Most important take away

AI doesn’t have to replace your whole job to put it on thin ice — it just has to hollow out enough tasks that, when the next budget freeze or reorg hits, the company asks “why is this role bundled this way?” The travel-agent pattern is the right mental model: the visible break came years after the technology arrived, when downturns forced the industry to admit what had already changed. The fix is to audit your actual last 10 days of work into four buckets — Theater, Commodity, on the Line, Durable — and start migrating your hours toward Durable before the org reorgs you.

Summary

The thesis: Performance review systems still measure visible output, but the economics underneath have already shifted. The dangerous window is when the work still looks fine, your manager is happy, and the org hasn’t caught up — but less and less of that work needs you.

Data the host cites:

• OpenAI/UPenn: ~80% of US workers could have ≥10% of tasks affected by LLMs; ~1 in 5 could see half their tasks affected.
• Anthropic Economic Index: ~49% of jobs have already had at least a quarter of tasks performed using Claude.
• Microsoft research on 200K Bing Copilot conversations: most common work humans bring to AI is gathering information and writing; most common AI output is writing, teaching, providing information, and advising.

The four-bucket audit (TCLD) — actionable insights:

Open your last 10 business days of calendar, sent email, Slack DMs, docs, tickets, code commits. Tag every individual item (not the role, not the project) with one of four letters:

• T (Theater): Work that exists because the org performs it, not because it produces examined value. Status meetings where nothing changed. Decks nobody reads. Reviews that exist for political cover. The hardest tag to use honestly because it admits you wasted time. AI absorbs theater first because adequate-quality theater is what the theater already was.

• C (Commodity): Real work that produces real value but doesn’t need you specifically. Summarizing, routing, applying known rules, writing first drafts in known shapes. The test: could you write a spec and have someone else produce a roughly equivalent output? Not an insult — companies run on it — but it’s where AI compresses fastest.

• L (On the line): The uncomfortable middle. Pattern recognition with structured patterns. Editorial work in established formats. Synthesis across familiar inputs. Work where a strong junior could do 70%, the last 30% feels yours, but you’d struggle to articulate exactly what judgment you applied.

• D (Durable): Work where the output depends on something you can’t fully describe in advance. You changed the question more than you answered it. You read the room. You saw that the stated problem wasn’t the real problem. Your presence visibly changed the outcome beyond competence.

What the audit will reveal (predictions):

• Theater is bigger than you want it to be. People conflate “professionally expected” with “valuable.”
• Commodity is bigger than you want it to be — and this hurts because that skill took years to build. But markets don’t care how hard a skill was to learn.
• Durable is smaller than you want it to be. Your self-image is built around it, but how many hours did you actually spend there?
• Theater + Commodity = the fraction of your week on thin ice.

What durable work actually looks like — career-defining insights:

1. Durable work is mostly question-holding, not question-answering. Most orgs reward answers. AI absorbs answers fastest because the question is given and the output can be judged against the prompt. Durable work starts before the question is set: knowing when “we’re asking the wrong question” is the move.

2. Different work compounds differently. Theater compounds to nothing. Commodity work compounds to the organization (the system captures the pattern and you become replaceable). Durable work compounds to you — calibration, scar tissue, intuition that can’t be transferred.

3. The legibility paradox. Durable work must be legible enough that the system values it, but not so legible that the system can run it without you. Show outcomes (“I got us to ask a different question and we changed the plan”); don’t expose mechanism (“here’s my decision tree”). Once specified, work can be delegated and commoditized.

4. Identity is the real obstacle. The mechanics of the audit are easy. The hard part is what the tags do to the story you tell yourself about your value. Identity updates are psychologically expensive, but the advantage goes to people who update self-image before the org forces it on them.

Six concrete next moves (career advice):

1. Stop performing the theater you can stop without consequences. Cancel the recurring report nobody reads. Skip the meeting where you’re the third senior person. Send three sentences instead of the full update. Mostly nothing happens — that’s the point.

2. Don’t pour recovered time into more commodity work. This is the trap: AI lets you write faster, so you write more updates. You become twice as productive at work whose value is collapsing.

3. Pour recovered time into durable skill development. Take cases that don’t fit known patterns. Choose projects where the framing is unclear, not just the execution. Sit in conversations where you’re not the expert and have to read the room.

4. Build a private track record of durability calls. Each week, write down one call you made where the outcome depended on judgment you can’t fully reduce to rules — context, decision, result/date. After a year you have ~50 entries; after three years a real portfolio you can use to renegotiate role and routine load.

5. Use that record to refuse commodity work that doesn’t fit your trajectory. Don’t declare “I don’t do routine work anymore.” First become visibly valuable on non-routine work, then renegotiate via project selection.

6. Be honest if the audit says you’re in the wrong role. If most of your week is Theater + Commodity and L is drifting commodity-ward with no path to Durable inside this role, the answer is moving. When evaluating a new role, ask current incumbents what they spent last week on, where the ambiguous questions are, what calls they made that couldn’t have been made by a process.

How to use AI to do the audit: Codex with computer use in a browser (across email, calendar, Slack), with chunking — too many objects to one-shot. Build prompts with explicit definitions of T/C/L/D and your own preferences. The host has guides on his Substack.

Closing distinction: Two paths. (a) Use AI to do more of the work whose value is collapsing — feels productive in the short term. (b) Use AI to do more of the work whose value compounds — harder, less legible, may not get immediate credit, but pays off in years.

Stocks/companies named (not investment recommendations): OpenAI (private), Anthropic (private), Microsoft (MSFT) — Bing Copilot research, Expedia (EXPE) — travel-agent disruption analogy.

Chapter Summaries

1. Cold open: thin ice means a full calendar with no clue what’s happening — The travel-agent pattern: tech changes economics first, downturns force admission later. Don’t be surprised.

2. Hollowing out happens piece by piece — Most people overestimate “will AI replace my job?” and underestimate “how much of my last two weeks still needed me?”

3. Why orgs lag the change — Performance systems measure visible output and old metrics. There’s a dangerous window where reviews say “you’re fine” while the role economics have already shifted.

4. The audit (TCLD) — full setup — Open last 10 business days. Tag every item, not the role.

5. T — Theater — Status meetings, decks no one reads, reviews for political cover. AI absorbs this first because the theater was already operating below the threshold of human attention.

6. C — Commodity — Real work, real value, doesn’t need you specifically. Companies run on it; just don’t build your career on it.

7. L — On the Line — The uncomfortable middle where most career anxiety lives. Some becomes commodity, some becomes durable.

8. D — Durable — Work that depends on context, taste, courage, and pattern recognition that can’t be specified in advance.

9. What the audit reveals — predictions — T is bigger than you want; C is bigger than you want; D is smaller than you want.

10. Durable work = question-holding, not question-answering — The skill of saying “I think we’re asking the wrong question” is the most defensible skill.

11. How work compounds — Theater → nothing. Commodity → organization (you’re replaceable). Durable → you (calibration, intuition, scar tissue).

12. The legibility paradox — Make outcomes legible; keep the mechanism partially hidden.

13. Identity as the real obstacle — Tagging tasks T or C threatens self-image. Update before the org does.

14. Six concrete next moves — Stop theater, don’t reinvest into more commodity, develop durable skills, build a private track record, use it to refuse commodity work, leave if the role can’t support D growth.

15. Closing call — The differentiator is no longer “do you use AI tools.” It’s “what do you invest the time AI gives you back into.”