Hey Reader,

Welcome to the 24th edition of the 3-2-1 (check out previous issues here).

Last edition I argued that judgment gets built through reps. This week I go one level deeper and try to answer what judgment actually is. A new HBR article gave me a starting taxonomy, AI helped me stress-test it, and now I need your judgment to finish the job. This edition ends with a chance to share your perspective, so read to the end!

Let's get into it.

---

3 Things for Work (in L&D)

How Do Workers Develop Good Judgment in the AI Era? (HBR)

• The article that inspired this edition's 1 Idea. David S. Duncan names five distinct forms of judgment and shows why AI creates a judgment gap: experienced people can tell when AI output is sound, junior people can't, and AI is absorbing the work juniors used to learn on.

• 🏋 Effort ≈ 5 min read

The Tragedy of the Cognitive Commons (Human Resource Development Review)

• Every profession shares this but no single company owns the Cognitive Commons: the pool of deeply skilled professionals with the tacit knowledge and judgment to work effectively with AI and validate its outputs. That pool regenerates through entry-level jobs and years of supervised, complex work, and AI threatens the regeneration in two ways: it eliminates junior roles, and it makes it easy for the juniors who remain to outsource thinking to AI, skipping the cognitive struggle that builds deep competence. His ask: stop reskilling firm by firm and start stewarding the commons together.

• 🏋 Effort ≈ 30 min read (academic paper; the abstract alone earns the click)

Don't Be a Meat Proxy (Niklas Gruhn)

• To bottom-line in three words: don't be a meat proxy, a human conduit who copy-pastes AI output to other humans without reading it, validating it, or owning it. Gruhn's rule for messages, code reviews, and feedback: if you didn't engage with it enough to say it in your own words, you haven't added anything, and you're spending someone else's time to prove it.

• 🏋 Effort ≈ 3 min read

---

2 Things for Life

A thread of voice-to-text app recommendations (X)

• Probably one of the biggest quality of life improvements I've made: dictating my thoughts into my computer or my phone and having them lightly formatted as text in whatever app I'm using. My friend KP's thread collects a bunch of tool recommendations so you can pick one and start talking.

• 🏋 ~2 min scroll

What to Make of a Life by Jim Collins

• I've almost finished this, on recommendations from two different friends. It's a great study in how we all go through fog, we all fall off cliffs, and no life is a straight line up. I particularly like his concept of simplex stepping as a way of navigating the fog.

• 🏋 A book, but a fast one

---

1 Idea from Me

AI helped me generate eight new forms of judgment. Now I need your judgment to choose four.

I've been thinking about judgment for months now.

First, I wrote about curiosity as a way to suspend judgment. Then I argued that AI shifts the cognitive load from production to judgment. Most recently, I described performance as judgment-in-context.

But I never tried to define it.

Then I found the work of David S. Duncan, a former McKinsey consultant and disciple of Clayton Christensen. In the HBR article linked above, he names five forms:

1. Evaluative judgment: Is this good or bad?
2. Contextual judgment: Does the usual rule apply here?
3. Tradeoff judgment: Which competing priority matters more?
4. Anticipatory judgment: What happens after we make this choice?
5. Ownership judgment: Should I make this call or escalate it?

It was the most useful definition of judgment I had seen. Duncan's five cover the core of individual decision-making. But judgment in organizations also requires knowing what to ask, when to act, who to involve, and how much is enough.

So I tried to fit it to Seek-Sense-Ship, the model we teach, to identify gaps.

My first pass looked like this:

• Seek: Contextual judgment
• Sense: Evaluative, tradeoff, and anticipatory judgment
• Ship: Ownership judgment

That looked lopsided.

Three forms clustered around making sense of information. Only one helped you gather the right inputs. Only one helped you turn a recommendation into action.

That may be because Sense is where most judgment happens. Or it may be because the model was missing something.

So I set myself a mission: find the gaps and fill them in without turning the model into my garage at the end of the summer.

Of course, I asked AI to help.

But I didn't ask it to "give me more types of judgment." That would have produced a long list of clever-sounding synonyms. I asked it to find functions the original five did not perform, test each candidate for overlap, and create a more balanced model across Seek, Sense, and Ship.

We ended up with eight candidates:

#	Candidate	What it means
1	Inquiry judgment	Knowing which questions will move thinking forward
2	Timing judgment	Knowing when to act and when to wait
3	Scope judgment	Knowing how deep and wide to go, and when enough is enough
4	Stakeholder judgment	Knowing who needs to be involved and what matters to each of them
5	Synthesis judgment	Knowing how to combine conflicting ideas into a coherent point of view
6	Ethical judgment	Knowing what is right when values or laws conflict
7	Delegation judgment	Knowing who, or what, should do the work, with how much autonomy
8	Pattern-recognition judgment	Knowing what kind of situation you are actually in

I don't think all eight survive.

Scope may be a form of tradeoff judgment. Delegation may be the other side of ownership.

I'm also not convinced these are all new.

People needed inquiry, timing, and stakeholder judgment long before AI arrived. What AI changes is we now have to decide what to delegate to a machine, which outputs look right, and when to override a confident recommendation.

AI can weaken judgment when it removes the struggle that develops it. But it can also strengthen judgment when we use it to open the problem instead of close it.

Across Seek-Sense-Ship, that looks like:

• Seek: Ask AI to surface missing questions, dissenting views, and analogous situations.
• Sense: Ask it to attack your categories, expose overlaps, and find counterexamples.
• Ship: Ask it to model how different people may receive the idea and challenge whether the timing is right.

That is how I used it here. AI expanded the possibilities, but it didn't make the final call.

That part belongs to me and you.

My goal is to add four forms to Duncan's five, creating a working model of nine.

Which four would you keep?

Reply with four numbers.

---

That's it for this week - enjoy your Sunday!

I'll be back in two weeks ✌️

Andrew

P.S. If you want to see another demo of how I use AI to think better, check out the playback of my Maven session from last week.