I Went to a Demand Gen Conference. It Was an AI Conference.

Nothing else matters except AI right now.

Nothing else matters except AI right now.

I flew to San Francisco for Demand + Expand expecting two days of demand gen tactics. Pipeline optimization. Conversion strategy. The usual stuff.

What I got instead was an AI conference with a demand gen label on it. Every major presentation circled back to AI. Every hallway conversation ended up there. I think that tells you everything about where B2B is headed right now.

The crowd told the story before the first speaker took the stage. I talked to probably 20 or 30 people over two days. Almost all of them from slightly more established SaaS companies, 50 to 200+ employees. Mix of customer success, marketing, demand gen, and GTM people.

Not the leanest AI-native crowd. These are the companies one tier up from the five-person teams running everything on Claude and Make. Emerging and scaling SaaS. The ones that represent where the broader market is actually moving.

Speaker lineup was stacked. Craig Rosenberg, Guy Yalif from Webflow, Chris Rudegrapp from Sendoso, Trisha Gelman (CMO of Box). I didn't get to every session, but those are the ones that stuck with me.

There's a massive alpha opportunity sitting in plain sight right now. One slide showed the top decile of startups and their growth rates. These companies are deep in AI. Absolute AI natives. Growing at rates that make everything else look slow.

Rest of SaaS is growing at about 100% per quarter, which is incredible by any historical standard. But the gap between the AI-native top and everyone else is visible and widening.

I keep coming back to this. Technology adoption takes forever. We're 15 years into mobile. 20 years into SaaS. Still building new solutions for both.

AI is going to take a decade to really permeate the economy.

If the companies at this conference, scaling SaaS companies actively leaning into AI, represent the current frontier of adoption, think about how far behind the regular economy is right now. There is so much room to capture real competitive advantage.

AI agents were THE topic. About half the presentations were specifically about deploying agents inside organizations. Not whether to do it. How to actually make it work.

Single prompt gets you 80% of the way there. The last mile is what separates useful agents from abandoned side projects.

I compare it to food delivery. Getting the food from the restaurant to your neighborhood is the easy part. The hard part is the driver figuring out your apartment building lobby, finding the elevator, walking the hallway, getting to your actual door.

Same problem with agents. They break when APIs change. They need updating when connected products ship new versions. The organization's needs evolve and the agent has to evolve with them.

Somebody has to own that ongoing work. And that's where the four approaches that came up across multiple presentations matter:

1. Ruthless prioritization. Box runs an actual pipeline of agents, project managed the way you'd manage software builds and product rollouts. They prioritize which agents get built and when. This can't be willy nilly.
2. Treating agents like employees. Sendoso gives every agent a name, a job description, and specific requirements. Each one reports to a person. If that agent goes down, a human is responsible for getting it back up.
3. Budgeting for agents. How many tokens does this agent use? What human effort is it replacing? Is the AI actually better than a person for this job? Per-token costs are extremely efficient for fractional work, but this still requires active management.
4. AI centers of excellence. Finding people inside the company, from any department, who have the curiosity and skill set to build AI on their own. That person becomes the internal champion. Eventually it becomes their actual job title.

Sendoso's own AI journey was one of the more honest moments of the conference. They were early adopters. Did what everybody does. Started building hundreds of custom GPTs, tried to automate everything, ended up with about 400 agents.

Then the Pareto principle showed up. 20% of those agents were delivering 80% of the value. They've since scaled back to a handful and put real structure around planning which ones get built.

The human element stays. On the GTM side, speaker after speaker showed how much of the sales process can be automated now. Lead prioritization with Clay, research, talk tracks, follow-up sequences. Probably 90% of the pre-sale work.

But the actual sales conversation? A human still picks up the phone and connects with another person. I'm sure AI will keep getting better at this, but right now that part is still ours.

I think about this the same way for paid ads agencies. The architect role stays. The account manager who designs strategy, makes judgment calls, holds the client relationship.

All the background work should get fully automated. What remains is the creative work that requires original thinking. Our jobs get harder and more creatively demanding, not easier.

What this means for paid ads. This is where I bring the conference back to the world I work in every day.

Trisha Gelman from Box put out a stat. 94% of LLM users still use traditional search. That tracks with what we're seeing across our accounts. We have not felt a meaningful impact on paid ads performance from AI search. Not yet.

Competitor search is the #1 place for meaningful last-click revenue at the bottom of the funnel right now. Going after your competitors' prospects is an entire strategy on its own, and I think most advertisers are underusing it.

Search query length tells an interesting story. Average Google search is 4 words → average LLM query is 23 words. And growing.

We now have two ad products inside LLMs, AI Max on Google and ChatGPT ads. We've started rolling both out, but they're both so early that we honestly don't have much insight yet.

One thing I feel strongly about. Stop over-controlling match types. The algorithms are genuinely good at showing ads for complex search journeys now. People search in so many different ways that trying to lock down every match is a losing game. Retain control of your ad copy and let the algorithms handle distribution.

On SEO, you need extremely fresh content to stay relevant. For B2B, YouTube, Reddit, and G2 are where you get cited by AI. There was a full presentation on this at the conference.

Appify is worth watching. It's basically a marketplace for AIs with scraping capabilities that are now really accessible. For paid ads, the interesting part is being able to scrape B2B reviews and understand what a client's competitors are actually saying, then use that intelligence to inform ad copy. That kind of competitive research used to take hours of manual work.

Quick hits. Review sites are the #2 source for software shortlists, with G2 leading the pack (per Profound, December 2025). Events are still a major channel, especially intimate ones with personalized gifts. Customized outreach works.

The invisible B2B buyer is a topic big enough that I'm going to cover it in a separate post.

Outbound only works if you lead with pain. For outbound emails, don't mention your product. Only talk about the prospect's problem.

AI amplifies your message and your positioning. If the positioning is broken, AI amplifies the broken thing faster and at higher volume. It doesn't fix bad messaging. It scales it.

Use product usage data and CRM history to personalize outbound. The data is already sitting there. The question I keep asking our team is how we use what we already know about our clients' industries to do smarter outbound into specific verticals.

The reality is that AI does the same work cheaper, faster, and at higher volume. The positioning still has to be right. The copy still has to be sharp. AI just makes the gap between good work and bad work visible a lot faster.