Skip to content
Dx
04Applied pack

Building & Agents

5 modules · 315 minutes of real work · after Core Judgment

04 — Building & Agents

Status: Optional · take this after Core Judgment · do not take it first

Purpose

You will build, integrate, or maintain AI systems — prototypes, integrations, tools, agents. This pack is the engineering side of the program. It assumes you already know why you're building (Core Judgment) and focuses on how to build something that survives contact with real users, real data, and real failure.

The modules

  1. Solution Contract Before Any Build — most AI builds start with a tool or a demo; the contract is the correction
  2. Isolation and Least Privilege — broad access is convenient until the day the system does something you did not expect
  3. One Job Per Agent — narrow agents with deliberate handoffs beat one large autonomous system
  4. Observability, Drift, and Maintenance — a system that cannot be inspected will fail in ways you do not notice until the damage is done
  5. Verification Before "Done" — "it worked when I tried it" is not evidence

Who this is for

People who ship. Engineers, automation builders, technical leads — and non-engineers who need to understand enough to commission, review, and own a build.

What you will be able to decide after this

  • Which parts of a system should be deterministic code and which parts genuinely need a model.
  • How much autonomy an agent deserves — and where the guardrails go.
  • What the system must prove before it earns another user.
  • When a build is worth maintaining and when it's time to retire it with dignity.

Rules of this pack

  1. Every module produces a working artifact or a decision document with a verification requirement. No slides.
  2. Core Judgment is assumed. If you haven't done it, go do it.
  3. 45–90 minutes per module. The builds stay small on purpose — the point is the judgment, not the repository.

The checkpoint

You are getting value from this pack when you can:

  • Write a precise solution contract before building
  • Limit access and irreversible actions by design
  • Keep agents narrow and composable
  • Make behavior observable and failure detectable
  • Refuse to call something "done" without real verification

If you are still starting from a tool or a demo and trying to add safety later, stay here and do the contract and isolation work first.

The modules

Solution Contract Before Any Build

~75 min

Artifact: A one-page solution contract for a real, bounded use case — leaving no major ambiguity about what the system may and may not do

Verify: Have someone who will live with the consequences read the contract. If they can find a way the system could take an action you did not intend, tighten the prohibitions and permissions

Isolation and Least Privilege

~60 min

Artifact: A permission list for the Module 1 use case — every tool, system, and data source the solution would need, the minimum permission required for each, how access would be revoked, and how irreversible actions are gated

Verify: Walk the permission list with someone technical enough to spot over-scoping. If any access is broader than the job requires, reduce it

One Job Per Agent

~60 min

Artifact: The Module 1 solution redesigned as one or more narrow agents — each with a single job, allowed actions, prohibited actions, and the human interface

Verify: Ask: "If this agent starts doing something unexpected, can I explain exactly what it was supposed to be doing and shut only that capability down?" If the answer is no, narrow it further

Observability, Drift, and Maintenance

~60 min

Artifact: A short observability and maintenance plan for the solution — what will be logged, what signals indicate trouble, who watches the signals, how often the system is reviewed, and how changes are tested and released

Verify: Simulate a quiet failure (for example, the agent starts returning plausible but incorrect results on a subset of cases). Confirm that your plan would surface it before major damage occurs. If not, strengthen the signals and ownership

Verification Before "Done"

~60 min

Artifact: A verification record for the solution — test cases run (including at least two failure or edge cases), results against the original acceptance criteria, any gaps that remain, and an explicit decision: ready for limited use / needs changes / not ready

Verify: Have someone who did not build the system review the verification record and try to break it. If they succeed in a way the tests did not catch, add the case and fix the gap

The gate

Done with the modules? The checkpoint decides whether you leave this pack. No badge — the right to continue.

Building & Agents checkpoint →