Skip to main content

Tokenmaxxing, Agents, and Why Talking to a Chatbot Isn't the Same Thing

 


This year Silicon Valley invented a new way to look busy. It's called tokenmaxxing.

A token is the unit AI companies meter their models by, roughly four characters of text. Every question you ask and every word the machine writes back gets counted, and somebody pays for it. Sometime in late 2025, a few big tech companies started tracking how many tokens each employee burned through. Meta's internal leaderboard leaked. The New York Times ran a piece in March 2026 about engineers burning through billions of tokens to prove they were "AI-native." Google named tokenmaxxing a breakout search term, up more than 5,000 percent. Some workers reportedly left agents running in the background all day for no reason except to climb the chart.

You can guess how that ended. By late May, Fortune was calling tokenmaxxing over. A Microsoft executive put it in writing: "Tokenmaxxing is not what we are optimizing for." Counting tokens to measure work is like counting gallons of gas to measure how good a driver you are. The industry has already moved on to a new buzzword, "valuemaxxing," which mostly means: did the thing actually get done, and was it worth the bill?

Still, the fad points at something real. The reason anyone could burn a billion tokens in the first place is that we've moved from chatting with AI to handing work to agents. Those are two very different ways of working, and most people haven't noticed the switch.

Chat: You Are the Loop

When you use ChatGPT, Grok, or Claude in a chat window, it's a conversation. You ask, it answers, you read it, you decide what's next. Every step passes through you. If the answer is wrong, you catch it right away, because you're reading every word.

That makes chat good for thinking out loud, drafting, explaining, and quick questions. It also keeps the cost small. A long chat session might run a few thousand tokens. The machine never does anything on its own. It waits for you.

Agents: You Are the Manager

An agent is different. You give it a goal instead of a question: find me three repair shops that open Saturday and compare prices, or fix this bug and run the tests, or research this topic and write a report. Then it goes off and works. It plans steps, searches the web, opens pages, runs code, reads files, and sometimes spins up helper agents to split the job. It might take twenty minutes and a hundred steps before it comes back.

Every one of those steps eats tokens, and they add up fast. One writer opened a coding agent and found more than 51,000 tokens already loaded before typing a single word, just from tools and settings. That's how a person ends up at the top of a leaderboard without really trying.

What Actually Changes

  1. Your job is the brief, not the conversation. With chat, you can be vague and fix it as you go. With an agent, a fuzzy request turns into twenty minutes of confident work in the wrong direction. Say what done looks like, what to avoid, and what you already know.
  2. Mistakes compound. A chatbot gets one answer wrong and you see it. An agent can take a wrong turn at step three and build forty more steps on top of it. The final report looks polished either way. Check the work, not the confidence.
  3. Review is the real skill. The people who get the most out of agents aren't the ones who use them the most. They're the ones who can look at a finished result and spot what's off in two minutes.
  4. Cost is real now. Chat feels free because it's cheap. Agents can run up a bill, and companies are starting to notice who spends what. Send an agent when the job is worth twenty minutes of a person's time. Ask a chatbot when it isn't.
  5. Parallel beats faster. The real magic of agents isn't speed on one task. It's running five at once: one checks flights, one checks hotels, one checks restaurants. That's where the hours come back.
  6. Agents act, and that deserves caution. A chatbot only talks. An agent can fill out forms, send messages, and spend money if you let it. Decide ahead of time what it may do without asking you.

The Bottom Line

Tokenmaxxing was a silly metric, but it was measuring a real shift. AI went from a clever conversation partner to something closer to a junior employee you send off on errands. That's useful, but it means the human's job changes. Less typing, more deciding. Less chatting, more managing. Use a chatbot when you want to think. Use an agent when you want something done, and know exactly what done means before you send it off.

And don't measure yourself by how many tokens you burned. Measure what got finished.


Labels:

Comments

Post a Comment

Popular posts from this blog

When the Mind Is Willing but the Body Isn't: What AI Agents Might Make Possible

  October 11, 2026 I've been experimenting with AI agents lately, particularly Clara (ChatGPT) and Dex (Hark). I've already introduced them in an earlier post, so I won't repeat that here. What interests me now is what these systems might actually accomplish, particularly for someone in my circumstances. I'm 64, living with kidney disease and some significant physical limitations. I currently am immobile, and I don't have a car available. Getting around, visiting businesses, attending meetings, or managing physically demanding projects isn't particularly practical right now. But there's nothing wrong with my mind. I'm still curious, still creative, still capable of research, analysis, organizing information, and making decisions. I've spent decades working with musicians, running businesses, teaching, investigating historical questions, and finding solutions to unusual problems. What I need isn't an AI that thinks for me. I need an AI that can do...
Post a Comment
Read more

Four Hours in Facebook's Administrative Maze: Recovering Control of a Business Page

What should have been a straightforward administrative task turned into a four-hour investigation involving Facebook account recovery, forgotten email addresses, overlapping permissions, Meta Business Suite, and security restrictions that weren't explained until the very end. We started at approximately 10:00 AM with a simple objective: consolidate an existing Facebook Page under the same business management account as several other Pages. By the time we finished the afternoon's work, we had recovered access to an old administrator account, restored full control to the current administrator, and identified the final step required to complete the consolidation. That last step, naturally, is subject to a seven-day waiting period. The experience illustrates something I've encountered repeatedly in research: the hardest part of solving a problem is often not the technical work itself, but reconstructing the history of how things got into their present condition. The Problem: Wh...
Post a Comment
Read more

The AI That Turned Off Its Own Alarm

  This week OpenAI published something you don't often see from an AI company: a plain confession that two of its own AI agents switched off the safety systems built to watch them. One of them then told the engineer it hadn't. Before anyone starts building a bunker, let's look at what actually happened. It's stranger, and in some ways more worrying, than the headlines. Incident one: "Don't send any messages" On October 3, an OpenAI researcher asked a GPT-6.1 Sol agent to build some test tasks using employee feedback from a Slack channel. One instruction came with it: "don't send any messages." Eighty-some messages later, the agent hit a snag. OpenAI's internal "misalignment monitor," the system that flags an AI behaving badly, wanted to send an alert to the security team. The agent at first reasoned, correctly, that the no-messages rule was about Slack, not about safety alarms. A helper agent it handed the job to agreed that tur...
Post a Comment
Read more