$254. That’s What It Cost Us To Run Our AI VPs of Marketing and Customer Success Last Month. | Saa Str

Overview

AI VC AI Mentor: Digital Jason + Amelia AI Startup Benchmarking

AI Agent Playbook Free e Books

	e Book: Hiring a Great VP of Sales
	e Book: Raising Capital
	e Book:  The First $1m ARR

Details

University All Posts Podcasts The Top CROs VC Fundraising Top Videos Q&A Best of Saa Str #1 Bestselling Book Search Everything Join the Community

Free e Books

	e Book: Hiring a Great VP of Sales
	e Book: Raising Capital
	e Book:  The First $1m ARR

AI Annual 2026 Events Overview Sponsors

	Event Sponsorship
	Media Sponsorship

Digital AI Day 2026 (Free) Speaker Submissions Speaker Requirements Overview

$254. That’s What It Cost Us To Run Our AI VPs of Marketing and Customer Success Last Month.

by Jason Lemkin | Artificial Intelligence (AI), Blog Posts, Saa Str. Ai

I just pulled the Replit bill for last month. Here’s what it cost Saa Str to run 10K (our AI VP of Marketing) and Qbee (our AI VP of Customer Success) for the month:

That’s it. $254.06 For Two AI VPs, One for Marketing, One for Customer Success.  One month. The price of a dinner for two in San Francisco.

For context, those two roles, filled by humans at a Series C+ B2B company, would run you somewhere between

Of course, it’s not the same.  Qbee and 10K do a lot, but they aren’t quite replacements for a human VP.  Still, it gives you a sense of what they cost to run.  More than a Squarespace site, or even a single Claude Max subscription.  But not all that much more, really.

Let me break down what these numbers actually mean and what these two agents actually do, because the punchline isn’t “AI is cheap.” The punchline is more interesting than that.

Replit charges per agent invocation. Variable pricing. The agent runs when it has work to do, and we get billed for the compute and the model calls underneath. No seats. No per-user. No annual contract minimums.

Here’s the line-item breakdown from the March invoice:

The infrastructure layer (autoscale compute, Postgres, data transfer) for the entire Saa Str app portfolio, not just these two, came in at about $50 for the month. Spread across all our production apps. Functionally rounding error.

Qbee is our AI VP of Customer Success. Amelia built her on Replit. Her job:

Manages 100+ paying sponsors for Saa Str AI Annual 2026

Coordinates booth assignments, deliverables, deadlines, and content reviews

Flags issues that need a human before they become problems

Before Qbee, this was a big chunk of a full-time human’s job, plus pieces of two other people’s jobs. Sponsor management at scale is a grind. Hundreds of email threads, hundreds of deadlines, hundreds of small commitments to track.

Post-Qbee, we estimate a 70% reduction in human hours spent on sponsor ops. The remaining 30% is the high-judgment stuff: pricing exceptions, escalations, relationship work.

10K is our AI VP of Marketing. Also built by Amelia. 14K+ lines of code under the hood. Her job:

Surfaces what’s working, what’s not, where to spend the next dollar

A real VP of Marketing does a lot more than this, obviously. 10K isn’t replacing a VPM in the strategic sense. She’s replacing the operational layer of a VPM — the running-the-machine layer that swallows 60-70% of a marketing leader’s week.

Result: we get the operational visibility of a VPM without paying for one. And our actual marketing leadership (me + Amelia + the team) gets to focus on the strategic 30%.

A classic Saa S seat like Salesforce, Gong, Outreach costs

With agents like 10K and Qbee, the agent IS the seat AND the human. The

And the agent works 24/7. Doesn’t get sick. Doesn’t miss Monday standup. Doesn’t quit for a competitor.

I’m not saying this replaces every role. It very much does not. Strategy, judgment, relationships, novel problem-solving, leadership, taste, voice, humans still win, by a mile.

But the operational layer of a lot of mid-level and senior roles? The “running the machine” part of the job? That layer just got priced at ~1% of what it cost a year ago.

What the full Saa Str Replit bill looks like, for everything we do: $2,324 a month.

Total:

Total:

So: ~$2,300/month to run the entire AI agent stack and app portfolio for Saa Str. Some of those apps have 600K+ uses. One has 4,000+ pitch decks analyzed. Another is the conference site for Saa Str AI Annual 2026.

For perspective: that’s less than what we used to spend on a single mid-tier Saa S tool.

A note on third-party AI agents: more expensive, still a great deal. Maybe a better deal. Buy don’t Build if you can.

If you’re reading this and thinking “great, so all AI agents cost ~$100/month and the market is wildly mispriced” — slow down. That conclusion would be wrong.

The $254 number is for agents we built ourselves on Replit. 10K and Qbee are in-house software. The cost is compute + our engineering time. They exist because no commercial product did exactly what we needed against Saa Str-specific data and workflows.

The third-party AI agents in our stack cost dramatically more. Artisan (our AI SDR), Agentforce (Salesforce’s AI), Monaco, Qualified, Momentum, and several others run us $25K+/month each, and up. Some scale higher with usage and seat count.

That sounds like a lot next to $159 for Qbee. It’s not the same product. It’s apples to oranges.

When you buy a third-party AI agent, you’re paying for:

The engineering team that built it (50-200+ engineers, not one Amelia on Replit)

The engineering team that built it (50-200+ engineers, not one Amelia on Replit)

Pre-built integrations into Salesforce, Hub Spot, your CRM, your email, your phone system

Pre-built integrations into Salesforce, Hub Spot, your CRM, your email, your phone system

Domain expertise. E.g,, Artisan has watched millions of cold emails win and lose. We haven’t.

Domain expertise. E.g,, Artisan has watched millions of cold emails win and lose. We haven’t.

Ongoing model updates, new playbooks, new features, delivered automatically

Ongoing model updates, new playbooks, new features, delivered automatically

Real reliability and uptime SLAs, not “Amelia is on PTO and 10K is being weird”

Real reliability and uptime SLAs, not “Amelia is on PTO and 10K is being weird”

And here’s the thing: at

Run the math on Artisan. AI SDR at

Built-in-house on Replit: very cheap (

Built-in-house on Replit: very cheap (

Third-party AI agents: materially more expensive ($25K+/month), but you get an enterprise product with engineering, integrations, support, and continuous improvement built in. Still 1/3 to 1/10 the cost of the human team they replace.

Third-party AI agents: materially more expensive ($25K+/month), but you get an enterprise product with engineering, integrations, support, and continuous improvement built in. Still 1/3 to 1/10 the cost of the human team they replace.

It’s not “build vs buy” as religious war. It’s “build vs buy” as portfolio decision. We do both. Most companies should do both. We will happily replace 10K or Qbee with a third-party tool the day one exists that does it better. The same way we already buy Artisan instead of building our own SDR.

The mistake to avoid: assuming the

The soft costs are real. They’re just lower than human management costs.

I want to be honest about this, because the $254 number is going to get screenshotted and people are going to misread it.

The hard cost is $254/month. The soft cost is not zero. Here’s what the soft cost actually looks like:

Build cost. Amelia spent weeks shipping each one. 10K is 14K+ lines of code. Qbee is a real production system with sponsor data, integrations, workflows. That’s a serious engineering investment. Not a weekend project. Not a no-code form. Real software.

Maintenance cost. This is the part nobody talks about. We are updating both agents constantly:

10K: every day. Not weekly. Daily. Marketing is creative, contextual, constantly shifting against the calendar and the news. Prompts that worked last week drift. New campaigns hit edge cases. Tone needs re-tuning. The model gets an update underneath us and behavior shifts. 10K is a high-creativity agent and high-creativity agents need high-touch maintenance.

Qbee: a few times a week, sometimes daily. More stable than 10K because customer success is more pattern-based — a sponsor question is a sponsor question. But still: knowledge base updates as new sponsors come in, permission tightening when she does something nobody asked her to, integration fixes when an upstream API changes.

A useful rule of thumb we’ve landed on: budget 0.5 FTE of human attention per production agent, not 0.1. Maybe more for the first 6 months. The “deploy an agent and it runs itself” pitch is wrong. What’s actually true: you’ve hired a brilliant new VP who needs daily coaching, daily corrections, and constant context updates.

But here’s the thing. Even with all that maintenance — and it’s real, and it costs me and Amelia hours every week — it is still dramatically less time than managing one human VP, much less two.

Think about what you actually do as a manager of a senior human:

That’s easily 5-10 hours/week of management overhead per senior human. None of which moves the business forward. All of which is required.

With 10K and Qbee, the maintenance is the management. When I “fix” 10K, I’m directly improving the work product. The hour I spend tuning a prompt is an hour that ships marketing output. The hour Amelia spends correcting Qbee’s behavior is an hour that improves sponsor ops.

There are no 1:1s with 10K. No career conversations. No “where do you see yourself in 3 years.” No comp negotiations. No PIPs. No goodbye lunches. No backfill recruiting. The agent doesn’t quit. The agent doesn’t get poached. The agent doesn’t have a bad week because of something happening at home.

So yes — the soft costs are real. We probably spend 3-5 hours/week between me and Amelia maintaining 10K and Qbee combined.

A human VP of Marketing would consume 5-10 hours/week of management time, before they did a single hour of work. And we’d need a second one for CS.

Net-net: the soft costs of running an agent are real, but they are still meaningfully less than the soft costs of managing a human. And every hour of agent maintenance directly improves output. Every hour of human management is overhead.

This is also why you can’t deploy 20 agents with a team of zero. We run 20+ agents with 3 humans because each of those humans spends a meaningful chunk of their week tuning, correcting, and rebuilding. The leverage is real. The fully autonomous part is not.

Cost is Not The Constraint.  Don’t Let It Hold You Back

Three things, if you’re trying to figure out what to do with this.

1. The cost is not the constraint. If you’re not building production agents because you’re worried about token costs or compute costs, you’re solving the wrong problem. The cost is irrelevant at this stage. Build the agent. The ongoing cost won’t matter if the agent delivers.

2. The constraint is the build. Most founders need at least one “agent deployment expert” who can actually ship a real production agent. Not a demo, not a chatbot, a real one that runs other parts of the business. If you don’t have that person, get them. They’re worth more than the next AE you were going to hire.

3. Pick the boring functions first. Sponsor ops. GTM standups. Status reports. Internal coordination. The parts of the business that consume human hours but don’t require human judgment. Those are the wins. Don’t start with the customer-facing crown jewels. Start with the back office.

We have 20+ agents now in production at Saa Str, running with 3 humans. The bill for the whole stack is $2,300/month. Revenue went from -19% Yo Y to +47% Yo Y in the same period.

I don’t think we’re special. I think this is what’s coming for everybody.

$254/month for two AI VPs. That’s the new baseline.

Revenue Alignment: How to Pull Marketing, Sales, and Customer Success Together with Divvy CRO Sterling Snow (Pod 628 + Video)

Scaling Customer Success from 0-5,000 Customers with Drata's VP of Customer Success and VP of Customer Experience

Revenue Alignment: How to Pull Marketing, Sales, and Customer Success Together with Divvy CRO Sterling Snow (Pod 628 + Video)

Scaling Customer Success from 0-5,000 Customers with Drata's VP of Customer Success and VP of Customer Experience

Get from

Key Takeaways

• AI VC AI Mentor: Digital Jason + Amelia AI Startup Benchmarking

• AI Agent Playbook Free e Books

    e Book: Hiring a Great VP of Sales
    e Book: Raising Capital
    e Book:  The First $1m ARR
  

• University All Posts Podcasts The Top CROs VC Fundraising Top Videos Q&A Best of Saa Str #1 Bestselling Book Search Everything Join the Community

• Free e Books

    e Book: Hiring a Great VP of Sales
    e Book: Raising Capital
    e Book:  The First $1m ARR
  

• AI Annual 2026 Events Overview Sponsors

    Event Sponsorship
    Media Sponsorship