Your product has a new user. It’s not human.

Should MCP be in your ICP?

Everyone (I think?) agrees that defining your ICP (Ideal Customer Profile) is important.

It’s the thing that makes the whole company go around - marketing and sales build their entire personality around it, founders try to change it every week, and engineers nod while secretly thinking ‘this feels made up.’

Define your customer → build for them → grow. Easy. In theory.

But there’s an assumption baked into all of this: Your user is human. I think that assumption is breaking.

As agents begin to interact with products on our behalf - often via protocols like Model Context Protocol (MCP) - your ‘user’ may never actually touch your product.

Which changes pretty much everything.

In the past, even API-first products had a clear answer. Your ICP was still a human, usually a developer. Someone chose your product, integrated it, and used it. But now, LLMs will definitely be involved in every step of that process, and may even replace it. This is the question I think every company should be asking:

What happens when your main ICP is an MCP?

---

Elena’s Picks

Hey, I’m trying out a new format! I’ll be specifically featuring products and services I personally recommend to this community. Plus, sometimes a few that I’m exploring.

→ Stripe: The GOAT for payment processing. They also crush it with billing, invoicing, and pretty much the whole monetization stack. They’re an essential part of our growth work at Lovable, and I use them to manage payments for all of my solopreneurship stuff, too. If you’re considering alternatives… don’t. Just use Stripe.
🎁 Offer: Playbook for scaling AI product pricing. How are we supposed to price AI products, exactly? Stripe put together the 5-step framework you need.

→ Lovable: What can I say? Yes, I work here. But… I work here because I was obsessed with the product, first. It’s the best way to build an app or site, even if you have no technical experience. Try it out!
🎁 Offer: Get a year of Lovable for free! (For paid subscribers to Growth Scoop)

→ Factors.ai: Here’s one I’m exploring: The problem they’re going after? Running ABM campaigns manually and across a bunch of different platforms. Factors.ai has an agent that pulls together visitor identification, intent data, ABM activation, and multi-touch attribution. How much better could all this be if it’s all in one platform?
🎁 Offer: LinkedIn Benchmarks report. The Factors.ai team analyzed 100+ b2b companies and they’re giving away the insights for free.

---

Good products have always been about understanding the quirks of how humans actually operate and interact with things.

We built onboarding flows to reduce friction, polished UI to feel intuitive, added empty states and tooltips and delightful microcopy so users feel guided and successful. Entire growth teams exist to improve the conversion between steps in a flow that assumes a human is clicking through it.

But what happens when none of that is seen?

Agents don’t care about onboarding. They don’t care about your navigation. They don’t care about user psychology. They don’t care that you spent six weeks debating whether the CTA should say ‘Get Started’ or ‘Try Now.’

They care about whether your product returns the right output, quickly, reliably, and in a format they can use.

If it does, you stay in consideration. If it doesn’t, you will get replaced. Not churned. Replaced.

You have to open up

If agents are going to sit between users and software, then your product needs to be accessible to them. Which is where the MCPs come in.

If your product isn’t exposed through MCP (or something equivalent), agents can’t use it. And if agents can’t use it, you’re not part of the workflow anymore.

This means that every B2B company will be pressured to open up. And might be forced to, even if they don’t want to. Which means some uncomfortable tradeoffs.

Opening up your product means sharing functionality, data, and workflows in a structured, machine-consumable way. You are effectively saying: ‘Here is what we do. Here is how to use it. Go ahead.’

Which is great for accessibility. And terrible for any moat that depended on friction, UI complexity, or user habituation.

A lot of companies are going to discover that what they thought was defensibility… wasn’t

If your product required training, onboarding, or repeated human interaction to be valuable, that created a kind of stickiness. Some companies have gotten confused and thought the stickiness was based on their product being superior. But it wasn’t. It was just too painful to switch.

Agents remove that pain.

They don’t need to ‘learn’ your product. They don’t care how it feels. They just evaluate whether your output is better than the alternative.

Which leads to a pretty brutal question: If an agent can reproduce your core functionality without ever touching your product, what exactly are you defending?

In that world, your UI is not a moat. Your ease of use is not a moat. Your brand is not a moat. Even your workflow might not be a moat.

Your moat becomes whatever cannot be easily replicated at the output layer.

For some companies, that’s proprietary data. For others, it’s unique aggregation, network effects, or deep integration into systems of record.

For many, it’s… unclear.

This affects distribution, too

This shift in ICP affects your core product flows, but please also consider what Growth looks like when agents are a core part of the buying process, too. A lot of what we call ‘Growth’ today is really distribution and conversion optimization for human users. Ads, traffic, funnels, activation rates, onboarding completion.

With this in mind, you start designing differently. With agents and MCPs as part of the mix, distribution is almost reversed - you’re not taking it out to them, you’re making it easier for them to come get it from you. Are you part of the set of tools an agent can call? Are your endpoints discoverable, documented, and trusted?

You stop asking: ‘How do we get users into our product?’

And start asking: ‘How does our product fit into their workflow?’

You start prioritizing:

• clean, well-documented interfaces

• predictable, structured outputs

• reliability

Do you even need a traditional landing page anymore? Clean, thorough documentation may do the trick.

Does ranking on Search matter? Now we’re not just talking about ‘People ask ChatGPT instead of Google’ - we’re talking about people not asking at all, because an agent is going out to find the answers for them. And that agent is wayyy more thorough. They’re not going to stop at the first few options. There’s no ‘below the fold’ for them. They’re on a relentless mission to evaluate every option and then choose based purely on performance.

Once you start thinking through this, you’ll realize that some of the things you used to invest in may not matter as much in this new layer. Non-human users just have totally different priorities.

This doesn’t mean human experience goes away

So, invisible, mechanical replacements for every product?? Nah. But I do think this will split the market.

A good analogy is what happened with ghost kitchens during Covid. Restaurants that went all-in on delivery optimized everything for that model. Speed, packaging, menu design, pricing. They didn’t need ambiance or location or even a particularly strong brand outside of the delivery apps. They survived and thrived during that time while restaurants focused on the in-person dining experience got destroyed.

I think we will see something very similar with software.

Some products will double down on being genuinely enjoyable to use. These are your “minimum lovable product” companies. They win because people choose them, advocate for them, and build affinity with them.

Others will become almost entirely invisible. They exist as infrastructure. As a codified set of rules that is hard to reproduce. They are never opened directly, never explored, never ‘used’ in the traditional sense. They are just… there, powering outcomes. And you know what, I think most of the B2B will fall here. Gasp.

Why?

Because most people don’t actually enjoy using business software. Most of B2B product experiences… suck. They use it because they have to. If an agent can take over those interactions and deliver the same or better outcomes, most users will happily step out of the loop.

B2C will likely lag, because there are parts of consumer behavior that are tied to taste, identity, and exploration. But even there, pieces of the experience will get abstracted over time.

What now?

Every company has some decisions to make. In order to evaluate your product and the value you can provide, agents and MCPs need to have the full picture - your structure, systems, processes. Most companies don’t want to share these elements… but if you don’t, you may be replaced by an alternative that does. And even more than that: If all of that process doesn’t actually lead to a better output, we’re not just talking about a threat to your efficient distribution. We’re talking about a threat to your company’s existence.

So start thinking about this. The shift hasn’t fully played out yet, but the direction is pretty clear. If people don’t want to use your software or product, they’re going to ask their agents to do it for them. And if another upstart company can provide the same value to those non-human ‘users’ at a fraction of the cost? The LLMs may not be great at math, but that’s an equation I’m pretty sure they can solve.

So yes, you probably still need a clear human ICP. But if you’re not also thinking about how an agent evaluates, selects, and uses your product… you’re optimizing for a user that may not be in the loop much longer.

---

Edited by Jonathan Yagel.

---