When your next trial user is not human...

AI assistants such as Claude Code, Codex and Cursor now write a lot of software. And more and more, they also choose the tools that software runs on. A developer asks for a place to store data or a way to send emails, and the agent compares the options, picks one and moves on.

So who’s reading your pricing page first? More and more often, it’s an agent.

​Armature, who helps developer-tool companies get recommended by these assistants, ran nearly 17,000 test sessions to see which products the agents chose and why. Two examples stood out to me.

​Mailgun and Postmark both send emails from apps. Mailgun regularly lost to Postmark when the assistant saw that its free plan kept only one day of message history. Supabase, a database service that also bundles login, file storage and other extras, almost always lost when the agent wanted just a database, because its pricing presented everything as one package. And in many of the sessions where cost or effort came up, Armature noticed the problem had more to do with how a company presented the information than with anything that actually ruled the product out.

In those cases, the copy cost them the pick.

This is early evidence, and it mostly covers tools developers use. Armature also has a stake in the answer and used an AI to play the human user, so I’d treat it as a strong signal rather than proof. But the behavior itself is already common: Supabase says AI tools now launch more than 60% of its new databases.

Why would an AI agent skip a product that fits?

Free trials and self-serve sign-ups are built for a curious person. They click around, try a feature, hit a wall, find a workaround, and at some point the product clicks and they get why it’s worth paying for. That’s the aha moment every onboarding flow is designed to reach.

An agent never gets there. It shows up with one job, checks that job against what your page actually says about the product, the limits and the price, and makes a call. If your copy is vague, gives five jobs equal weight or makes one limit look scary, the assistant skips you, even if your product would have been the better choice once someone tried it.

That’s why I see this as a messaging problem first. You can make your site easier for an agent to read, but that won’t help if the message hides which job matters most. A bundle can feel like more value to your team and look like a mismatch to an assistant that came for one thing. A limit you never explained can look like a reason to rule you out.

I often describe buying decisions as three yesses: Is this for me? Does it solve my actual problem? Do I trust them? The agent is starting to answer the first two on the buyer’s behalf. And it takes your words literally. It won’t fill the gaps the way a sales call would, and it gives you no credit for context that only lives inside your company.

Who still has to say yes?

A person, for now. Resend, another email service, says a human has to create the account before an assistant can take over the setup. And in G2’s survey of more than 1,000 software buyers, 51% said they start research in an AI chatbot more often than in Google, yet G2 is clear that AI “is not replacing the buying process.”

The agent narrows the options. A person still signs off on the account, the cost and the risk. That changes an important trust question. Your buyer isn’t only asking, “Do I like this?” They’re asking, “Do I trust the agent’s choice?” To say yes, they need to see what they’re signing up for: what it will cost once the free plan runs out, what happens to their data, and how much control they keep.

None of this means people will stop visiting your site. The first reader is changing, that’s all. And the things that help an assistant pick you (a clear job, honest limits, one obvious main offer) are the same things that have always helped people understand what they’re buying.

So when you read your own page, ask:

Can a reader tell exactly what job your product does and who it’s for?

Could a price, feature limitation or bundle make your product look like the wrong fit to agents?

Can the human decision maker quickly see what it costs, what happens to their data and what controls they get?

If you can’t answer those quickly, better copy alone won’t fix it. The answers come from positioning (which job you win, for whom, and against what) and from a messaging strategy that puts that answer first on every page an agent or a person might read.

Discovery

Distribution is what competitors can’t copy

What happens when anyone can clone your product? Shann Holmberg’s guide to building distribution starts there. Agents can now rebuild a lot of your features fast, he argues, but they can’t rebuild the people who chose to follow you, the subscribers who opted in or the partners who trust you.

So he sorts every channel by two things: where it reaches the buyer, and how you get it (build, partner, pay or acquire). Then he maps 24 of them across the funnel. One of them is the integrations and plug-ins AI assistants can use, which is a channel I didn’t expect to see on a list like this. His advice for picking your first two is refreshingly practical: different funnel stages, one you build and one you partner or pay for.

I’d add one thing. If agents are a channel now, your message is what travels through it.

The storyteller can’t make your choices for you

Lots of tech companies are hiring “storytellers” right now. In “The Storyteller Trap: You Can’t Outsource Conviction”, Smit Patel asks what problem they’re actually hiring for.

Slow growth often gets diagnosed as an awareness problem. Sometimes it is. Often, though, the real issue sits in the product, the people or the process, and quite often it’s a founder who hasn’t decided which customer to serve. The roadmap sprawls, and marketing gets asked to make it all sound coherent.

His closing line stuck with me: “sometimes the most important thing you can do for your company’s story is decide what your product is no longer going to try to do.” Same bundling problem as today’s newsletter, one level up.

Resonance