Applied Building

How to design a working AI system as a set of decisions made in order: the job, how much freedom to give, what it needs to know, what it can reach and do, when to split the work and where a person stays in the loop, written down as a short design that any builder could carry out.

1. 01What building means hereWhat a design is: a short, ordered set of decisions about the job, freedom, knowledge, reach, splits and checkpoints. Why it is written before any tool is chosen, and why earlier decisions constrain later ones.about 7 minutes
2. 02Start from the job, not the toolStating the job: what comes in, what goes out, who receives it, what "done" means and what is out of scope. Why starting from a capability produces a system looking for a job.about 7 minutes
3. 03How much freedom to giveChoosing between fixed steps and an agent choosing its own steps, and the points in between, and what the job's variety and the cost of a mistake say about that choice.about 8 minutes
4. 04What the system needs to knowStanding instructions versus material the system must look up: what can go stale, and who keeps it current.about 7 minutes
5. 05What it can reach and doListing the access the job needs, separating reading from acting, and deciding which actions the system takes and which it drafts for a person.about 8 minutes
6. 06When to split the workWhen one system is enough and when to split the work into parts: what each part receives, and what each handoff costs.about 7 minutes
7. 07Where a person stays in the loopWhich outputs a person sees before they leave and which get reviewed afterward, what the system does when it is unsure, and who it escalates to.about 8 minutes
8. 08Write it down, try it smallWriting the one-page design from the earlier decisions, running it once, and noting what the design missed.about 8 minutes

Opening a lesson needs membership. During the preview it's free for invited members. See plans