Key Takeaways
- Jason Lemkin at SaaStr argues B2B buying is shifting from "ask your developer" to "ask your AI agent, " with his own AI VP of Revenue now gating which vendor tools even get evaluated.
- If a buyer's agent does the first-pass shortlist, your website, pricing page, docs, and changelog have to be legible to a machine before a human ever opens them.
- The new proof point isn't a logo wall, it's whether your product is actually running inside another company's agent stack in production.
- CMOs should stop writing only for the human reader and start publishing structured, agent-readable evidence: capability lists, integration specs, pricing logic, and real usage data.
- This shift favors small, specific, well-documented products over big fuzzy platforms, which is a reversal of the last fifteen years of B2B marketing.
Jason Lemkin wrote something last week that's been sitting with me. He said the new "ask your developer" is "ask your agent, " and he described how his own AI VP of Revenue, which he calls 10K, is now the thing deciding which vendor tools make it to a real evaluation at SaaStr. A founder is telling you his agent is already the gate.
If you run marketing at a B2B software company, the question that falls out of this is uncomfortable and immediate. If a buyer's agent is doing the first screen, is any of what you're publishing actually readable by that agent, or is it all beautiful hero sections and gated PDFs that a model will skip past on its way to a competitor who wrote their capabilities in plain text?
Most of us are in the second group right now. And we have about a quarter, maybe two, to fix it before this becomes the default motion rather than the Lemkin edge case. So let me walk through what I'd tell a CMO to do this month.
The shift Lemkin is naming
Read the SaaStr post carefully, because the headline undersells the move. Lemkin isn't saying buyers will consult ChatGPT the way they used to consult Google. He's saying the agent becomes a persistent internal stakeholder, like a junior analyst who never leaves, and that analyst builds the shortlist, reads the docs, runs the trial, and tells the human what's worth a demo.
He also drops a diagnostic I think is the sharpest thing in the piece: track which competitors' agents are actually in production use at your target accounts. Not which ones have press releases about AI, not which ones have a chatbot on the pricing page. Which agents are actually doing work inside the buyer's stack right now. That's the new reference check, and it's measurable in a way that "brand consideration" never was.
The agent isn't just a channel anymore. It's increasingly the buyer.
What this actually looks like in practice
Here's the pattern I keep seeing when operators use agents to evaluate tools, which is now basically how a lot of technical buyers shop. You ask Claude or a project in ChatGPT to compare three options for, say, a transactional email sender, or a Postgres host, or a workflow tool. The agent goes and reads. It reads docs, it reads changelogs, it reads GitHub issues, it reads pricing pages, it reads Reddit threads. It does not read your beautiful homepage animation, it does not watch your 90-second product video, and it does not download your gated ebook, because it can't, and even if it could, it wouldn't quote from a PDF when it has crisp Markdown docs available from a competitor.
So the ranking the agent returns is almost entirely a function of which vendor made themselves most legible. The best-documented product wins the shortlist, and the shortlist is now most of the decision, because the human downstream is busy and trusts the agent more every week.
Your competitor with the uglier site and the deeper public docs is beating you in agent-mediated evaluations today, and you don't have a dashboard that shows it.
What "legible to a machine" actually looks like
Let me get specific, because "write for AI" is the kind of phrase that means nothing and gets you nowhere. When I say make yourself legible to a buyer's agent, I mean concrete things, and they stack together: your capabilities should exist as a plain list somewhere on your site, in HTML, not inside a carousel or a Figma embed, so an agent crawling your site can answer "what does this product do" in one page without inference; your pricing should include the logic, not just the numbers, because agents are very good at math and very bad at guessing, and a pricing page that forces a sales call is a pricing page that gets dropped from the shortlist; your integrations should be a real list with real endpoints, not a logo soup, because the agent is going to check whether you connect to the three things already in the buyer's stack, and if that answer isn't on your site in text, the agent will assume no.
Then there's the changelog.
A public, dated changelog is the single highest-signal thing an agent reads to decide if a product is alive. Weekly entries beat a glossy "what's new" page every time, because the agent can count the entries. And your docs should be in Markdown or at least in clean semantic HTML, hosted on your domain, indexable, with code samples that copy cleanly. If your docs are in a JavaScript-rendered knowledge base that requires login, you are invisible. Teams that rewrite their capabilities page this way report a noticeable difference in how agents summarize them back to buyers.
The scene that makes it click
Picture the walkthrough. A team is picking a background job runner for an internal dashboard. Someone opens a Claude project, pastes in the stack, and asks for a recommendation with tradeoffs. The agent comes back with three options, ranked, with reasons.
The winner isn't the one anyone would have picked from brand recognition. It's the one whose docs are cleanest and whose pricing page has an actual formula on it. Ask the agent why it ranked the better-known competitor second, and it tells you, in effect, that it couldn't verify the pricing without a sales call and the integration docs required an account to view. So it discounted that option.
That's the whole shift in one anecdote. The agent didn't dislike the bigger brand, it just couldn't confirm the facts, so it moved on. A year ago the buyer would have booked that demo anyway because they'd heard of the vendor. Last week, they didn't, because the agent had already done the work and the answer was good enough.
What CMOs should actually do this quarter
Three things, in order.
First, do an agent audit of your own site. Open Claude or ChatGPT, give it your URL, and ask it to evaluate your product for a specific buyer persona against two named competitors. Read the output honestly. Where it hedges, where it gets you wrong, where it ranks you lower, that is your content roadmap for the next ninety days. This takes an afternoon and it will be the most useful afternoon your team spends this quarter.
Second, publish the things an agent needs and a human doesn't want to write. A real capabilities page. A real pricing logic page. A real integrations list with versions. A public changelog. Think of these as machine infrastructure for your brand, not thought leadership, and almost nobody on your team wants to own them, which is exactly why you have to.
Third, start measuring presence in agent outputs the way you used to measure share of search. Pick ten real evaluation prompts your buyers would use, run them weekly across the major models, and track whether you show up, how you're described, and who you lose to. If you want help thinking through how to structure that measurement system, we put together a blueprint for teams working on exactly this shift.
When this is the wrong thing to focus on
Honest caveat, because this piece will age badly if I pretend the shift is total. If you sell into regulated enterprise accounts where procurement still runs a six-month RFP and the CISO personally interviews vendors, agent legibility is a secondary concern this year. The human gate is still the real gate. Keep doing the enterprise motion.
And if your product genuinely requires a consultative sale because the implementation is custom, forcing pricing transparency to please an agent will cost you more than it earns. Agents are shortlisting commodity-ish tools fastest. The more bespoke your work, the slower this hits you.
But if you sell self-serve or product-led or any tool a technical buyer could adopt without a committee, the agent is already in the room. Spend Monday morning running the audit on your own site and see what comes back.
Frequently Asked Questions
Does this mean SEO is dead?
No, it means SEO is becoming a subset of a bigger discipline. The same structural clarity that helps Google rank you helps an agent summarize you. The difference is the agent won't click through ten results, so being on page one isn't enough. You have to be the answer the agent uses.
How do I know if my competitors' agents are in production?
Ask your customers directly. In your next ten customer conversations, ask what AI tools they've connected to their stack and which vendor agents are actually running. You'll get a clearer map in two weeks than any analyst report will give you in a year.
Should we build our own agent for buyers to use?
Maybe, but not first. First, make yourself readable to the agents your buyers already have. Building a branded agent before your docs are machine-legible is like building a showroom before you have a product.
What about gated content and lead capture?
Gating still works for genuinely high-value assets a human will trade an email for. It does not work for product information an agent needs to evaluate you. Separate the two. Gate the research report, ungate the integration docs.
How fast is this actually moving?
Faster at the top of the market than the middle. Founders and technical buyers are using agents to evaluate tools now. Mid-market marketing teams are six to twelve months behind. Enterprise procurement is probably two years out. Plan accordingly, but start now.