Our Agents Drove Our Salesforce Bill Up 80%, Marketo Hates Agents, and We Stealth-Churned Off Notion
Jason and Amelia discuss running a 20-agent AI organization at SaaStr, covering their AI VP of Marketing (10K) autonomously generating campaigns, unexpected Salesforce cost increases despite fewer human seats, Marketo's broken unsubscribe flow that they fixed themselves in 20 minutes, and findings from a Redpoint survey showing 54% of CIOs are consolidating vendors.
Summary
Jason and Amelia provide a detailed operational update on running SaaStr with approximately 20 AI agents and 3 humans. Their AI VP of Marketing, called '10K', has evolved from a simple dashboard into a system that autonomously identifies campaign opportunities, segments lists, suppresses unqualified contacts, and drafts email copy — all without human prompting. Over a weekend, 10K identified that single-ticket buyers could be upsold to team packs, pulled the relevant data, suppressed unlikely converters, and drafted the email, stopping just short of hitting send due to API limitations with legacy tools like Marketo.
Marketo emerged as a major pain point. The platform has been failing to process unsubscribes properly, causing deleted contacts to still receive emails — a potential CAN-SPAM violation. After 10 days with no resolution from Marketo's support team, Amelia used Replit to build a custom unsubscribe page that hooks directly into Marketo's API, solving in 20 minutes what Marketo's engineering team couldn't fix in over a week. This episode highlights a broader theme: as vibe-coding tools make it easy for non-engineers to patch legacy software bugs, dated SaaS platforms look increasingly incompetent and replaceable.
On Salesforce, the pair discuss how they run it 'headless' — Jason has no login or password — with all data surfaced through 10K's dashboard. Despite cutting Salesforce seats by 60-70%, their bill increased 80% year-over-year due to heavy API and Agent Force usage. They argue this is a net positive: Salesforce data is now integrated with every metric across the organization. They also reference Owner.com's CTO who independently built a similar headless Salesforce system to analyze won/lost calls and manage thousands of restaurant customers.
The episode covers the discovery that they have effectively 'stealth churned' off Notion — neither Jason nor Amelia has logged in for months, as 10K replaced it as their standup dashboard. They also discuss building an N=1 AI parking pass distribution app for SaaStr Annual using Claude and Replit, which automated a task that previously took a human a full week.
On talent, they discuss Aaron Levie's prediction of a new $500K-$1M operator role that is somewhat technical and deep in AI. They argue finding such people is extremely difficult, suggesting that internal promotion (like Salesforce's FDE Lindsay), ex-technical founders, or poaching from Replit/Lovable/Vercel are the best paths. They criticize the trend of AI companies shipping without real support infrastructure, noting that a chatbot is not a replacement for actual customer success.
Finally, they reference a Redpoint survey of 141 CIOs where 54% said they're consolidating vendors and 45% of AI budget is being displaced from existing tools. The categories most at risk are customer service, finance operations, project management, and sales automation — all areas where SaaStr has already replaced legacy tools with agents.
About this episode
This week on The Agents, SaaStr AI CEO & Founder Jason Lemkin and Chief AI Officer Amelia Lerutte dig into what it actually looks like when your AI VP of Marketing starts running campaigns on its own, segmenting lists, writing copy, and asking if you've done the work yet. They also get into why SaaStr AI's Salesforce bill went up 80% after cutting seats by 60%, and why that's actually a good deal. Plus: Amelia vibed a fix to a critical Marketo unsubscribe bug in under an hour that Marketo couldn't solve in 30+ days, how headless Salesforce is already running inside their stack, and what an n=1 AI parking pass app reveals about the future of event ops. They also cover the Redpoint report showing 54% of CIOs are consolidating vendors, and which categories are getting cut first, plus the hardest hiring problem in B2B right now: finding the operator who can actually build and run your agents. __________________ Want to join the SaaStr AI community? We're the 🌎largest community for B2B + B2B software. Subscribe for weekly updates: https://www.saastr.com/subscribeform Twitter: https://twitter.com/saastr LinkedIn: https://www.linkedin.com/company/2724976 Quora Group: https://www.quora.com/q/cloud Facebook: https://www.facebook.com/SaaStr/ Instagram: https://www.instagram.com/saastr/ Our North American Event: https://bit.ly/2OXeAYh Our European Event: https://bit.ly/2OZTad8
Key Insights
- 10K, their AI VP of Marketing, autonomously identified a single-ticket buyer upsell opportunity over the weekend, segmented the list, suppressed low-propensity contacts, and drafted email copy — all without being prompted — stopping just short of sending due to Marketo API limitations.
- Despite reducing Salesforce seats by 60-70%, SaaStr's Salesforce bill increased roughly 80% year-over-year because agents are making heavy API and Agent Force calls, demonstrating that the 'death of the seat' doesn't mean lower costs — data and action usage replaces seat costs.
- Marketo failed to resolve a CAN-SPAM-level unsubscribe bug for over 10 days, while Amelia built a working replacement unsubscribe flow using Replit and the Marketo API in approximately 20 minutes — illustrating that legacy vendors become visibly incompetent when customers can vibe-code their own fixes.
- Jason and Amelia discovered mid-conversation that they have effectively stealth-churned off Notion — neither has logged in for months — because 10K replaced Notion as their standup dashboard once it reliably pulled live Salesforce and ticket data automatically.
- Claude separated 4,000 individual parking pass PDFs from a single file, appended street addresses to each, and completed the task in roughly 2 minutes — a job that previously took a human contractor a full week working manually in Acrobat.
- Amelia argues that finding agentic operators internally may be easier than external hiring because external candidates typically lack deep product knowledge, and without understanding the business, even technically capable people produce 'slop' agent deployments.
- A Redpoint survey of 141 CIOs found that 54% are consolidating vendors and 45% of AI budget is being displaced from existing tool spend rather than coming from net-new budget — meaning legacy SaaS vendors are directly funding the AI transition.
- Owner.com's CTO used Salesforce headlessly combined with Momentum call recordings to analyze every won and lost deal across thousands of restaurant customers in real time, identifying top feature gaps and win reasons — a use case the CTO said he previously found Salesforce too clunky to support.
- Jason argues that the consistency of AI agents is becoming a competitive advantage over humans — 10K never forgets campaigns it suggested, tracks whether the human has acted on its ideas, and proactively follows up asking 'have you done the work yet?'
- SaaStr's AI agents have closed what is likely seven figures in ticket revenue autonomously, but always at low ACV (sub-$1,000 per ticket); Amelia's stated next challenge is getting agents to close deals in the $25,000–$50,000 ACV range without human involvement.
- Jason criticizes AI-native companies for shipping without real support infrastructure, arguing that placing an untrained chatbot on a website is not a substitute for support, and that customers building their own workarounds to product bugs — as SaaStr did with Marketo — makes the vendor look worse, not better.
- Amelia identifies that the top personas for agentic operator roles are: internally promoted employees who are intellectually curious and already experimenting with agents on their own time (like their Salesforce FDE Lindsay), or burnt-but-recovered ex-technical founders who understand every business function.
Topics
Transcript
[0:00] Welcome everybody back to your favorite show, The Agents with Amelio Larut, our chief a officer and me. This week, our AIVP of marketing 10K shipped three campaigns on its own. Are we ready to go fully autonomous with no humans in the loop? Why we pay Salesforce more than we did before all our agents, even with less humans? I signed up for a new product this week, agent mail, because Cloud told me to. Is this the future? And 54% of CIOS told Redpoint they're switching their vendors out. They're [0:30] consolidating, what that means, and what we've learned on it. We'll get into that and we'll get into who's getting cut first. And before we get…
Full transcript available for MurmurCast members
Sign Up to AccessMore from SaaStr AI
Salesforce Headless: Cut Costs, Boost Data Access! #shorts
The speakers discuss their experience running Salesforce in a headless configuration within an AI-driven marketing platform, which has increased daily usage despite fewer human seats. Paradoxically, their Salesforce bill has grown compared to a year ago, even with fewer human users, due to heavy data and API consumption by AI agents.
Resend vs Marketo: Email API Battle! #shorts
The speaker compares Resend and Marketo as email platforms in an AI-driven marketing context. Resend is praised for its elegant API integration with agentic workflows, while Marketo is criticized harshly for its inaccessible API. Some friction with email domain verification is noted even for Resend.
Figma Fails: Why It Breaks for Print Graphics! #shorts
The speaker discusses recurring failures when using Figma for print production graphics at SaaStr events. Despite warnings, sponsors repeatedly attempt to use Figma for booth graphics, and the files consistently break. The speaker argues Adobe Illustrator remains the gold standard for print production work.
Build vs. Buy Apps: Avoid Salesforce & Workday Nightmares #shorts
The speaker and Amelia discuss the practical challenges of building vs. buying software applications, arguing that even if you could replicate tools like Salesforce or Workday, the ongoing maintenance burden makes buying far more cost-effective. They use a real-time database connectivity issue as a concrete example of the hidden costs of self-maintained software.