.