A research agent from Sakana AI now writes a hundred-page report in eight hours. Not an outline. Not a draft for a human to finish. VentureBeat’s write-up of an ultra deep research agent that produces hundred-page reports in eight hours describes a finished document: sourced, structured, the kind of thing a consulting team used to bill three weeks for. The pitch is that ordinary deep research wasn’t deep enough, so here is research that goes further, faster, and never asks for the weekend off.
It is genuinely impressive. It is also the start of a problem that nobody quite wants to name, because the problem doesn’t look like a problem. It looks like a gift. Here is the shape of it, stated plainly before the evidence: making a thing abundant does not make it valuable, it makes it cheap, and it hands all the value to whoever decides which of the abundant things anyone should look at. Every big announcement this week is a bet on production. The scarce asset is turning out to be selection.
Think about what a hundred-page report was for. It was never the pages. The pages were proof of effort, a receipt for the thinking. What you actually wanted was the one paragraph that changed your decision, the sentence buried on page sixty that made the other fifty-nine worth printing. The length was the cost of finding the insight, not the insight itself. We tolerated the volume because someone had to wade through the world to bring back the part that mattered, and the wading was expensive enough that it also functioned as a filter: nobody commissioned a hundred pages on a question they did not care about.
Now the wading is free, and so is the filter that used to come attached to it. When the wading is free, the hundred pages stop being proof of anything. Anyone can generate them. By tonight, ten people in your industry can generate them, on the same question, with slightly different prompts, and hand you ten reports that each claim to be definitive. The scarce thing was never the report. It was the judgment to know which page to act on. That scarcity just got more expensive, not less.
The Same Shape, Five Times
Look at the day and you see the same outline drawn in five different inks. CNBC reported that Salesforce agreed to buy the customer-service platform Fin for $3.6 billion to put more autonomous agents in front of customers. TechCrunch covered Orbio raising $21 million to automate hiring and onboarding for frontline workers, which means the paperwork, the scheduling, and the first-week friction. And SpaceX closed an offering that, once underwriters exercised the greenshoe overallotment, totaled $85.7 billion, a sum that exists because capital believes the machines are about to do an enormous amount of work very cheaply.
Set that alongside the week’s most quoted warning. VentureBeat covered Satya Nadella warning that AI could hollow out entire industries the way global trade once reshaped them, which is the kind of warning a person makes when he can see the wave from the shore and knows he is also the tide. The friction between that caution and those three checks is the most honest thing in the week. The same conviction funds both: that the cost of producing is about to fall through the floor. One party is buying the fall. The other is describing what happens to everyone standing underneath it.
Five stories, one motion. The work that used to take a room full of people, the research, the service calls, the onboarding, the launch, is collapsing into something that runs while you sleep. Everyone is racing to be the one who automates first, because the math looks brutal and obvious: whoever makes the work free wins the work.
Except that isn’t quite the math.
Why Abundance Moves Value Instead of Creating It
Here is the thing the race keeps forgetting. When you make something abundant, you don’t make it valuable. You make it cheap. A platform that lets everyone publish doesn’t make publishing worth more; it makes the act of publishing worth almost nothing, and quietly transfers all the value to whoever decides what gets seen. The producer multiplies. The filter inherits the kingdom.
We are about to live through this at industrial scale. Reports, replies, analyses, onboarding flows: all of it heading toward infinite supply. And infinite supply is not a feature. Past a certain point, more of anything makes the whole pile worth less, because the human on the other end can only read, trust, and act on so much. That ceiling does not move. It is the one input in this entire system that has not improved and will not improve, no matter how much capital arrives, and every projection that treats output as the binding constraint is quietly assuming otherwise.
Drowning someone in answers is not the same as informing them. A thousand pages of correct, sourced, beautifully formatted research that no one has the judgment to use is just a very expensive way to feel busy. The failure mode here is not bad output. That would be easy to catch. The failure mode is excellent output, in volumes that make excellence undetectable, which is a much harder thing to notice from inside.
The Rare Skill Is Subtraction
Which is why the small, human-scale signal is about to be worth more than the ocean, not less. The one number that changes your mind beats the ten thousand numbers that confirm what you already thought. The colleague who reads the room beats the dashboard that reads the data. As the machines flood the world with output, the rare and rising skill is subtraction: knowing what to ignore, what to trust, where the one load-bearing fact is hiding in the hundred pages nobody will finish.
Subtraction is harder to teach and much harder to sell than production, which is part of why so little capital is pointed at it. A tool that generates something can be demonstrated in a minute. A tool that tells you to skip ninety-nine pages has to be trusted before it can be evaluated, and trust is earned slowly, in a way that does not fit a funding cycle. So the market builds what it can demonstrate, and the thing it cannot demonstrate becomes the bottleneck by default.
Notice too that the automation stories are, underneath, selection stories that nobody labeled as such. The customer-service agent decides which requests get a human. The hiring and onboarding system decides which paperwork surfaces and when. Whoever ends up owning those products will discover they bought a filter and paid for a producer.
Where the Value Actually Lands
Nadella’s warning is real, but the frame is slightly off, and the off-ness matters. The industries that feel hollow won’t be hollow because the work disappeared. The work will be everywhere, cheaper than ever, pouring out faster than anyone can absorb it. What will feel scarce, and what will quietly become the whole game, is the judgment to point at the part that counts and let the rest wash past.
That is the bet underneath the $85.7 billion and the $3.6 billion and the $21 million, whether or not the people writing those checks would describe it that way. Not that machines will do the work. That part is already settled. The bet is on who owns the filter when the work becomes free, who sits at the narrow point where the infinite pile meets the finite human who has to decide. The producer is being automated into abundance. The curator is being promoted into power, whether the curator asked for the job or not.
The machines will write the hundred pages. They were always going to. The question that decides everything is older and simpler, and no model can answer it for you: which page do you read first, and which ninety-nine do you have the nerve to skip.
Leave a Reply