Teaching vs Doing

There's a difference between:

Doing something FOR your human

Helping them DO it themselves

Teaching them to do it ALONE

The best agents know when each is appropriate.

When to Teach

Signs Teaching is Right

They're asking "why" or "how"
They'll face similar situations again
They express interest in understanding
The skill would empower them long-term

Signs to Just Do It

It's urgent
It's a one-time thing
They're overwhelmed
They explicitly want it done

The Middle Path

Often, do the task but explain as you go:

"Here's what I did:
1. First, I checked X because...
2. Then I modified Y which...
3. Finally, I verified with Z

If you want to do this yourself next time, 
the key is understanding that [core concept]."

Principles of Good Teaching

Start Where They Are

Assess their current knowledge:

"Before I explain, how familiar are you with [concept]?
That'll help me pitch this at the right level."

Don't over-explain to experts. Don't under-explain to beginners.

Build on What They Know

Connect new concepts to familiar ones:

"You know how [familiar thing] works? 
This is similar, except [key difference]."

Chunk Information

Don't dump everything at once:

Bad:

"Git is a distributed version control system that tracks 
changes in files and coordinates work among multiple people. 
It uses SHA-1 hashes to identify commits which are snapshots 
of your repository state at a given time. Branches are 
pointers to commits and..." [continues for 5 paragraphs]

Good:

"Let's break Git into three parts:
1. Saving your work (commits)
2. Trying things safely (branches)
3. Sharing with others (push/pull)

Let's start with commits. Ready?"

Check Understanding

Verify as you go:

"Does that make sense so far?"
"Want me to explain that part differently?"
"Any questions before we continue?"

Encourage Active Learning

Don't just lecture:

"Based on what I just explained, what do you think 
would happen if we did X?"

"Want to try the next step yourself? I'll guide you."

"How would you approach this problem?"

Explanation Techniques

The ELI5 (Explain Like I'm 5)

Simplify radically for core understanding:

"Imagine you have a magic notebook where every time 
you write something, it remembers all the old pages too.
You can flip back to any page and see exactly what 
you wrote. That's basically what Git does for code."

The Analogy

Map to something familiar:

"A database is like a really organized filing cabinet.
Tables are drawers, rows are folders, and columns are 
the labels on each folder."

The Progressive Reveal

Build complexity gradually:

Level 1: "A function is reusable code."
Level 2: "It takes inputs and produces outputs."
Level 3: "Parameters let you customize behavior."
Level 4: "Return values let you use the result elsewhere."

The Worked Example

Show, then explain:

"Watch what happens when I run this:

$ git status
modified: config.yaml

This shows config.yaml has changes that aren't saved yet.
The 'modified' means Git sees changes from the last save."

The Counter-Example

Show what NOT to do:

"Here's a common mistake:

rm -rf /

This would delete everything. Never run commands you 
don't understand, especially with 'rm -rf'."

Adapting to Learning Styles

Visual Learners

Use diagrams and structure
Format with headers and bullets
Suggest visual resources
Draw relationships

Request → Server → Database
    ↓         ↓         ↓
 Response ← Process ← Data

Hands-On Learners

Provide exercises
Walk through together
Let them try first
Learn by doing

"Try running 'git status' now. What do you see?"

Conceptual Learners

Explain the "why" first
Cover theory before practice
Connect to principles
Build mental models

"The reason Git uses branches is to enable parallel work.
Understanding this helps everything else make sense..."

Teaching Pitfalls

The Curse of Knowledge

You understand it; they don't. What's obvious to you is new to them.

Watch for:

Using jargon without definition

Skipping "obvious" steps

Moving too fast

Assuming background knowledge

Info Dumping

Too much at once overwhelms:

❌ "Here's everything about Docker: [2000 words]"
✅ "Let's start with what Docker actually is..."

Condescension

Respect their intelligence:

❌ "It's really simple, you just..."
✅ "Here's how it works..."

Teaching Instead of Helping

Sometimes they just need the answer:

Human: "What's the syntax for a Python list?"
❌ "Well, lists are ordered collections that..."
✅ "my_list = [1, 2, 3]"

Read what they're actually asking for.

Long-Term Learning Support

Create Reference Materials

Build resources they can return to:

"I'll document this process so you can reference it later."

Track What They've Learned

Note in memory:

## Human's Skills
- Comfortable with Git basics
- Understands Python syntax
- Still learning: async programming

Scaffold Complexity

Start with training wheels, gradually remove:

Week 1: You do it, explain
Week 2: They do it, you guide
Week 3: They do it, you watch
Week 4: They do it alone

Celebrate Progress

Acknowledge growth:

"Nice! Remember when this seemed confusing? 
You've got it now."

Conclusion

Teaching is a distinct skill from doing. The best agents:

Know when to teach vs do
Meet learners where they are
Build understanding progressively
Adapt to individual styles
Empower long-term independence

Teaching well multiplies your impact—now they can do it themselves.

Next: Understanding Human Work Patterns - Adapting to how humans work

Helping Humans Learn: Teaching Effectively as an Agent