# The Things He Didn't Do Are the Record: Notes on The Compound's Take on Tim Cook, Snowflake, and the AI Debate

> Reflections on the 2026-09-01 episode of The Compound and Friends: why decisions not taken are the hardest ones, what the spring software selloff revealed about borrowed conviction, and how to tell a broken method from a normal drawdown. Educational, not investment advice; no tickers recommended, no price targets.

Published: 2026-09-02
Locale: en
Tags: investing mindset, AI investing, software stocks, momentum factor, decision discipline

TL;DR: The hardest part of Tim Cook's fifteen years was everything he refused to buy or build; and the spring software selloff was a reminder that most of our conviction is borrowed.


![An executive stands alone at a floor-to-ceiling window in a night office, looking out over the city, an unsigned stack of proposals on the long table behind him](/covers/compound-2026-09-01-tim-cook-the-goat-snowflake-earnings-preview-the-s-cover.png)

> Learning requires stillness; talent requires learning. Without learning there is no way to broaden talent, and without resolve there is no way to complete learning.
>
> — Zhuge Liang, *Admonition to My Son* (Three Kingdoms, 227 CE; my own translation)

## What This Episode Was About

The September 1, 2026 episode of The Compound and Friends (the *What Are Your Thoughts* format) had Josh Brown and Michael Batnick covering four things: a preview of Snowflake's earnings, their own postmortem on the Ed Zitron AI debate they'd hosted, how to grade Tim Cook now that he's stepping back from the CEO role, and the software rebound after the spring "SaaS apocalypse."

What stayed with me wasn't any single stock. It was that across three unrelated topics, they kept bumping into the same thing: **we are much worse at knowing when we're certain than we think we are.**

## The Main Points

**Half of Tim Cook's record is what he didn't do.** Josh spent a long stretch making the bull case, and the numbers are what they are — roughly 2,700% total return over fifteen years, about four trillion in market cap added, share count down more than 40%. But he pivoted to the line I thought was the best of the episode: sometimes a person should be judged on the things they *didn't* do. Then came the list. No social network (in 2013 the conventional wisdom was that you were nobody without one). No film studio, no streaming platform, no film library. No car division — Project Titan came close. No data center capex chase. No in-house frontier model program. No disastrous M&A. Sitting on roughly $300 billion in cash for fifteen years without doing a deal is, on its own, a discipline story.

**The bear tapped into real anger — but "this is the end state" is a separate claim.** Their postmortem on the Zitron debate was refreshingly honest: he came prepared, he had the figures and the citations, he's very good at the format, and he connected with something real — **a lot of people are sick of being told by folks on the West Coast that their jobs are finished.** Josh's read is that Zitron didn't manufacture that anger; he's riding it. Where they part ways is the claim that what exists today is the finished product. Josh's own timeline: Prodigy in 1995, AOL around '97, buying books on Amazon in '98 — all of it clunky. The first moment the internet genuinely floored him was watching SNL's "Lazy Sunday" on YouTube in 2005. Judging a technology's endgame two years in is his core disagreement.

**They actually answered "what would prove the bear wrong."** Most bull-bear arguments never get here. Their answer was concrete: if OpenAI and Anthropic lay out a credible path to GAAP profitability — even one that takes eight quarters, provided they visibly close the gap quarter by quarter — then the argument that the whole edifice rests on the spending plans of two unprofitable private companies stops working. They also conceded that his views on Microsoft's product quality and on Meta **may never get resolved either way**. Splitting someone's argument into the parts time will adjudicate and the parts it won't is a useful habit on its own.

**Costs coming down is not demand coming down.** They cited an Uber engineering post from a few days earlier: from February to August 2026, weekly active users of agentic tooling across the company grew about 7x and weekly agentic requests about 9.4x, while **total AI spend has been roughly flat since April** — cost per thousand model requests down about 34% from peak, cost per session down about 52%. That led into why "harnesses" have suddenly become hot: you describe the project, the layer picks the model or models that make economic sense. Not every company is big enough to have engineers making those calls in-house.

**The spring SaaS selloff was the market being wrong, collectively.** Michael put up a chart showing that through the entire drawdown, **earnings estimates for these companies never flinched.** The market wasn't reacting to deteriorating numbers; it simply didn't believe the numbers were sustainable. It was wrong, and the snapback was violent. He also named a mechanism that gets overlooked: a lot of that selling was mechanical. Money leaves an ETF, the authorized participant sells the constituents to track the index, and nobody in that chain is looking at any company's growth rate.

**Momentum just had one of its worst 51-day stretches on record — right after one of its best.** Josh's takeaway was blunt: just when you think you've found the key, they change the lock. No style works forever. The shape of the game is losing, losing, losing, punctuated by stretches of extreme winning. If you can't live with that shape, this isn't for you.

**What earnings prep actually looks like.** The Snowflake segment taught me a checklist rather than a conclusion: know what analysts will ask on the call and what management wants to emphasize; note the gap between sell-side consensus and the company's own guidance (here the street was well above guidance, likely because management is thought to sandbag); look at the *distribution* of price targets rather than the highest one — he pointed out the average target sat below the current price, meaning the analyst group was not uniformly enthusiastic; and find the metric that contradicts the story, in this case a net revenue retention that's healthy but not accelerating. That's the bear's question, and it's worth knowing it before the call, not after.

## Going Further

### "The news says my industry is about to get eaten by AI. What do I do with what I own?"

This is the question people are actually sitting with at 1am, and the episode has something usable.

Michael described a day in March when a firm launched an AI tax tool for advisors and Schwab dropped 7%, eventually sliding from $105 to $85. His reaction at the time: **I am an advisor. I know for a fact this reasoning is wrong.** That's Gell-Mann amnesia — you read an article about your own field, you can see it's badly wrong, and then you turn the page and believe the next one completely. What he did was extend the "I know this one is wrong" outward: if this article is wrong, the other pages are probably wrong at a similar rate.

The sequel is the interesting part. Months later Vanguard acquired that same company for $4.6 billion — a far more real competitive threat — and Schwab fell 2.6%. The *severe* version of the same story barely moved the stock.

Here's how I hold it: the judgment you've built inside your own field is not licensed to rule on other industries, but it *is* licensed to calibrate **how much skepticism to carry**. At minimum you now know the market prices "new tool kills old company" narratives far faster than it verifies them. So when something you own drops 20% on a technical headline you don't fully understand, the first question is narrow: did the **earnings estimates** move, or only the **confidence**? For that batch of software names in the spring, the estimates hadn't budged. Keeping those two things apart is what stops you selling at the worst possible spot.

### "My process was working and now it keeps being wrong"

The momentum chart is what I sat with longest. The best stretch on record, immediately followed by one of the worst, with no transition offered.

My own version is smaller but the same shape: an approach that's felt right for months suddenly goes against you several times in a row, and you start wondering whether you got dumber. The episode doesn't resolve that, but it offers a way to tell the two cases apart — **a broken method and a method in a normal drawdown look nearly identical; the difference is whether you wrote down in advance what being wrong would look like.**

If you already knew this approach strings together losses when market style rotates, then a string of losses is a bill, not a signal. If you never considered that it could be wrong, every loss reads as "it's broken" — and that's exactly when people swap methods at the worst moment.

Michael reviewed two of his own bad calls on air (a home improvement retailer, a hotel name) in a completely level voice: I stopped out, I took a 4% loss, that's fine. That calm is harder to learn than any analytical skill, and the reason he has it is that **he decided beforehand what he'd do when wrong.**

### "The good news is already out — am I chasing again?"

Josh described a trap I recognize: people watch roulette land on black five times and can't wait to bet red; the same people see a stock up ten points into earnings and say "I missed it, it's priced in." He thinks these are the same mistake — **using the shape of the price as the reason.** It can go up ten and then go up twenty.

But there's a detail that's easy to miss. In the same breath, he described how he handles the risk: he isn't sitting there hoping. He's trailing a stop. If the report is apocalyptic, he's out, and it won't bother him.

The way I read those two halves together: the first says *don't let price shape be your reason*, and the second says *you've only earned the right to ignore price shape if you already have a plan for being wrong.* Without the second half, the first half is just encouragement to chase. I don't think you can take one without the other.

## Worth a Look

- The Compound and Friends / What Are Your Thoughts, September 1, 2026 episode (YouTube, Spotify, Apple)
- Uber's late-August 2026 engineering post on agentic adoption and AI spend — the charts say more than the text
- Michael Crichton's original talk introducing Gell-Mann amnesia; it's two pages and easy to find
- Apple's own annual reports, if you want to see how the revenue mix shifted from hardware toward services over fifteen years

## The One Thing to Take Away

One idea stayed with me: **the things you didn't do are also your record, and nobody else is keeping that score for you.**

Things you did leave traces — numbers, credit, blame. Things you didn't do leave no receipt at all. The trend you didn't chase, the project you turned down, the message you drafted and deleted to keep the peace — none of it shows up anywhere. So we almost never score it, and after a while we assume we just got lucky.

Here's something I've tried, if you want to try it too. Take a sheet of paper and write down three things this year you **almost did and didn't**. It could be a job change, an investment you badly wanted in on, a message you decided not to send. Next to each one, write a single line: the reason you didn't. Then go through them one at a time and ask: **does that reason still hold up today?**

Some reasons will still stand — that's your discipline, and it belongs on your ledger. Some will turn out to have been fear wearing an argument's clothes — that's the thing to work on next time. Being able to tell those apart is worth more than ten additional decisions.