The key-man test14 minNvidia's founder resigned effective immediately after a week of scandal, leaving the company at the center of the AI economy without the man who built it. The real question now is whether Nvidia's dominance was the product of one person, or a machine that can run without him.Read now →

Alex Hormozi Froze His Team's AI Budget. His Reason Should Worry Anyone With a Token Bill
Alex Hormozi's company spends 16 times more on AI per employee than the average tech-forward firm, yet revenue stayed flat. So he capped the budget. Here's how his team actually uses AI, the $350,000 automation mistake he keeps seeing, his test for whether AI is paying off, and the one job he says never to hand to a machine.
On September 30, Alex Hormozi sent his team a memo with a blunt opening line: requests for more AI budget "aren't coming."
That's a surprising sentence from someone who, earlier this year, told his audience that learning AI should be their number one priority. And their number two. And their number ten.
The reason was a single uncomfortable comparison.
Hormozi's company, Acquisition.com, was spending 16 times more on AI tokens per employee than the average tech-forward company, based on spend data from Ramp covering 70,000 businesses. And yet, in his words, company revenue was flat relative to the increase in AI usage.
More AI. Same money.
His diagnosis applies to almost every business paying an AI bill right now. Here's what he said, how his team really uses AI, and the test he wants everyone to run.
The memo, in plain English
Hormozi published the memo publicly. Its core points:
- No more budget increases. The company already spends far above average.
- Unlimited budgets create unlimited waste. Tokens should go only to projects with a clear revenue or cost-saving payoff.
- The data suggests a skill gap. Either people are using AI on projects that don't matter, or they're using it on things that do matter and then spending the saved time on things that don't.
- Training, not money. The team gets lessons on what's worth working on, how to use fewer tokens for the same output (he calls it "AI hygiene"), and how to use saved time well.
And the postscript, which is solid career advice for anyone on a team: the way to get more tokens is to show how little you can use and how much you can get from it.
He is not anti-AI. Far from it
It would be easy to read the memo as a backlash. It isn't.
Hormozi has said AI takes up about half of what he thinks about for his business. The memo is about aim, not abandonment. Look at how his team uses it.
Five agents, 120,000 support tickets
For a book launch, the team spun up five AI agents to handle customer support. Most tickets were simple, people asking where to get their free materials. By Hormozi's account, the agents handled somewhere around 120,000 tickets and resolved about 90% without any human involvement.
That's AI at its best: high volume, repetitive questions, clear right answers.
Triple the media, same headcount
He has said his media team's output roughly tripled in a month without adding staff, because prompts now handle 30 to 50% of the workload. One person does the work of three. His expectation: soon five, then ten.
Break jobs into workflows, not hires
His framework works for almost any role. Take video editing.
- Old thinking: "I need to hire an editor."
- New thinking: "What are the five things an editor actually does to create a video? Each one should be a workflow."
For his short clips, an AI workflow reads the transcript, finds the highest-tension moments, starts the clip right there, and strips out the ums and dead air.
Writing: assist, don't automate
Hormozi is emphatic that AI isn't writing his posts. Tweets used to take him about 90 minutes a day. Now it's about 90 minutes a week, roughly seven times more efficient. But the AI generates ideas he iterates on. It doesn't write the final words.
His content ideas come from AI too. He records and transcribes his meetings, then has AI go through them alongside his calendar to pull the most interesting real moments from his week. Instead of inventing stories, he talks about things that actually happened.
AI in sales: a useful scorecard
Sales is where you can track Hormozi's predictions over time, because he's made them out loud.
- January 2025: he said an AI appointment setter was about three months away.
- April: he said the company had an AI sales development rep (SDR) matching its human team.
- More recently: he got more specific. The team is still hiring human SDRs because humans are still better than the best AI on the phone. But one AI text agent sells around 200 units a month on its own.
So AI wins over text. On voice, humans still win, even after about two years of work on voice AI.
If you build AI systems for clients, that's a useful real-world benchmark: text-based selling is ready; phone selling, at least at this level, isn't yet.
The $350,000 mistake
For every business using AI well, Hormozi says he sees many more using it badly. His favorite cautionary tale:
A company had 11 virtual assistants doing data-cleaning work, at roughly $11,000 a month total. They did a good job. Everything worked.
The company then spent $350,000 building an AI system to replace them.
Two things went wrong:
- The math. At $11,000 a month, the assistants cost about $132,000 a year. The AI build cost nearly three years of their pay before saving a cent.
- The target. Data cleaning wasn't what limited the business. Every business has one thing capping growth at any moment, usually leads, sales or delivery. The money went somewhere else.
His fix is a question to ask before automating anything: should this task exist at all? In his words, it's more efficient to decide something isn't a priority than to automate something that isn't a priority.
The test: views, leads, money
The day before the memo, Hormozi posted the logic behind it. Paraphrased:
If you're using AI all day and you're not making more money, you have the same problem you had before AI. You're doing stuff that doesn't matter. Except now it's costing you tokens.
He's also pointed out the obvious thing vendors don't advertise: AI companies bill you whether what you built was useful or not. The meter runs regardless of outcome.
So here's the three-question audit his approach implies:
- Since I started using AI, am I getting more views, more leads or more money? (Translate to your function: tickets resolved, qualified candidates, days to close.)
- If the bill is rising and those numbers aren't, what is my real constraint?
- Should this task exist at all?
If your token bill keeps climbing and none of the three numbers move, the problem isn't AI. It's aim.
So, does he fire people because of AI?
This is the question everyone wants answered.
Looking across his public comments, there's no clear instance of Hormozi saying he let someone go because of AI. What he does say is that his company, despite heavy AI use across ads and content, is still hiring. He points out that the AI labs themselves, OpenAI and Anthropic, employ thousands of people.
But he's also direct about what he'd tell other owners: raise the bar for the whole company. People who meet the new bar stay. People who don't, don't. He acknowledges that's harsh.
He's given that advice on stage. When the owner of a $20 million website company told him AI was killing his business, Hormozi's prescription was to spend six months rebuilding workflows and cut headcount in half with AI, which would roughly double the margin.
"Bring your own agent"
For employees, his advice is to get there first.
He predicts the job market of the future will be BYOS: bring your own software, or bring your own agent. You train agents on what you do, then go to a business and say, "Instead of hiring a department, hire me. I bring the system."
The goal: replace yourself before someone else does.
He's shared an example. A woman named Tina was running cohort programs through Telegram, and her team kept answering the same questions over and over. Hormozi told her to vibe-code an agent that searches past answers and suggests one, so the team only checks it and hits send. A couple of months later, she reported back: she'd moved her team out of Telegram entirely onto an AI dashboard and, inspired by the first win, built nine other tools for her business.
His starter routine: block four hours a week. In each block, automate or amplify one specific task you do. You get faster every time.
The one job never to give AI
There's one thing Hormozi says not to delegate: big decisions.
Every so often, he tests the models with an obvious decision, asking the same question to several leading chatbots. He says the answers are "all over the place." His conclusion: if you delegate your decision-making to AI, you get dumber.
His rule: don't outsource your thinking, only the typing. A closely related version that circulates among AI builders: you can outsource your thinking, but never your understanding.
What AI can't copy: risk and reality
If judgment stays human, what else does? Hormozi's answer is risk.
He's argued that in a world of infinite AI labor and intelligence, the last thing humans will get paid to do is take risk. Someone has to own the decision. Someone pays for the tokens, owns the company and takes the loss.
He illustrates it with entertainment. If a famous creator's video promised $5 million and a Lamborghini, and none of it was real, there would be no stakes, and no one would care. Humans want stakes.
The same applies to content. If two people post the same six tips for scaling a sales team, and one asked a chatbot while the other actually built ten sales teams, the second wins every time. His phrase: reality is the moat. Your brand and reputation can only be built in reality.
His war on AI slop
He has zero patience for AI-written content and has called out its tells, especially forced negation ("it's not X, it's Y") and choppy staccato sentences. His fix if you write with AI: give it about 12 rules it can never break and 16 samples of your own writing. He says that makes the output roughly five times better.
Don't sell the AI. Sell the outcome
He's equally blunt about AI businesses. Building a company around a tool you made is probably a bad idea, he says. Advertise what the customer wants. Customers still want to lose weight, get results fast and have it be easy. Nobody cares that you have an AI that analyzes transcripts. Just say you'll help fix their problem.
If superintelligence is a few years away
Asked what he'd do if superintelligence really were only a few years off, Hormozi gave a two-part answer. As an entrepreneur: keep building trust and distribution. Personally: think hard about what it means to be human and why you're here.
Try it: the audit on a small business
To make the framework concrete, here's how it might look for a hypothetical five-person marketing agency. The numbers are illustrative, not from Hormozi.
Step 1: list every AI expense. Say the agency pays for five chatbot seats, a coding assistant, an image generator and API usage for a client-reporting bot. Total: around $900 a month and rising.
Step 2: map each to a number. The reporting bot saves about ten hours a month of manual work. The image tool helped land two new clients through faster mockups. The coding assistant built an internal tool nobody uses.
Step 3: name the constraint. The agency isn't short on delivery capacity. It's short on leads. Proposals go out slowly, and follow-ups get forgotten.
Step 4: reallocate. Cancel the unused tool. Point the saved budget and hours at a lead workflow: an agent that drafts tailored proposals from discovery-call transcripts and reminds the team to follow up.
Step 5: re-measure in 30 days. If proposals sent and replies received go up, the spend earned its keep. If not, cut again.
That's the whole method. It isn't glamorous. It's exactly what Hormozi's memo asks his own team to do: connect every token to a result, and stop paying for activity that only looks like progress.
Why this matters
Hormozi's memo is a public version of a conversation happening inside thousands of companies. AI spending is climbing fast. Measurable returns often aren't. MIT's 2025 "GenAI Divide" research made the same point at scale, finding that most organizations it studied saw no measurable profit-and-loss impact from generative AI pilots.
The gap usually isn't the technology. It's that AI makes it cheap to do more work, including more of the wrong work. Hormozi's insight is that AI amplifies whatever you point it at. Point it at busywork and you get expensive busywork.
Practical takeaways
- Name your constraint first. Leads, sales or delivery. Point AI there before anywhere else.
- Run the three-number audit monthly: views, leads, money (or your function's equivalent) against your AI bill.
- Do the build-versus-keep math. Compare automation cost to years of current cost before building.
- Delete before you automate. Ask whether the task should exist.
- Practice AI hygiene. Shorter prompts, smaller models where possible, fewer retries.
- Keep decisions human. Use AI for options and drafts; own the call.
- Build your own agent stack for your role, four hours a week.
Counterpoints and risks
- Flat revenue has many causes. A quarter of flat sales doesn't prove AI spend was wasted; benefits like faster output may lag or show up as avoided hires.
- Survivor stories. The 120,000-ticket and tripled-output figures are self-reported.
- "Raise the bar" can mean layoffs. Advice to cut headcount in half to double margins has real human costs.
- Spend caps can backfire. Starving experimentation may stop teams from finding the next big win.
- Model disagreement isn't permanent. Today's inconsistent answers on decisions may improve, though accountability will still sit with people.
FAQ
Why did Alex Hormozi freeze his AI budget? His company was spending 16 times more on AI per employee than average tech-forward firms while revenue stayed flat relative to AI usage, which he took as a sign people were using AI on work that didn't matter.
Is Hormozi against AI? No. He uses AI heavily for support, content and sales, and says it occupies about half his business thinking. His point is to aim AI at real constraints.
What is the $350,000 AI mistake? A company spent $350,000 to replace 11 virtual assistants costing about $11,000 a month, on a task that wasn't limiting its growth.
Does AI beat humans at sales? Over text, his team has an AI agent selling around 200 units a month. On the phone, he says humans still win.
What should never be handed to AI? Big decisions. His rule is to outsource the typing, not the thinking.
The bottom line
Hormozi didn't freeze his AI budget because AI failed. He froze it because his team was succeeding at the wrong things, faster.
Every token bill hides the same question: is this making us more views, more leads or more money? If not, the AI isn't the problem. The aim is.
Use it everywhere. Point it at your one real constraint. Automate your own job before someone else does. Keep the big decisions. And spend the time it saves building proof that no AI can fake.
Sources
- Alex Hormozi's internal AI usage cap memo (post on X, via Nitter mirror)
- How Much Should You Spend on AI? Hormozi's Budget Lesson (Remy)
- Alex Hormozi's AI Strategy: What His Company Actually Does in 2026 (MindStudio)
- MIT report: 95% of corporate generative AI pilots fail to deliver returns (Computing)





