Falling in love with the hammer

There's a conversation that keeps repeating in every meeting, every LinkedIn feed, every AI sales call in 2026.

—"I've got 20 agents running my company in parallel."

—"I downloaded 150 free agents from GitHub, configured them, and they're already operating."

—"I built a whole organization: a CEO agent, a CMO agent, a battalion of executors. Working 24/7."

You listen, you nod, and for a second you think: that's impressive.

Then you realize that in the entire conversation, nobody mentioned the customer. Or the business. Or the problem being solved.

The world filled up with hammers

We're at a moment where building agents is trivial. There are marketplaces, open-source frameworks, ready-to-use templates. What used to require an engineering team can now be assembled on a Saturday afternoon with a couple of prompts.

And that — which is genuinely extraordinary — brought a side effect that few people are naming out loud: the conversation stopped being about value and became about the toolbox.

The other day I heard someone explain, with legitimate pride:

"I've got an agent that reads my calendar and emails every morning, gives me a summary, I trained it to auto-reply to operational emails, and I have a dashboard showing me everything in real time."

It sounded flawless. I asked two questions:

—How many meetings do you have per day?

—Three, four at most.

—How many emails per day actually need your attention?

—Twenty, thirty.

—And how much value, concretely, does it add that the agent replies to your emails?

Silence.

That's the problem right there. The tool ate the value-generation process. The agent exists. It works. It's elegant. But the question of what it's actually for — in business terms, recovered time, faster decisions, money — went unanswered.

The uncomfortable question

Let me say something that has quietly become taboo over the last few months:

Does everything that can be automated have to be automated?

Because that's the logic settling in. If it can be done, it gets done. If an agent is available, it gets connected. If the framework allows it, it gets deployed. The silent decision criterion is technical availability, not business impact.

But a company's success has never depended on having the best hammer. It has depended on knowing which nail matters.

Do I need to buy every tool today, or as my business actually requires it?

The most important question a founder, CEO, or operations leader can ask right now isn't "what AI tools should I buy?". It's:

• Which process, if I improve it, changes my business?
• How will I know it improved?
• Who owns this still working six months from now?

If those three questions don't have clear answers, having 20 agents running in parallel isn't digital transformation. It's expensive decoration.

The bias nobody talks about

There's a cognitive bias I see everywhere: people confuse sophistication with results. If the solution is complex, it must be working. If it involves many components, it must be adding value. If it has a beautiful dashboard, it must be moving the business.

Not necessarily.

Sometimes the best agent is the one you don't build. Sometimes the right automation covers a boring, repetitive, glamour-free process that nobody wants to show off at a conference — but that saves four real hours a week to someone whose time costs money.

Sometimes it's not about how many agents you have. It's about whether the single agent you put into production is moving a metric the business owner can defend in front of the board.

The toolbox is dazzling the buyer

Something else is happening, and it's worth naming: the toolbox is dazzling the customer.

The demos are stunning. The marketplaces are infinite. Every week brings a new framework, a more capable agent, a more elegant pattern. And the customer, understandably, gets excited. Wants everything. Wants it now.

The problem is that fascination with the hammer is not a strategy. It's a very well-designed distraction.

Well-implemented AI is transformational. That's not up for debate. What's up for debate is what "well-implemented" actually means.

It doesn't mean having the newest stack. It doesn't mean having downloaded all 150 GitHub agents. It doesn't mean having built a parallel AI organization operating in the shadows.

It means having identified a process that matters, having agreed on how success is measured before starting, having implemented something that sustains itself over time, and having left someone inside the organization in charge of keeping it running.

That's boring to post on LinkedIn. But it's the only thing that moves the needle.

The invitation

If you're evaluating an AI initiative in your company — internal or outsourced — ask three questions before approving anything:

1. What specific process are we tackling, and why this one?
2. What's the metric that will change, and when will we know?
3. What happens to this six months from now if the person who implemented it leaves?

If all three have answers, you're doing AI. If they don't, you're buying hammers.

And hammers, no matter how impressive, don't build houses on their own.

---

I see too many people in love with the hammer instead of looking at the customer. And that, more than a technical problem, is a problem of focus.

---