We didn’t want to build it.
Let’s be clear about that upfront. We wanted to buy an AI agentic solution that does what we need in marketing. We looked. We demoed. We really, really tried.
But we just couldn’t find one that does what we actually need.
Why We Usually Buy (The 90/10 Rule)
Here’s the thing: at SaaStr, we’ve been running 20+ AI agents in production since May. We’ve invested $500K+ this year in AI infrastructure. We’re running 3 instances of Artisan for outbound (5-7% response rates, $1M+ in pipeline). We’ve got Qualified handling inbound ($1M+ closed revenue in three months, 70% of our October closed-won came from the AI agent). We built Digital Jason on Delphi (139,000+ advisory conversations). We use Gamma for decks, Momentum and Attention for RevOps, Opus Pro for video.
We know AI agents. We’re the #1 performing customer for both Artisan and Qualified across their entire customer bases.
And we follow one rule religiously: Buy 90% of what you need. Only build the 10% where no solution exists.
Why? Because you cannot build something better than what 10,000+ customers are already paying for. The opportunity cost is too high. The maintenance burden is real. Don’t build your own SDR platform, CRM integration, or general-purpose chat.
We’ve built custom tools on Replit before—our speaker review system (eliminated $180K in annual agency costs), our valuation calculator (300,000+ uses), our pitch deck reviewer (1,000+ founders helped). But only when we literally could not buy a solution AND it was a P1-P2 priority.
So when it came to marketing orchestration, we really wanted to buy.
The Problem: AI Marketing Agents Just Write Mediocre Content
Here’s what we discovered after evaluating every AI marketing agent solution we could find:
They all just write content.
Blog posts. Social captions. Email copy. Maybe some SEO optimization. That’s it.
And look—we already have more content than we know what to do with. SaaStr has 5,000+ pieces of content at this point. 20 million words. Content generation isn’t our bottleneck.
The real problem in marketing is orchestration. Strategy. Knowing what to do, when to do it, and actually executing against a coherent plan every single day across every channel.
Not a single AI marketing agent we found could do that. They could write a blog post. They couldn’t tell us which blog post to write today, how it connects to our email campaign tomorrow, which sponsors we should be reaching out to this week based on pipeline data, and how all of it ladders up to getting 10,000 attendees to SaaStr AI 2026.
So we reluctantly built our own. We nicknamed it “10K.”
Why Did We Build “10K”?
Because 10K has two jobs:
- Get us to the first 10,000 attendees for SaaStr AI 2026 in May
- Help drive the first $10,000,000 of revenue this year
That’s it. Clear goals. Clear accountability. Even for an AI agent.
What 10K Actually Does (It’s Not Magic, But It Works)
What 10K does isn’t that complicated or even that clever per se. But it’s comprehensive in a way no human on our team ever was—and no AI marketing agent we could buy attempted to be.
Here’s the architecture:
- Deep Historical Analysis: 10K uses Claude Opus to do deep analysis of all of our data for the past 4+ years. Every campaign. Every conversion rate. Every email open rate. Every sponsor interaction. Every registration pattern. All of it.
- Dynamic Campaign Planning: That analysis feeds into an app we built on Replit that has designed every campaign, every offer, for each single day of the year through mid-2026. I mean every single marketing action. Not quarterly themes. Not monthly initiatives. Daily executable tasks.
- Real-Time Updates: And here’s where it gets interesting—it updates in real-time. Daily. New data comes in, the plan adjusts. A campaign underperforms, it recalibrates. It’s not a static spreadsheet someone made in January and forgot about by March.
- Automation Where Possible: To the extent it can automate the marketing actions, it does. Not everything. But a lot.
- Predictive Intelligence: It also uses APIs from all the vendors we use and synthesizes them into real-time predictive data. Which sponsors should we be talking to right now? Who should we be targeting to attend SaaStr AI 2026 this week? When should we reach out to whom?
Basically, it auto-updates a deep Claude analysis of our entire dataset with real-time inputs and creates a daily workflow and calendar to execute against.
What 10K Can’t Do (And What We’ve Learned About Managing AI Agents)
Let me be honest about the limitations.
10K can’t do it all. Just like our AI SDR agents can’t fully replace human SDRs. There are still judgment calls. There are still relationships that need a human touch. There are still creative decisions where you need someone to make a call.
And here’s the brutal truth we’ve learned from managing 20+ agents: AI agents require about as much management time as humans. It’s different work—training, monitoring, iterating on prompts, catching hallucinations, reviewing outputs—but it’s still 30% of our Chief AI Officer’s time. Every day.
The Biggest Challenge: Maintenance
Here’s the thing nobody talks about when they hype “vibe coding” and building your own AI tools:
Vibe coding a tool is one thing. Maintaining it for years is another.
10K is now another agent that needs 30+ minutes a day to keep training it, working with it, adjusting it, fixing what breaks. Every single day. It’s not a “build it and ship it” situation. It’s a living system that requires constant care and feeding.
But then again—so do humans. A human VP of Marketing needs your time too. Weekly 1:1s. Quarterly planning. Strategy debates. Coaching. Course corrections. The difference is the type of work, not the amount of work.
And here’s my honest confession: The moment we can find an off-the-shelf AI VP of Marketing tool that does what 10K does, we will switch. In a heartbeat.
We don’t want to maintain custom-built software. We’re not a software company. We’re an events and media company that happens to need software to operate. The maintenance burden is real. Every time Replit updates something, every time an API changes, every time Claude gets smarter and our prompts need adjusting—that’s work we’d rather not do.
The 90/10 rule exists for a reason. We built 10K because we had to, not because we wanted to. And the second someone builds a product that handles marketing orchestration the way Artisan handles outbound or Qualified handles inbound, we’re buying it.
So if you’re a founder building AI marketing agents: Don’t just build another content generator. Build the orchestration layer. Build the thing that tells marketing teams what to do every single day based on real data. Build the AI VP of Marketing. We’ll be your first customer.
This week alone we had four major AI agent incidents across our portfolio. One agent ran an unauthorized A/B test and gave away free tickets. Another told people to come to an event that already happened (AI agents notoriously struggle with dates). A third broke entirely when a vendor pushed a “hot fix.” A fourth app just stopped loading.
We fixed them all. But it’s work.
The Real Bar for AI Agents in 2026
And maybe that’s the actual bar for AI agents in 2026.
Not “can it do everything?”
Not “is it perfect?”
But: Does it do more than any human ever did in this function?
10K does. Despite its limitations. Despite needing oversight. Despite the occasional failure mode.
It holds more context than any human marketer could. It recalibrates faster. It doesn’t forget. It doesn’t get sick. It doesn’t have drama. It executes against a plan with consistency no human team ever achieved.
And here’s what surprised me most:
10K does more than any human on our team ever did in this role.
Not because our team wasn’t talented. They were. But because the sheer volume of data synthesis, the daily recalibration, the constant optimization across every channel simultaneously—no human can actually do that at the scale 10K does.
What This Means for You
If you’re a SaaS founder or marketing leader reading this, here’s my honest take:
Don’t wait for the perfect AI marketing agent to exist. The ones out there that just generate content? That’s table stakes now. That’s not the unlock.
The unlock is using AI to do the hard work of marketing: the data synthesis, the pattern recognition across years of campaigns, the daily orchestration of dozens of channels and tactics, the predictive modeling of who to talk to and when.
If you can’t find a vendor that does this for your specific needs (and you probably can’t yet), consider building something yourself. Claude + Replit + your existing vendor APIs can get you surprisingly far. Our speaker review system took days, not months. Our valuation calculator has processed 300,000+ valuations.
Remember the 90/10 rule: Buy everything you can. Build only what doesn’t exist and matters enough to justify the maintenance burden.
But when you hit that 10%—when you need something that truly doesn’t exist—go build it. That’s where the magic is. Apps no one has built. Niches, even niches of n=1, that no one has built great software for.
The goal isn’t to replace your marketing team. The goal is to give them a co-pilot that never forgets, never gets tired, and can hold more context in its head than any human ever could.
That’s what 10K does for us.
And honestly? That might be enough from an AI agent in 2026.
We’re sharing everything we’ve learned about AI agents at SaaStr AI 2026 in May. All the tools, all the tactics, all the failures. See the full agent stack at saastr.ai/agents. And if you figure out how to build an AI VP of Marketing before we perfect 10K, let us know.
