New Three-Legged Stool: Humans, Hardware & Tokens

The infamous project-management three-legged stool—we’ll call it the 3LS today, because it sounds fun—has been around for decades. Many a management decision has come down to a simple, brutal reality: you can have two, but you can’t have three.

Is this ringing any bells? Have you been in the meeting arguing over cost, schedule, and performance? Eventually, someone says: ‘You can have it cheap. You can have it fast. Or you can have all the capabilities. Pick two.’

I don’t know about you, but these models frustrated me when I first encountered them. What management-school physics genius created the mathematical formula proving this was true? But after delivering hundreds—if not thousands—of projects, it sure seems to hold up. Pick two.

The new 3LS: Humans, Hardware & Tokens

Over the past few months, I’ve noticed a new three-legged stool emerging. This one defines how organizations squeeze value out of AI: Humans, Hardware, and Tokens—HHT.

• Humans: I sit firmly in the Humans camp. We still need people, and we will for a long time. Humans bring unique perspectives, visions, frustrations, style, grit, judgment, and a whole collection of intangibles that may be mimicked but not replaced. I put enormous value on humans.
• Hardware: AI needs a place to live. That can be in the cloud, locally, or at the edge—and we will need all three. Ironically, choosing among them forces you right back into the old cost, schedule, and performance conversation. Someone should probably build an AI agent to evaluate that for us.
• Tokens: If you’ve been caring for your AI and watching it grow, you know it eats tokens. Models will consume as many tokens as they can get their digital hands on—APIs and MCP servers included. We’ve even invented words like tokenmaxing to describe both their appetite and our attempts to feed them efficiently.

The old stool asks leaders to balance the economics of a project. The new stool asks them to balance the economics of intelligence.

Why now?

For the first time I can remember, nearly every industry is trying to learn the same new capability at once: AI.

If you aren’t already feeling the pressure of the HHT stool, you will. I won’t spend this article cataloging the million ways AI acts as a force multiplier. I’m going to assume you already know your people are asking for more funding, more hardware, and more tokens.

Look at what Humans—the subject-matter experts—are doing right now:

• A friend in cybersecurity research is using AI for automated, intelligent penetration testing. His team is chaining vulnerabilities together to compromise systems and prove that threats are real.
• Another friend is using AI to train and coach teams in real time, keeping workflows moving with something like video-game tooltips for real life.
• A DJ friend with thousands of sound clips is using AI to categorize, tag, and curate his library to create entirely new musical experiences.

These humans are geniuses in their domains. Now they have to choose how to supply the other two legs of the stool.

If you aren’t paying for tokens yet, this might sound silly. But those of us who constantly refresh usage screens to see how much compute remains in the current five-hour block understand it. You may even be following Tibo and Boris on X, cheering whenever the limits reset.

Coding is one of AI’s clearest superpowers right now. Stack Overflow’s 2025 Developer Survey found that 84% of respondents were using or planning to use AI tools in their development process, and 47.1% said they used them daily. AI is helping millions of people codify their systems. But intelligence has a cost. Tokens are like really, really expensive cookies.

This scarcity is one reason increasingly capable open-weight models—from China and elsewhere—are entering the conversation. Organizations will experiment with them to gain more control over cost, capacity, and deployment. The cat is out of the bag. As you become more sophisticated and build swarms of agents, you need places to run them. You need policies. You need to test your logic. And you need to decide how cloud, local, and edge hardware fit together before the architecture starts making those decisions for you.

My Fourth of July wish

America marked its 250th birthday on July 4, 2026. For the occasion, I baked a massive cake for US—get it, the United States and us.

When I blew out the candles, here was my wish: I want to spend $5,000 to perfectly balance the new 3LS for my team. I want to invest in the Humans by giving them dedicated Hardware and enough Tokens to run models that produce high-quality logic and code, 24 hours a day, for an entire year.

I know $5,000 sounds impossibly low for that kind of around-the-clock power today. But that’s the wish. I want to invest in my people and let their agents run without constantly negotiating with limits.

When we finally reach that price point, the real fireworks will begin.

Source note

The AI usage figures above come from the Stack Overflow 2025 Developer Survey, AI section: https://survey.stackoverflow.co/2025/ai/