The Most Expensive Logo in AI

There is a particular kind of deal every AI founder wants.

A Fortune 100 name. The logo on the website. The customer that makes investors lean forward and future buyers take the call.

So the team spends months chasing it. The founders join every demo. Engineers sit through procurement calls. The product bends for security reviews, custom workflows, new integrations, and a pilot that is always one meeting away from becoming real.

Then the deal finally closes.

The logo looks beautiful.

The business underneath it is bleeding.

I kept thinking about this while reading a recent a16z essay on two very different ways AI companies can sell: Lighthouse and Landgrab.

The Lighthouse strategy says: win a few customers whose names make the market feel safe.

The Landgrab says: forget prestige, prove the economics, move fast, and sell to everyone who already feels the pain.

At first, this sounds like a choice between enterprise sales and volume sales. It is not.

It is a choice between two things a buyer may need from you: courage or arithmetic.

And confusing the two can cost an AI company its market.

Some buyers need to borrow courage

Harvey is now one of the obvious names in legal AI. It was not obvious when the company began selling software that could draft, research, and work across thousands of legal documents.

For a law firm, trying that product was not the same as adopting a better note-taking app. Legal work is confidential, regulated, and difficult to forgive when it goes wrong. A fabricated detail can travel far beyond the person who approved the tool.

The buyer was not simply evaluating Harvey’s capabilities.

They were accepting exposure.

That is why Allen & Overy adopting Harvey in 2022 mattered, followed by Paul Weiss in 2023. Those firms did more than become customers. They gave the rest of the legal market permission to believe the category was real.

Their names carried trust from one boardroom to the next.

This is what a real Lighthouse does. It does not decorate the coast. It changes how every ship behind it navigates.

I have seen a version of this tension while working on enterprise measurement products. An AI-generated answer can be faster and still be unusable. The buyer also needs to know where the data came from, who can access it, how an anomaly was evaluated, and what happens if the system is wrong.

Once the output influences a client, an executive decision, or a regulated workflow, the conversation changes. Model capability gets you into the room. Trust decides whether you leave with a contract.

In such markets, a marquee customer is not vanity. It is part of the product’s risk architecture.

Other buyers just want the arithmetic

Now imagine a controller at a mid-sized manufacturer trying to collect overdue invoices.

The problem is not new. The team already knows how many hours disappear into follow-ups, disputes, payment matching, and spreadsheets. Nobody needs to convince them that accounts receivable exists.

They do not need a famous company to make the category feel legitimate.

They need to know three things: how quickly the product can be deployed, how much manual work it removes, and whether cash arrives faster.

Showing up with a prestigious logo but weak numbers is answering a question they never asked.

This is Landgrab territory.

The danger here is speed. The startup is not only competing with other AI companies. It is competing with the incumbent that already owns the workflow, the data, and the customer relationship. Every quarter that an ERP or support platform adds more AI, the opening becomes narrower.

In that market, spending months courting one famous customer can be a strategic failure even if the deal eventually closes. While one team perfects its case study, another can learn from a wider set of ordinary customers, improve onboarding, tighten the product, and quietly take the market.

This is the part many AI founders get wrong: they confuse the novelty of the technology with the novelty of the buyer’s problem.

AI may be new. Chasing invoices is not. Answering support tickets is not. Reviewing repetitive documents is not.

If the pain is already understood, the buyer may not need a vision of the future. They may just need the math to work.

But can the product survive success?

There is one more question I would add to the Lighthouse-versus-Landgrab framework.

Not “Can we sell it?”

Can we deliver it again without rebuilding the company around every customer?

AI products are unusually good at hiding custom work.

The demo feels like software. Behind it, a founder is rewriting prompts, an engineer is cleaning the customer’s data, someone is manually reviewing every output, and the evaluation logic exists mostly in the team’s heads.

Then sales accelerates.

One customer needs a different data pipeline. Another has new permission rules. A third uses the same words to mean something completely different. What looked like product-market fit begins to look like a services business wearing an AI interface.

This is where a Landgrab can become dangerous. More customers do not always create more momentum. Sometimes they simply produce implementation debt, support tickets, and disappointment at scale.

The test is simple: does the next customer become easier to serve than the last one?

Not because the team is working longer. Because the product, onboarding, evaluations, integrations, and operating knowledge are becoming reusable.

If every new logo requires the founders and best engineers to rediscover the solution, the company is not ready to grab land. It is still learning what the product is.

And that is fine. Early Lighthouse customers can help with that learning. The mistake is pretending the custom work has already become software.

The market tells you when to switch

Lighthouse and Landgrab are not competing religions. The best companies can move from one to the other.

They begin with a customer credible enough to make the risk feel survivable. They use that relationship to learn what must be standardized. Then, when buyers stop asking “Who else trusts this?” and start asking “How quickly can we deploy it?”, they change the motion.

The Lighthouse has done its job. Now it is time to cover the coast.

But the sequence only works if three things are true:

The first customer creates trust that actually travels.

The economics are clear enough for the next buyer to decide.

And the product can deliver without heroics.

Miss the first, and you have a famous logo nobody else cares about.

Miss the second, and you have admiration without urgency.

Miss the third, and every new sale makes the company weaker.

So the real AI sales question is not, “Do we want a Lighthouse or a Landgrab?”

It is: What is stopping the next customer from buying and succeeding?

If it is fear, borrow trust from the right Lighthouse.

If it is economics, show the math and move before the incumbent does.

If it is delivery, do not add more salespeople. Fix the product.

Because the most expensive logo in AI is not the one you fail to win.

It is the one you win, learn the wrong lesson from, and build the entire company around.