In the Short Term, AI Is a Complexity Machine, Not an Efficiency Tool
AI lets everyone write and do research, and the result is more emails, more decks, more decisions. Jim VandeHei says that in the short term it manufactures complexity, and only deliberate deletion gets your time back.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Six-month planning is already a fantasy
Jim says he used to mock one-, two-, three-year plans; now even a six-month plan strikes him as laughable — he can only have high confidence in things six to seven weeks out. He gives a concrete number: six months ago, 30% to 40% of Axios traffic came from Google Search; now it has fallen to 5% to 10%, involving millions of users, and it happened almost overnight. He calls this state post-processable velocity: the AI world is moving faster than humans can understand it. This is not abstract anxiety; it is a traffic structure rewritten within a few months.
— Jim VandeHeiMost companies are stuck in the messy middle
Jim calls the current phase the messy middle: companies start using AI, build prototypes, see the promise, and then discover that "I have to connect this to my data pool," which is far harder than expected. He describes a split: people in Silicon Valley feel this is 1999 and everything is about to explode; but most people in Toledo, Topeka, Oshkosh feel "I'm just running my company, AI is probably that new thing, my employees are talking about it, we'll get to it eventually." Introducing any technology into an existing company is hard, because you have to wire it into the clunky systems built over the past decade.
— Jim VandeHeiIn the short term, AI is a complexity machine
Jim says flatly that for most people, in the short term, AI is a complexity creation machine. The mechanism: now anyone can use AI to write something decent, and anyone can do deep research and dump it into an email to their boss. The result is that everyone receives more emails, more decks, more ideas — some clearly thought through, many not. So it's "oh God, now I have more decisions to make." This is exactly why he wrote the book: how to help you delete the things you do out of habit and inertia, so you can amplify what you actually want to do.
— Jim VandeHeiThe 25% wasted each week is habit, not ability
Jim cites data from The Economist and others: on average, 25% of each week is spent on things that shouldn't be done at all, or that have no value to the company, and the cause is habit and inertia. He does the math: 25% means two and a half years wasted out of every decade. He stresses that AI will not solve this right now — "I thought AI would make us more efficient; right now it isn't." In the short term people are anxious and confused, and he thinks you should be anxious and confused, because this is a strange time to be alive.
— Jim VandeHeiFirst confront, then delete, and only then amplify
The framework has three steps: confront is the psychological move, asking "why am I doing this" and "does this meeting really need to happen"; delete is cutting it or saying no; amplify is doing what you're good at, what you enjoy, what brings a promotion or a payoff. Jim says this system has been running at Axios for two or three years, and the effect is that people are more focused and more efficient, but most importantly happier. He gives the example of the "let's kill shit" meeting: gather the leadership team, spend an hour or two listing every stupid meeting, process and product, then kill them. That's how the Axios app got cut — tens of millions of people come to Axios, but only 30,000 use the app regularly, while the maintenance cost is extremely high.
— Jim VandeHeiMake more decisions and your decision-making gets worse
Jim says the average person receives about 120 texts and 150 emails a day, and this mental strain is underestimated. He cites brain-scan research: the more decisions you make, the worse your ability to make decisions, and this is called cognitive fatigue. So unless you reduce the number of decisions you have to make, you won't perform well on the decisions you actually care about. He gives a personal example: he set recurring purchases like toothpaste and deodorant to Amazon auto-subscribe so he never has to think about them again. Add up these small things and you have fewer decisions and more time — enough to read, to have coffee with someone.
— Jim VandeHeiAI won't replace reporters, it will replace generic consultants
Jim says he doesn't think AI will replace reporters, and doesn't think it will replace most jobs. He names one that will be replaced: the generic consultant — AI is already decent at general consulting and business analysis. But for anyone with passion, as long as they can treat this technology as a force multiplier for their specific work and get very fluent with it, they'll be fine, even more in demand. He judges that fluency with AI will soon become table stakes in the workplace. The logic behind the Axios Local expansion is here too: use AI for distribution, copy editing and other back-end work that isn't core to journalism, drive costs down, and then hire hundreds of reporters.
— Jim VandeHeiNever bet your business model on a third party's goodwill
Axios has a content licensing deal with OpenAI. Jim's logic is direct: ChatGPT, Google and Anthropic have already eaten all the public information on the internet, including everything Axios has ever written. So there are only two paths — sue like The New York Times and hope to win, or accept a licensing fee. He judges that Axios can't win a lawsuit, and that its content volume isn't large enough for the damages to be worth it even if it did win. He stresses the difference from BuzzFeed betting on Facebook: this revenue could disappear tomorrow and the company's overall performance wouldn't be affected. His principle: never bet your business model on the generosity of a third party that has no vested interest in your business.
— Jim VandeHeiThe Claude project he built for his wife beats doctors
Jim built a project for his wife in Claude: input all her health information, specify that the AI should only pull from sources like Johns Hopkins and Mayo, have it review new medical literature every two weeks, and cross-reference online with people who have similar symptoms. He says it was better than any doctor he had ever seen, until they finally went to Mayo and met a doctor who could approach the level of that system. His own use is to feed all his blood tests, MRIs and family history into ChatGPT. He admits he has a concierge doctor and takes the results in to check; the doctor says "that's pretty good" every time, and one of his doctors now uses this setup for most of his work.
— Jim VandeHeiIn their own words · checked verbatim
I've always laughed at like year or two or three-year planning. I'm like, how the hell do you do that? But now I kind of laugh at six-month planning.
Jim VandeHei6:07
I think for most people in the short term, it is a complexity creation machine.
Jim VandeHei9:10
Well, think of a 25%, man. That's two and a half years out of every decade that I pissed away doing things I don't want to do or shouldn't be doing.
Jim VandeHei10:16
There's been tons of studies, tons of brain scans that show the more decisions that you make, the worse you get at making decisions. It's called cognitive fatigue, mental fatigue.
Jim VandeHei15:28
don't ever bet your model on the generosity of a third party that has no vested stake in your business.
Jim VandeHei25:47
I have found it to be astronomically better than any doctor I've ever met with.
Jim VandeHei28:51
You need to be that ruthless, constantly thinking about fewer, fewer, less, less, quicker, quicker.
Jim VandeHei31:59
Figures
| Change in Axios's share of traffic from Google Search | from 30%-40% to 5%-10% | 6:07 |
| Average work time wasted each week | 25% | 10:16 |
| Average number of texts received per person per day | about 120 | 15:28 |
| Average number of emails received per person per day | about 150 | 15:28 |
| Probability of AI destroying humanity given by an Anthropic executive | 10% | 4:07 |
| Trump approval rating | about 40% | 1:04 |
Glossary
- post-processable velocity
- A term Jim coined for the speed at which the AI world changes, exceeding humans' ability to understand it.
- messy middle
- The stage where companies start using AI and build prototypes, but find that connecting it to old systems is harder than expected.
- cognitive fatigue
- The more decisions you make, the worse your ability to make decisions.
- force multiplier
- Treating AI as a tool that amplifies your ability to do your specific work.
- table stakes
- Fluency with AI will soon become a basic workplace requirement.
How to listen
Founders and managers who are bringing AI into their organizations, and anyone crushed by the information overload AI brings who wants to find their time again.
The opening section on the midterms and the political toxicity of data centers has little to do with the AI efficiency theme.