I recently had one of the best sales experiences I’ve ever had with an enterprise software company. I didn’t talk to a salesperson. I didn’t book a demo. I didn’t even buy the thing.

The company was Viktor, which makes an AI employee that works in Slack and Teams. I got curious after I saw Matt Swulinski had joined its growth team. I installed Viktor in Slack and started asking the questions I actually needed answered. How does it handle memory? Can it connect to GitHub? What happens if someone on my team teaches it a way of working I don't want it to keep? How much control would I have over that?

I build agents for clients, so these were not the questions on a typical “book a demo” form. I could have spent a couple of weeks trading emails and sitting through calls to get that level of detail. Instead, I kept asking questions when I had them, sometimes very specific ones, and got a much clearer picture of whether the product fit my work.

At one point I asked what would stop a teammate from teaching it a mediocre way of working and having that become the team's way. Viktor's answer was unusually direct: nothing today. It said there was no review gate, test, or quality bar beyond its own judgment, and called that a genuine weakness compared with my PR-reviewed repo.

That was the moment I trusted the conversation most. It had described a limitation that mattered to the way I work, in enough detail for me to judge the fit myself. I didn't immediately adopt it for that use case, but I left knowing more about the product than I usually know after a sales call.

Then OpenAI announced it was testing Sponsored Agents, and I thought: yes, this is where buying is going. Maybe the more useful question for marketers is not “How do I get someone to fill out this form?” but “How do I help them work out whether we're right for them while they're interested enough to ask?”

I've been saying “landing pages are dead,” which is a bit dramatic even for me. I don't mean companies should delete their websites. I mean the landing page should stop being the main place buyers are expected to find every answer, especially when it gives them a little information and then asks for their name, email, company, job title, and availability before they'll get anything more useful.

As a buyer, that feels transactional and opaque. I don't know who will read the form. I don't know when they'll respond. And I still don't know whether the product solves my problem.

If I could chat with the body of knowledge a salesperson will eventually use to talk to me, I'd rather have that chat right now. Let me ask the odd, specific questions that matter to my situation. Let me do it at 10 p.m. without putting a call on my calendar or repeating myself to three people. If I realize it isn't a fit, that's useful too. The stakes of asking a question should be close to zero.

I've heard versions of this problem in client conversations. Should a prospect still move through a delayed, multi-step email nurture sequence before they can get a useful answer? If free-trial activation matters, should the first experience be another form and onboarding flow, or a conversation that helps someone start doing the thing? Those are design questions, not evidence that every company should rip out its current funnel tomorrow. But they are questions we should be asking.

Buyers are already learning to expect this kind of interaction. In G2's 2026 survey of B2B software buyers, 51% said they start software research with an AI chatbot more often than with Google. That's a measure of how those surveyed buyers research, not proof that they want a brand's agent to close every purchase. Still, it should get the attention of anyone whose ad destination assumes the buyer will patiently click through tabs, filters, PDFs, and a contact form to find out whether the product fits.

The website still matters. It needs clear product information, pricing context where possible, documentation, proof, and a way to verify what an agent says. I want the destination to be an information-finding journey: useful material I can inspect, plus an invitation to ask my own questions immediately. A form can exist when someone needs a human follow-up. It should not be the price of admission to an answer.

OpenAI's September 16 announcement describes Sponsored Agents as a test with select US advertisers. After someone clicks an ad in ChatGPT, they can choose to start a clearly labeled conversation with a business-sponsored agent. That conversation is separate from both ChatGPT's independent answers and the original chat. The person can ask follow-up questions and follow a link to the business's site when they're ready.

That is a different shape for an ad destination. The placement doesn't have to end at a fixed page that tries to guess every possible objection in advance. It can lead to an exchange shaped by what the buyer actually wants to know.

We don't yet have public evidence from that announcement that Sponsored Agents convert better, work for every category, or solve the messy parts of sales. It's a test. What interests me is the behavior it makes possible.

Imagine clicking an ad because a product looks promising, then asking: Will it work with our existing setup? What happens to our data? Can I try it without involving the whole team? What does it not do? A decent agent should give direct, current answers. A really good one should be comfortable saying, “That's not a fit,” or “I don't know; here's who can help.”

That last part matters more than another layer of personalization. A bot trained to turn every question into a pitch would have ruined my Viktor experience. The honest answer was what made it useful.

The company also needs somewhere to get those answers from. The raw material is probably sitting across product documentation, FAQs, support tickets, sales calls, implementation notes, pricing rules, reviews, and the questions people type into search. Some of it is approved to share publicly; some of it absolutely isn't. Building a selling agent means deciding which information is reliable, current, and safe for that agent to use, then testing how it behaves when a buyer asks something the source material doesn't cover.

I'd want marketing, sales, product, support, and security in that work. Marketing can see the questions that bring people in. Sales hears the objections. Product knows the actual limitations. Support knows what customers struggle with after the sale. Security and legal can tell you what the agent should never promise. If those teams don't agree on the answers, putting a chat window on the page won't fix the disagreement. It will just make it visible to buyers faster.

There may also be a product-learning opportunity in the questions people ask and the reasons they decide not to buy. I found myself wondering what Viktor could learn from the questions I asked. But I don't know what it records or how it uses those conversations, and I wouldn't assume any company has permission to treat a private exchange as free product research. If you want that feedback loop, you need to design it with appropriate notice, consent, access, and review.

My excitement about conversational buying doesn't mean I want an agent negotiating a large enterprise rollout on its own.

If I were buying Viktor for a big team, I'd eventually want to talk with someone about pricing, logins, data privacy, security governance, and what implementation would require. I'd bring an IT person. I'd want the human conversation to start with what I'd already learned, not with a salesperson asking me to repeat the last two days of questions.

The research points to that tension, too. In a Gartner survey of 645 B2B buyers, 67% preferred a sales-rep-free experience, while 69% preferred to validate AI-generated insights with a sales rep. Those answers can coexist. Buyers may want to research on their own and still need a person when the decision affects a whole organization.

So I'd start with one journey rather than announcing that the funnel has been replaced. Pick a product and a group of buyers with real, recurring questions. Gather the questions from calls, support conversations, search data, and customer interviews. Build an answer base with owners who can keep it current. Put an agent in front of target buyers and watch where it helps, where it gets confused, where it overstates the product, and where someone asks for a person. Compare that experience with the existing page and form. Measure whether buyers understand fit sooner, whether the handoff gets better, and what happens after the first interaction. A chat count by itself won't tell you any of that.

Look at your current ad destination. What would your best salesperson answer immediately that the page makes someone wait to ask? Start there.

What question do your buyers keep asking that your website still makes them book a call to answer?

Leave a comment below. I’d like to hear where people get stuck, and what you’ve tried.