Schmoozing Is Dead, Agents Are Hitting 120% of Humans, and Growth Is the Only Thing That Matters: 10 Things From the Closing Q&A at SaaStr AI Annual 2026
Schmoozing Is Dead, Agents Are Hitting 120% of Humans, and Growth Is the Only Thing That Matters: 10 Things From the Closing Q&A at SaaStr AI Annual 2026
by | Blog Posts, Featured Q&A, Q&A, Videos
We just wrapped SaaStr AI Annual 2026, and the closing Q&A on Day 3 ended up being one of the most engaged of the whole gathering. No script, no prepared deck, just an hour-and-a-half of questions from the audience on the topics most on their mind.
We went off on a lot of tangents. Some of you were there. For those who weren’t, here are the 10 things that mattered most. Some of these I’ve been saying for a year. Some we figured out in real time on stage. A few I’m still chewing on.
Let’s go.
1. Your Sales Team Needs to Be Product Experts. Schmoozing Is Officially Dead.
There’s a guy at Replit, Kody, whose title is something like Head of Forward Deployed Engineering. When I first hit a wall using Replit, I talked to him briefly. The first call didn’t really help. The second call did. He sat down, explained exactly why I was having the issue, and walked me through how to get around it. That’s it. That’s the whole story. And that single interaction is why I’m a Replit super-fan today and have brought them millions of dollars in business.
Compare that to Marketo. I had a bug. They forced Amelia onto the phone for 4 hours, blamed her for the problem, and then told me they wouldn’t talk to me again unless she agreed to another 4 hours. They didn’t help. At all.
That contrast is the whole game in 2026.
In the age of AI, where every product is changing on a weekly basis, your sales team has to know the effing product cold. If they can’t answer a technical question the way Kody can, I’ll bypass them. I’ll buy from the competitor whose rep actually understands the product. And so will every other founder I know.
The old model where the AE is a “people person” and the SE handles the technical questions is breaking. Not gone. But breaking. The reps who win now are the ones who know the product so well that customers can’t get rid of them, who hound them for months and years after the deal closes. That is the sign of a great rep. The customer keeps coming back, even after they’ve been handed off to someone else. The rep can’t escape them.
If your rep can’t do that, they’re just not part of the discussion.
2. The Marc Benioff Unguarded Moment
Last summer, Marc Benioff came on the 20VC podcast with me, Harry, and Rory. I asked him what he’d learned from Palantir. First he gave the line: that Palantir’s nine-figure deals show him he’s undercharging.
Then I caught him in an unguarded moment and asked what he thought about FDEs (Forward Deployed Engineers). He paused. And he said something I haven’t been able to stop thinking about:
“I can’t do this, but I wish every single customer at Salesforce had an FDE and did not have to pay me until they were deployed.”
Read what he actually said. The CEO of the most successful B2B software company in history wishes every customer could go live before they signed a contract. Not after a 5-year deployment. Not after Accenture got it working. Before. The. Contract.
That’s the standard Benioff wants to hit in the agentic era. Of course it doesn’t work for Salesforce’s existing motion. But it’s what a great founder wants. The agent has to work when they go live. That’s the new bar.
Whether you call them FDEs, SEs, SAs, or Forward Deployed Lawyers (like Harvey does with law firms, which I learned about from Anik at the CPO Summit), don’t over-index on the title. The mix is whatever it needs to be so that the agent already works on Day 30, not Year 3.
3. You Cannot Build a Real Relationship on Zoom
Every time a rep tells me they “have a great relationship” with a prospect, I ask one question: When did you meet them in person?
Today the answer is almost always: never.
Those relationships are fake. They are LinkedIn DMs and Zoom calls and a vague sense of familiarity. They are not relationships. And in a world where your buyer can replace you in 90 days with a better agentic vendor, fake-relationship inertia is not going to save you.
Dinners. Events. Bring your customers together. That still works. Schmoozing on text does not. It hasn’t for a while. We just kept pretending.
4. The Slop-to-Value Ratio Is the Only Frame That Matters
A founder at the Q&A pushed back on me. She said the SaaStr AI Annual site has slop. Things break. The mobile app had bugs. A lot of these new products are sloppy.
She’s right.
But she’s also looking at it wrong. The right frame isn’t is there slop. The right frame is what is the slop-to-value ratio, and which direction is it moving.
I started working on the SaaStr AI Annual site in November on Replit. It was 16,000 lines of code by March. By the time Amelia took it over and we got to the event itself, it was 48,000. Somewhere in those 32,000 new lines, there’s slop. I’ll stipulate that. But this event we just ran could not have existed on Squarespace, which is what we were on for the last 4 years. The value to slop ratio for us is 50 to 1, and the slope of improvement is steep.
The same is true for LinkedIn comments. Three months ago, automated LinkedIn comments were horrible. “Great job, team!” Garbage. Today the comments are getting really good. In 90 days they’ll be so good you’ll wonder if you should just have your AI talk to the slop AI and skip the humans. The slop quality is going up. (Whether that’s good for LinkedIn as a network is a different question. The value of those comments is still zero.)
But for our products, for the agents we deploy, the value is 50 times the slop. I can prove it. Monaco GTM, which we launched two months ago, has some AI slop in it, not everything is perfect yet. Sam, the CEO, will tell you that himself. But the first day we went live, Monaco found the right person at Anthropic, figured out Anthropic was in our ICP, reached out, and set a meeting. Whatever slop is in Monaco, that one outcome paid for it 100x over.
AI B2B software today is sometimes hacky, basic, primitive. It’s still early, despite the massive gains since just December. The AI inside it is sometimes better than the software itself. Both can be true. Give them a year.
5. The First Time I Saw an Agent Beat a Human
I had my real epiphany on this on-site at Annual, doing the Replit podcast with Kody. We were looking at the outbound emails 10K (our AI VP of Marketing, built on Replit) had written to get the right people to the event.
The copy was good. Better than most humans I’ve worked with. But the copy wasn’t the point. The point was the analytics. The amount of work that went into identifying the top 40 people to invite, the personalization, the targeting, the prioritization. A human just couldn’t do that. Not in that time, not at that quality, not at that volume.
That was the first time in a year of running 21 agents in production that I’d seen one clearly do something better than a human could. Not 80% as good. Not at parity. Better.
Once you see it once, you start looking for it everywhere.
6. Aiming for AI SDRs to Hit 80% of Your Best Rep. But In Some Parts of GTM, Agents Are Already at 120% of Humans.
Most founders building agents are anchored on “make it as good as 80% of our best human and then scale it.” That was the right frame a year ago. It’s the wrong frame now.
The new frame: find the slivers where an agent can be 120% of your best human, and start there.
A few examples from our stack:
Inbound. Qualified (now owned by Salesforce) set up 682 strong meetings for us in the run-up to Annual. That is more meetings than any human BDR I’ve ever worked with could have set, at higher quality, with zero quota-driven sandbagging. Better than any human BDR. Full stop.
Social selling. In 2021, people sold courses on social selling. The play was to manually monitor 100 prospects on LinkedIn, write “great job” when they posted, and pretend you were working. It mostly didn’t work. When someone I’ve never heard of writes “great job, Jason,” I don’t take a meeting.
But now an agent can actually do this with value. Imagine I post “I love Replit but I had an issue getting OAuth working today.” An agent can read that, look at my account, identify the actual issue, and reply with a real answer. A human could never do that at scale. An agent can do real social help, not social selling. That is 120%.
Outbound for the right ICP. When a prospect says “I’m looking for an AI SDR that integrates with HubSpot and does X,” an agent can reply with “Here are 22 customers with that exact workflow. Payhawk uses this configuration. Want me to introduce you?” That is instant, specific value. A human rep saying “Want to grab coffee?” cannot compete.
Find your slivers. Aim for 120%, not 80%.
7. The Worst Buying Experience in B2B Still Exists. AI Should Kill It Tomorrow.
You go to a vendor’s website. You don’t want to talk to sales. But there’s no other option, so you click “Contact Sales.” You get on a call. It’s a kid named Jason who just graduated from San Mateo Community College. He’s there to qualify you to decide if you’re worth the time of an AE who might have a slot next week.
He can’t answer any of your questions. You ask “Do you integrate with Replit?” He says, “I’ll have to ask Bob about that.”
That role still exists at thousands of B2B companies. AI can do 110% of it today. That role should go extinct as soon as humanly possible. It’s actively destroying pipelines.
8. If You’re Asking How to Get Your Leadership AI-Fluent, They’re Already Behind
Someone at the Q&A asked the best advice for leaders who aren’t in the loop on AI. My first reaction: if you have to ask the question, the org is by definition not AI-fluent.
You know when you are. You’re shipping code the way Andrew at Lovable is, where every product designer commits code. You’re not arguing with engineering about whether the tools work. The team isn’t telling you “these tools aren’t ready.” (Maybe they are. But if they’re saying that, you’re not fluent.)
If you have to ask, here’s the only thing that works in my experience: bring your leadership into a room and show them the single most magical thing your team does with AI. Live. Not slides. Not a deck. The actual workflow. The actual agent doing the actual thing.
If their jaws don’t drop, you’ve got a deeper problem. If they do drop, you’ve created the aha moment, and now they want to know how to bring it into the company. That’s how you drive change.
You can’t lecture people into AI fluency. You can only show them.
9. 2020-2021 Was an Aberration. Stop Trying to Apply Anything You Learned Then.
I keep hearing people compare today to 2020-2021. They are not comparable. Not even close.
What actually happened in 2020-2021: a virus caused everyone to buy 20-year-old software like it was going out of fashion. There was no new technology. There was no platform shift. Demand spiked because budgets were temporarily infinite. And then it ended. Almost instantly.
The clearest proof is Shopify. Shopify said in the summer of 2020 that it was already over. Their data, which is the pulse of e-commerce SMB, showed the spike ending in May/June 2020. The reason it looked like the boom lasted until Q1/Q2 2022 was long contract terms hiding a market that had already snapped back.
Hopin raised at $6B running digital events. By September 2020, the demand was gone. They just kept showing ARR because customers had signed multi-year contracts. The net new was zero.
We learned nothing from that era because nothing real changed. Software didn’t get better. There was no AI. We don’t even work from home that much anymore. Half my friends in the Bay Area are grinding back in the office every day.
This time is fundamentally different. It is a platform shift. The products are changing weekly. The buyer behavior is changing. The org structures are changing. None of the 2021 playbook applies. Throw it out.
10. The Engineering Job Market: B-Tier Is Dead. S-Tier Is in an Arms Race.
Engineering hiring is up 20-30% from the lows. Engineers are in high demand. And it’s never been worse for B-tier developers.
Both of those things are true. They’re not in conflict.
A simple example. My son is at Penn State, off the charts in math, doing real work in string theory and inner model theory. The labs find him. He has job offers.
None of his classmates in CS have a job. Zero. There is essentially no need for a B-tier, mid-grade CS engineer right now. It’s been devastating for that population. My daughter’s at Stanford. Nobody is going into CS anymore.
Why? Because in an agentic world, you can build basic software with way fewer engineers. So the bar has gone up. Replit and Lovable are in a death race, both at ~$500M in 18 months. Neither one can phone it in for a year. They have to ship harder than anyone has ever shipped. So they will hire any 10x or 100x engineer they can find. The arms race is at the top, not the bottom.
If the world were static, we could get by with fewer engineers. The world has never been less static. So great engineers have never been more valuable.
11. There’s No Land Grab in AI. There’s an Inertia Grab.
This is a counterintuitive one. Everyone in venture is talking about the AI land grab. I don’t think it exists, at least not in the classic sense.
The reason is that customers know the products are changing weekly. They’re explicitly signing 1-year deals (not 3-year) and telling vendors “we will look again in 10 months.” So you can’t lock customers in the way you could in 2015.
What does exist is an inertia grab. The desire to switch is real. The cost to switch is lower than it used to be. But the cost of retraining, of building muscle on a new platform, of trusting a new vendor’s roadmap, is high. Once we’re trained on a platform, the new vendor has to be twice as good to make us switch. Not 10% better. Twice as good.
So if you’re a vendor: this is the moment to be in market. Customers are still picking their default platforms. The defaults that get set now will compound for years. And in B2B, especially in the agentic era, customers aren’t just buying your software today. They’re buying the next 24 months of your roadmap. They’re buying your trajectory. Which is exactly why founders have to be in front of customers right now. The CEO sells the journey.
12. Venture in 2026: Growth Is the Only Thing That Matters. Full Stop.
Let me summarize all of venture capital today in one paragraph, because it’ll save you a lot of pitch meetings.
In 2026, nothing matters but growth.
People are writing checks into Anthropic at $950B. They don’t get to talk to anybody. You don’t get to talk to anybody. You have 48 hours to decide if you’re investing at $950B or not.
It is logical, even if it sounds like greed. The high end has no ceiling anymore. We’re going to see $3T+ IPOs in the coming months. The combined value of SpaceX, Anthropic, and OpenAI will exceed every tech IPO that has ever happened, combined. When the high end is that fat, no one cares about durability or quality of revenue. They care about growth.
The hard part: companies with very good growth that would have been easily funded 18 months ago won’t get funded today. 10-to-20 in a year used to be a slam dunk. With 120% NRR and durable, compounding revenue, you’d have term sheets in a week. The debate was whether next year was 36 or 42, not whether the company would exist.
In the SaaSpocalypse, those sure things look unsure. So 10-to-20 isn’t enough anymore. You need 5-to-30. 3-to-50. Replit, 2-to-500. That’s the bar.
If your growth isn’t in that zip code, you might take 100 beatings before you get a term sheet. That’s just the reality. Send me your growth. Maybe we’ll meet.
What We Learned at SaaStr AI Annual 2026
One year ago, Amelia and I came out of SaaStr Annual with one pseudo-agent, a Delphi clone of me. This year, we have 21 agents in production, 3 humans on the team, and $1M+ in AI-driven closed revenue.
Next year? I don’t know. Will we have 441 agents? Will we have meta-autonomous agents that don’t even exist yet? Will composable software replace traditional software entirely? I genuinely don’t know.
What I do know: every single thing we built, Amelia and I built without an engineer. You can do it. You can probably do it better. That is the mission. I want everyone at this event to be better than us in a year. When you are, we’ll have won.
See you at SaaStr AI Annual 2027.