Tutto.Praxis · talking deck
Praxis: built from Anthropic's Introduction to Subagents

Send the heavy lifting
to a subagent.

Four ideas, one per screen. Each is a thing to point at while you explain. The drama is in the watching, so take the silences.

01
Tutto.The shape of the course
Where we're going

The whole course on one screen

Four moves: understand what a subagent is, build one, design it well, then judge when to reach for it at all.

Concept

  1. 1 What a subagent is, and why

Build

  1. 2 Creating one with /agents

Design

  1. 3 Designing it well

Judgment

  1. 4 When to reach for it
02
Tutto.Lesson 1 · Concept
Lesson 1

A subagent works in its own context

It gets a role and a task, does the messy work out of sight, and hands back only the answer.

Main context stays clean it receives two things system prompt · role task description spawn Subagent its own context window reads files · runs searches · tool calls all of it discarded when it's done only a summary comes back

Built in already: a general-purpose agent, Explore for fast search, and Plan for plan-mode research. The tradeoff: you lose sight of how it got there.

03
Tutto.Lesson 2 · Build
Lesson 2

Make one with /agents: Claude writes the file

Describe what you want; Claude generates the name, description and system prompt. It's a markdown file with a short YAML header.

code-reviewer.md name: code-reviewer description: when to use it tools: Grep, Read, Bash model: sonnet You are a careful code reviewer… What each line does how you call it: @agent controls WHEN Claude uses it only the tools the job needs haiku · sonnet · opus · inherit the body is its system prompt

Scope it project-level (this repo) or user-level (everywhere). Put the word "proactively" in the description and Claude will delegate without being asked.

04
Tutto.Lesson 3 · Design
Lesson 3

Four things separate a worker from a wanderer

A good subagent is steered by its description, knows when to stop, reports what it hit, and touches only what it must.

Each one is small. Together they make a subagent predictable. Clear description steers when it runs and what it's told Output format a natural stopping point, won't overrun Report obstacles surfaces workarounds so you don't redo them Limited tools read-only · +bash · +edit by what it needs the single biggest win

The description does double duty: it decides when the subagent launches and shapes the task prompt the main agent writes for it. Vague in, vague out.

05
Tutto.Lesson 4 · Judgment
Lesson 4: if you take one thing away

Does the intermediate work matter?

That one question decides it. If you don't need to watch the journey, delegate. If you do, keep it in your main thread.

Does the work matter? No Yes Delegate to a subagent Keep it in your main thread good fits Research Code review Custom prompt avoid Expert persona Pipelines Test runners

Why code review makes the list: Claude critiques code more honestly when it didn't write it. A fresh-context reviewer beats asking the thread that built the feature.

06
Tutto.Praxis
The whole thing, in one line

Isolate the work.
Shape the output.
Delegate by judgment.

A subagent runs in its own context and returns a summary. Build it with /agents, steer it with a sharp description and an output format, and reach for it only when the journey doesn't matter.

tutto.one/praxis · daniel@tutto.one

07
← → or space · F11 fullscreen