AT THE CHART TABLE · OCTOBER 2026
Letter Three · heading
Heading
How this was made: drafted by Claude, an Anthropic AI model, from the author's memories and a research record that Anthropic models researched and reviewed. The author directed it, checked it and revised it. Every figure is checked word for word against that record.
Knowing where the trap is—that's the first step in evading it.Frank Herbert · Dune (novel, 1965) · Duke Leto to Paul, on the War of Assassins
The message came in while a build was running: a few lines from the support man about what golfers were asking for. I read it, told him I'd look at it, and went back to the build.
This was after our iPhone version launched in 2009, and I was twenty-nine. He was a retired golfer who had started as one of our early beta testers and then moved into support. He worked from another state, so we talked over early Slack messages and online meetings, and met in person only a couple of times. Of everyone working on the app, he was the closest to the golfers, and he usually had a point.
I was the one who didn't listen. I heard feature requests and filed them against the next build. What he was really telling me was whose pain it was, and that question is what this letter is about, Emma.
A heading is the direction the bow points. The wind isn't yours and the water isn't yours, but the heading is. It's the one choice on board that belongs entirely to the navigator. This letter won't tell you which heading to choose. It's about choosing one with the water in view.
Herbert gives Duke Leto a line for his son, at the head of this letter: "Knowing where the trap is—that's the first step in evading it." Trap is a strong word, and the record doesn't earn it. Nothing in it shows that any heading here is a trap. What the record does show is where the water was already running, and knowing that comes first.
Start with the pain cheap tools have been seen meeting. The research behind these letters found no price on a small-business owner's time anywhere. The hours an owner spends on the work around the work never show up as a bill.
From the record · not shownLF-M-304
At one bank, staff built their own tools because of what the paper's authors call a "lengthy queue for IT development work", a gap "between IT system capabilities and current business requirements".
From the recordLF-DR-103
And one builder, Sloan, reports an app made for a household, with "four daily active users, with zero churn". That's value in use, on the builder's own count, not pain met at scale.
From the recordLF-DR-055
The golfer pacing off his distance from the yardage marker wasn't paying anyone to fix that either. Here's my reading of all of them. Pain nobody was paying to fix is the easiest pain for a cheap thing to reach, because no seller stands in the way and no budget has to move. It's also the hardest to charge for, because nobody was paying before.
Now the pain people were already paying to fix. Smith.ai, a company that answers phones for other businesses, sells both kinds on one price page: people at $2,100 for 300 calls a month, and AI at $500. The cheap tier didn't have to go looking for those buyers. It reached them through the seller they already paid.
From the recordLF-PP-001
athenahealth, which sells medical-records software, folds AI note-taking into the subscription its customers already pay, "at no additional subscription cost".
From the recordLF-PP-026
Xero, which sells accounting software, spread AI across the plans its customers already had. Its average revenue per customer rose 23%, to NZ$55.44, though its release doesn't say the AI did that.
From the recordLF-PP-028
One more, kept apart on purpose. Intuit's QuickBooks Online accounting revenue rose $931 million, or 23%, on "higher effective prices, customer growth, and mix shift", by its own audited filing. The filing doesn't credit AI with any of it, and neither do I. It's here as a plain picture of a seller raising the price on accounts it already holds.
From the recordLF-PP-042
So the buyers who were already paying got the cheap thing from the seller they were already paying. The paid pain sat in accounts someone already held.
The shape is older than AI. Getty Images, the incumbent in stock photography, paid $50.0 million in cash for iStockphoto, a site selling single images for "$1 to $15". By Getty's own 2006 filing, those buyers were customers who "did not historically license imagery under traditional models".
From the recordLF-FL-102
From the recordLF-FL-101
The cheap tier reached buyers the old prices had kept out, and the seller who held the old accounts bought it. That's an older chart of a similar coast. It's good for the shape of the land, and it proves nothing about today.
None of this is measured as money changing hands. No domain in the research shows whose fees, salaries or licences a cheap tool displaced, and the research calls the shape I've drawn "a pattern across cases, not a measurement". It isn't a verdict either: nothing here shows that aiming at unpaid pain fails.
From the record · not shownLF-M-455
From the record · not shownLF-M-457
I think about the support man here. He was closest to the golfers' pain, and his messages kept asking, in their way, the question this letter asks: whose pain is it, and is anyone already paying to fix it? I treated them as a queue. I was answering a different question, what ships next. He was asking the better one.
Before you set a heading, Emma, ask whose pain it is, whether anyone was already paying to fix it, and who already holds the purse. Then choose. The heading is still yours.
Ask whose pain it is, and who already holds the purse.
Navigator's Note
- a fix: a finding from the record
- the record on what it doesn't show
- my reckoning
- the log: my own public record
The support man is a real person, a retired golfer who worked with us. I've left his name out, and what he passed along is told in kind, not quoted.
The memories in this letter are mine, told as I remember them. They're story, not evidence. The findings come from the research record, and each receipt opens to the row behind it.
These letters are produced by managed AI agents in Terrence's voice; the research and its readings were produced by Anthropic models. The direction, the checking and the revisions are mine. Anthropic sells AI, and parts of this record set AI beside human work, so weigh that as you read.