SaaStr built an AI VP of marketing

SaaStr spent over $500K on AI infrastructure before concluding that no off-the-shelf marketing agent could do what they actually needed. So they built one from scratch and named it 10K. What they found is worth paying attention to if you're building inbound pipeline right now.

The content factory mistake

SaaStr has 5,000-plus pieces of content across 13 years. Twenty million words. When they surveyed the AI marketing agent market, every tool they demoed did the same thing: write more content. Blog posts, social captions, email copy, some SEO optimization.

The problem is that content volume was never their bottleneck. It's rarely anyone's real bottleneck once they've been at this for more than a year.

Content generation is the easiest part of the problem to automate. It's also the least valuable part to automate. AI marketing tools defaulted to content because what buyers are easiest to impress with in a demo is a polished blog post in 30 seconds. A tool that tells you your Tuesday email campaign is cannibalizing your Friday LinkedIn window is harder to show in 20 minutes. So the industry optimized for the former.

SaaStr's 90/10 rule is worth stealing: buy 90% of what you need, build only the 10% where no off-the-shelf solution exists. They bought aggressively and found the 10% the market hadn't solved. That gap, between content creation and marketing orchestration, is where they built.

What 10K actually does

10K runs on Claude Opus analyzing five-plus years of SaaStr's historical data, fed into a Replit app. Every day it:

• Plans the week's and quarter's marketing activities
• Builds campaigns with specific offers, messaging, and targets
• Assigns specific tasks to specific humans with dates attached
• Connects to Salesforce and predicts which sponsors are likely to convert and when
• Updates in real time as new pipeline and vendor data comes in

The last point matters more than it looks. Most marketing plans are built monthly or quarterly and then defended against reality rather than updated by it. 10K updates daily. When a campaign underperforms, it flags it on day two, not day sixty.

That distinction is where most human marketing operations break down. There's a natural bias toward defending the plan you made, especially when someone on the team has been running a specific channel for three years. 10K has no sunk cost to protect. It reads the pipeline data and tells you which channel is outperforming by 4x without caring that someone built their identity around the other one.

Zero agenda is genuinely valuable. Most marketing teams don't have access to it and wouldn't easily admit they're missing it.

Is it better than a human VP?

Not exactly, and SaaStr is clear about this. Better than an average human marketer on data processing and unbiased analysis, yes. Better than a great VP of marketing, not yet.

The more interesting framing is that the question itself is wrong. SaaStr runs 10K in parallel with a human VP. Both produce recommendations. Both get compared. The pitch from most AI vendors is that AI replaces expensive headcount. At SaaStr, AI makes the human sharper by providing a second opinion with no political stake in surviving the next review cycle.

For operators building inbound pipeline, the useful translation starts here: the goal is not to generate more content. The goal is to show up in the right places at the right time with enough consistency that the right people start recognizing you. That requires sequencing intelligence, not just output volume. Understanding how LinkedIn inbound signals work is the foundation before any automation layer makes sense.

Where it falls short

SaaStr is honest about the gaps. There's no central dashboard yet for managing all AI agents across the business. 10K handles marketing data well; it doesn't yet integrate cleanly with every other agent running in parallel.

The subtler gap is data quality. 10K is as good as the data it has. Five years of SaaStr's historical data is a genuinely rich training set. Most operators don't have that. If you've been posting on LinkedIn for eight months with inconsistent formats and no systematic tracking of what landed, an AI orchestrator has thin material to work with.

This is why auditing your existing presence before adding any automation layer is the right order of operations. The operators who get the most out of systematic engagement have usually done the work of understanding their current pattern first: what's working, what's been buried, which comment threads generated actual conversations versus isolated likes. A quick bottleneck audit is the precondition for useful orchestration.

Every model improvement makes 10K more precise; every week of new data sharpens its predictions. The compounding is the point. The asset being built is the data layer, and that asset compounds in a way that content generation alone never does.

What operators should take from this

The market for AI marketing tools is optimized for demo-ability, not for the actual bottleneck. The easiest tools to buy are solving the wrong problem for most operators who have been at this for more than a year.

Agenda-free analysis is genuinely scarce. Anything you can use to get an honest read on what's working is worth the friction.

The data layer compounds. The reason to start tracking your engagement behavior systematically is not to optimize this week's comments. It's because six months of clean data is worth more than six months of scattered activity when you want to understand what's actually driving inbound. The patterns that drive LinkedIn inbound become visible once you have enough signal. Before that, you're guessing.

SaaStr's 10K is a specific solution to a specific problem. But the underlying logic applies broadly: stop automating the wrong thing, build the data asset, run AI recommendations alongside human judgment rather than replacing it.

“The market optimized AI marketing tools for demo-ability, not for the actual bottleneck. The easiest tools to buy are solving the wrong problem.”