Selling to AI Agents

AI tools aren’t just building products. They’re choosing them.

Mar 29, 2026

I’ve spent some time lately playing with AI tools to build small apps and workflows.

It’s easy to be impressed by the obvious things: how quickly code gets written, how cleanly features come together, how little friction there is between idea and output. But what stood out most is how third-party tools are baked into the build process.

These systems aren’t just generating code. They’re making product decisions.

Ask for email confirmations, Resend shows up.

Ask for scraping, Firecrawl is pulled in.

Payments? Obviously Stripe.

I didn’t explicitly ask for any of these tools — nor did I research anything, compare tools, click on an ad, entering into a marketing funnel.

The AI selected the tool and we moved, that was it.

There is a new growth channel if you’re selling B2B software products:

It’s selling to agents.

From discovery to selection

Every major tech shift creates a new growth channel.

Somewhat timely when I started writing this, but I read a post from Andrew Chen on LinkedIn - link here.

He talks about how each wave has had its playbook, examples:

Cloud and SaaS → SEO, content, email.
Mobile and social → feeds, creators, platforms.

Now we’re firmly in the next shift - AI isn’t just a change of interface, but a lot of the time, it’s also the decision-maker. The “user” choosing your product might not be human. It’s an agent—and you’re completely abstracted from the decision.

This new channel isn’t just about showing up in ChatGPT responses. It’s about embedding your product directly into the tools people use—Replit, Loveable, Claude Code. Whatever the interface is, how do you get inside it?

Why certain products get chosen

Let’s assume neutrality—no commercial bias from the AI provider. We’ll focus on what you can control here:

Humans choose based on clarity of value, perceived risk, ease of use, social proof, brand throughout the comparison phase. Agents aren’t as different as you might think. But the way they consume products is—and that’s where things start to shift.

Agents operate through MCPs, not UIs—which means they over-index on developer content: docs, GitHub, changelogs.

Products I see winning agent recommendations seem to share a profile:

Originally developer-first, so being agent-centric is a natural next step.
Docs are mature, with quickstart guides readily available. Low friction.
Strong presence in online communities like Reddit.

API-first products that gave a fuck about DX before AI have a head start.

If you’re just thinking about this now, you will naturally be a little behind.

What does agent-first GTM look like?

Translating this into things we can actually control, the companies that win this channel will have most of the following elements nailed:

If it’s not callable, it doesn’t exist

A clear API, CLI and MCP strategy is table stakes.

Without it, you’re irrelevant in this context.

Side note: if you’re still gating your API docs, karma’s coming for you.

Your product doesn’t need to be headless by default by the way — but the more callable it is, the more likely it is to be used.

Docs = distribution

I’d think of developer docs as a distribution channel.

Well-structured docs with clear, actionable quickstart guides for specific use cases. Minimise steps to the first API or MCP call.

Agents don’t tolerate friction. They’ll default to whatever works fastest, most predictably, with the least effort.

Structured reputability

Invest in things that be indexed and referenced—not brand, but hard artefacts; community-driven content, GitHub stars, changelogs etc.

I acknowledge this one is easier said than done…

Partnership is a land grab

Also, now’s the time to start building commercial partnerships with the companies that naturally sit at the interface layer.

If these products default to one provider, you should probably make sure it’s you.

If you don’t someone else will.

The risk if you ignore this

People buy from people—that’s true. But machines will buy from machines.

There’s a new waves of headless applications building agent-first or agent-only products with a structural advantage over incumbents and legacy software.

This isn’t black and white. Complex industries—enterprise, travel, healthcare—are more hesitant to trust AI, so adoption will lag. But in more open, progressive industries, this should probably be a key growth consideration.

Entire companies have been built by moving early on new growth channels.

This feels like one of those moments—get there first, and you likely win, given AI tends to default to a single option.

With that in mind, if you get there first and become the de facto choice, that’s a moat—which is increasingly rare as software gets easier to build.

Some closing thoughts

It wasn’t long ago that influencer marketing was considered niche—now entire agencies are built around it.

What might seem like a niche opportunity today will likely become a meaningful growth lever for software companies—not just for developer tools, but across categories like design, sales and recruitment.

The difference is this isn’t just a growth problem—it impacts strategy, product, and engineering. It goes deeper, which is probably why a lot of companies won’t bother.

The companies that win won’t just market directly to their audience.

They’ll market to the interface layers that make decisions on their behalf.

Over to you ✌️