When Everyone Has a Butler: Gooaye EP691 on What AI Actually Changed

The one who knows where the roof leaks is the one standing under the eaves; the one who knows where governance fails is the one out in the fields.

—— Wang Chong, Lunheng, “Shujie” chapter (Eastern Han; translated by the author)

Written nearly two thousand years ago, and still true: some information can only be obtained by the person standing in the place. No amount of reading from a distance substitutes for it.

Gooaye EP691 (26 August 2026) splits into a family trip to Hokkaido and a long, unusually candid stocktake of what two years of using AI has actually done to the host’s work. What follows is my own reading and extension of it, not a reconstruction of the episode.

What the episode is about

Two threads.

The first is the trip. Before leaving, his father warned him not to go to Hokkaido because “you’ll run into a bear.” He assumed this meant an actual bear. It turned out his father’s last Hokkaido trip was in July, when the market fell every single day. This time the index just chopped sideways, so he declared the superstition broken.

The funnier part is the service. The hotel emailed ahead repeatedly to ask how many men and how many women were in each room and which room the children would sleep in. He replied that they’d sort it out themselves; the staff kept asking anyway. At breakfast a staff member knocks to escort you to the dining room. He said they’d walk over themselves, opened the door, felt relieved to find nobody outside — then turned the corner and found a man standing at the end of the corridor, hands crossed, bowing deeply. Leaving the key at the desk, he mentioned the door wasn’t locked; the reply was “Please lock the door,” and back he went. He called the whole thing creepy beyond belief, while admitting the rooms, the food and the grounds were flawless — especially gardens deliberately built to look like untouched forest. You can tell they were arranged. They pretend they weren’t.

The second thread is AI: how he researches now, how he places orders, and a conclusion that sounds deflating. Two years in, AI has genuinely made him a stronger operator, and he still can’t tell you how much it helped.

Running through both is his eldest son, who starts his final kindergarten year soon. Only a handful of summers left, so this trip he deliberately sat next to him instead of hiding in the back of the van watching stocks — followed by the honest admission that by day four he was close to breaking, and would probably be back in the rear seat tomorrow.

The main points

1. Three expectations about AI, all wrong. Will it replace me? Anyone whose income depends directly on their own output can feel that judgment is still worth something. Can I use it to destroy my competitors? Everyone uses it; the few who don’t were already obsolete — and now there’s a new group who never worked in your field and can cross into it with the same tools. Will it make life easier? No, and arguably the opposite.

2. The time you saved didn’t become your time. Be honest with yourself, he says: you saved time on specific tasks, but your total hours went up. Nobody got more golf out of this. Salaried workers are more exposed — the report that took two hours takes five minutes now, so you can pocket the difference, except your boss also knows that generating reports got easy, and you may be the first person reviewed. He compares it to the arrival of the computer: a phenomenally powerful machine appears, society changes, and the work continues, longer than before.

3. Doing it yourself is sometimes still faster, and that isn’t nostalgia. For breaking news and unfolding events, social media — X in particular — beats asking a model, and the model burns tokens along the way. For a fixed number in an annual report whose location he already knows, Ctrl+F takes a second; handing it to a model means clarifying the request back and forth for several minutes. The criterion isn’t which tool is more sophisticated. It’s which one is faster.

4. Financial statements are where AI is strongest right now. He trades thematic names, which means constantly touching industries he doesn’t know well — and many accounting lines only mean something in context. Contract assets and contract liabilities point at different realities in different sectors; a lengthening collection cycle is a default warning in one industry and a non-issue in another because of who the customers are. That used to mean digging through filings by hand. Now you often don’t even need to supply the framework; the model knows how the sector’s statements should be read.

5. You still pull the trigger, and AI has never lost money. What’s moving intraday, which group is unusually strong — he insists on watching that with his own eyes rather than delegating it, and says this may be one of the main sources of his edge, while admitting it might just be a blind spot. His reason is blunt: AI has never been savaged by the market, so its instinct there isn’t there yet. He estimates three to five out of ten of his trades are improvised: he spots something intraday, immediately pulls data, queries the model, checks whether anyone on social media is discussing it, confirms his understanding and his numbers are ahead of the crowd, and buys.

6. The most honest line: some of this may be placebo. His metaphor is excellent — everyone now walks into the exam hall with a Jarvis. The butler sharpens your pencils and hands you the cheat sheet, and without one you can’t even enter the hall; but every answer inside is still yours. He used to buy first and research after. Now the homework fits into ten minutes intraday, and he places orders with far more confidence. Yet the market also seems to have become more efficient: things that historically shouldn’t move that much now spike all at once, so even with an informational edge you can outsmart yourself and take the damage on the pullback. By the end, he says, he feels slightly sheepish about the whole thing.

One more piece, from the Q&A: answering a question about quitting smoking, he argues that what smokers enjoy isn’t the cigarette but the deep breath they’re forced outdoors to take, because people rarely get to breathe deeply otherwise. He never connects that to the AI discussion. Put side by side, it’s the most useful thing in the episode — more on that below.

Going further

”I use AI for research now. Why hasn’t my performance improved?”

This is the question the episode lands on hardest. You feed in the filings, you get the sector framework, you run three valuation scenarios. The process is vastly more rigorous than a year ago. Then you open your statement and you’re roughly in line with the index.

The episode answers it plainly: once everyone has a tool, it stops being an advantage and becomes an entry fee. That’s not new in markets — real-time quotes were an edge once, so were financial databases, and both are entry fees today. What’s different is the speed. You’ve barely noticed yourself getting stronger before someone who never covered your sector has caught up using the same tools.

The more interesting question is where the edge went. If the research output is commoditised, three things are left to separate people: what you notice earlier than others, when you’re willing to act, and how much volatility your approach forces you to carry. The episode focuses on the first, watching price and volume with human eyes. But the second and third were changed by AI too — because when you feel more certain, you naturally size up.

There’s a side effect worth watching. He observes that things now “spike before they should.” If that’s real, an informational edge is worth less, not more: you were right, but the price finished reacting before you finished buying. That’s testable on your own records. Take your last few trades where you felt early, and measure how far price had already travelled between noticing and filling.

”I save time every day. Why am I more tired?”

The section on working hours is the most broadly applicable part of the episode, because it isn’t really about investing.

Saved time doesn’t turn into leisure because the saving gets reinvested. When the unit cost of a task falls, you don’t do the same amount and go home — you do more. You can cover more names, more industries, more detail, so you do. His phrasing: you can hunt more, you can uncover more of how the world actually works, and whether that converts into returns is a large question mark.

In investing this grows into something concrete. As research volume rises, the number of names that look decent rises with it. Positions spread out, decisions come more often, and the depth behind each one gets diluted. Same ten-hour day: a year ago you knew three companies cold, now you know twelve of them about seventy percent. Your statement never labels the cause. You just feel vaguely unfocused.

For employees there’s a second layer. He points out that the boss isn’t stupid either: the fact that report generation got easy is known to both of you simultaneously. So the question isn’t how much time you saved. It’s whether everyone else saves the same time on the same tasks — because whatever they do save is the least valuable part of your job.

”I researched it thoroughly. Why did I still lose?”

The admission at the end is rare. Markets are full of uncertainty and random events; you can understand an industry and an entry point completely, and if the leverage is large enough it still blows up. Take the leverage off and, stretched over time, you may end up roughly tracking a semiconductor index, or beating it slightly.

Underneath that is something rarely said out loud. Research depth improves how often you’re right. What determines survival is how large you are when you’re wrong. Those are separate variables, and AI only touches the first. Worse, improving the first feeds back into the second: the more thoroughly you’ve studied something, the more comfortable you are betting heavily on it. You end up with a higher hit rate and a higher cost per miss. That combination isn’t automatically better than what you had.

Which is why the smoking answer belongs here. Smokers think the pleasure is the cigarette; he argues it’s the deep breath you’re forced outdoors to take. The attribution is wrong, which is why quitting is hard — you think the thing to remove is nicotine, when the thing to keep is walking outside.

Investment attribution fails identically. You made more this year than last, and you credit “I started using AI.” Three things may have happened at once: the index rose, your sector caught a bid, and you were carrying more risk than last year. Isolating AI’s contribution doesn’t require more research. It requires a record: why you bought, what you expected, what actually happened. Without it you attribute by feel, and feel always hands the credit to whatever impressed you most recently.

He says as much himself: the next thing he wants to work on is turning “feel” into something repeatable, because he doesn’t know whether the last decade or so of instinct-driven results was skill or simply the high point of his life. That sentence is worth more than any conclusion in the episode.

Worth a look

• Gooaye EP691 (2026-08-26), on any podcast platform
• On why efficiency gains don’t reduce total consumption: Jevons paradox — nineteenth-century coal, unchanged logic
• On tools erasing the edge they created: any history of financial information — real-time quotes, electronic order entry, financial databases. Same script three times
• Two titles mentioned in the episode: the series The Bear, and the film Frequency

帶得走的一件事

Once everyone has a tool, it stops being an advantage and becomes an entry fee. What separates people is where they put the effort it freed up — specifically, on the thing nobody can do for them.

The weight is on the second half. Most people pour the freed effort back into the same activity, just more of it: three companies became twelve, one report became four. That’s paying the entry fee harder. It doesn’t create distance.

Something to do today: take one thing you’ve fully handed off to a tool or another person, do it yourself end to end, and write down three details you could only have learned by doing it.

It doesn’t have to involve investing. A dish you always order in, a route you always navigate, a tax return someone else files, a phone call you’ve replaced with messages. Then write the three details — not impressions, details: how many seconds after the boil before things go in the pot, that the third intersection is nearly impossible to turn left at, that column four of the form means something other than what you assumed, the half-second the other person paused on the phone.

If you can list three, there was more you didn’t know than you thought — the kind of information only available under the eaves. If you can’t, that task is safe to delegate forever, and now you know where not to spend your effort.

Raw Markdown

This article is an educational discussion of investment method. It is not advice to buy or sell any individual security, offers no target prices, and does not analyze any current holding. Investing carries risk; make your own decisions or consult a qualified professional.