May 12, 2026 · 1h 18m · saastr
Our AI is Hiring! Would You Work for One? And Are Autonomous Agents ... Safe? The Agents #005
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of 'The Agents,' Jason Lemkin and Amelia LaRoute explore the surprisingly low operational costs, technical architectures, and real-world workflows behind SaaStr's autonomous AI agents, 10K and QB.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Jason holds 60.1% of the talking time here. How this is scored →
Jason as informed peer 6.8 Guest teaching 4.3 Guest disagreement 1.2 Jason pushing back 1.5
speaking balance: gold is Jason, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 56:15 Amelia forcefully defends Salesforce against audience skeptics
Amelia rejects the common audience criticism that they only use Salesforce for sponsorship reasons, insisting it provides indispensable infrastructure that competitors cannot match.
Hardest push from Jason ▶ 8:15 Jason pushes back on worries over paying for agent code errorsJason refuses Amelia's framing that paying for agent missteps is a major problem, arguing that development speed and overall low dollar amounts make throwaway tokens inconsequential.
Biggest teaching moment ▶ 44:30 Amelia demonstrates how narrow context made the website agent outperform 10KAmelia walks Jason through how saastrainnual.com generated vastly superior marketing emails compared to 10K because it held the specific event assets and attendee context natively.
Jason holds their own ▶ 59:05 Jason details why custom Postgres CRMs collapse at enterprise scaleJason demonstrates deep operational expertise by detailing why building a bespoke CRM on Postgres is untenable when onboarding hundreds of sales reps who rely on Salesforce guardrails and workflows.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Jason as informed peer | Guest teaching | Guest disagreement | Jason pushing back | Why |
|---|---|---|---|---|---|---|
| The True Operational Costs of Running AI Agents | 7 | 4 | 2 | 3 | Jason breaks down the surprisingly low operational token costs of running their AI agents on OpenAI mini models. Amelia pushes back on the hidden software licensing costs and the frustration of being billed when vibe coding agents make mistakes, which Jason counters is negligible at scale. | |
| Is 10K Really an AI VP of Marketing? | 6 | 4 | 1 | 1 | Jason shares 10K's self-assessed scope of responsibilities and technical data on shifting attendee ICP demographics. Amelia connects 10K's tactical daily output directly to her original human job description as a director when she joined SaaStr. | |
| Operating Agents in Dev Environments and Real-World Campus Logistics | 7 | 5 | 1 | 1 | Jason explains the architectural advantages of keeping autonomous agents inside Replit development environments for infinite context retention. Amelia educates him on how she is actively using voice-enabled agent queries on-site to verify physical logistics across 40 acres. | |
| The Emergence of the SaaStr Annual Website as a New Agent | 6 | 6 | 1 | 1 | Amelia demonstrates how the event website evolved into an independent agent capable of parsing multi-page parking pass PDFs and constructing dynamic sponsor newsletters. Jason reflects on the need for orchestration and hierarchy layers across autonomous agents. | |
| Architecture Debate: Postgres vs. Salesforce for Agent Workflows | 8 | 4 | 2 | 2 | Amelia forcefully defends keeping Salesforce as a backend over pure Postgres due to integration stability and security. Jason backs this up with enterprise sales expertise, explaining why custom Postgres CRMs fail once sales teams scale to dozens of reps. | |
| Killing Point Solutions: Building an Automated Newsletter Generator | 7 | 3 | 0 | 1 | Jason describes building an in-house newsletter compiler in an hour that displaced an expensive standalone point solution. Amelia confirms how automating content force-ranking eliminated delegation friction with junior staff. |
Statements from this episode (15)
Lemkin: SaaStr's internal AI marketing and CS agents cost $257 monthly
“Our AI VP of Marketing customer success together cost 257 dollars.”
Lemkin: 95% of SaaStr AI agent calls run on OpenAI Mini
“About 95% of the OpenAI calls we're using are with Mini, which costs less than a penny per call.”
Lemkin: Buy off-the-shelf software rather than building custom AI agents
“The biggest takeaway is don't, still don't build it if you can buy it. We love Tenk and QB, they're part of our families, and I hate to say this, but we would dump both of them if there was an off the shelf tool. We wouldn't turn them off. This is the strange …”
Lemkin: Thoughtfully built GTM agents will cost under $1,000 a month
“For the stuff we're talking about, for building GTM agents, for building Autonomous dashboards, autonomous tracking. You're gonna struggle to, if you just do this reasonably thoughtfully, you're gonna, you're not even gonna spend a thousand bucks a month, righ…”
Lemkin: Tested AI CMO startups are just glorified ad managers
“The ones that we've tested are really marketing managers or ad managers on steroids. That's all they're trying. They're no, they're not CMOs. They, you can mock us for calling Tenk and AI VP of marketing, but mock the startups calling themselves AI CMOs that a…”
Lemkin: SaaStr no longer needs junior marketers due to AI agent
“Is it a partial replacement for junior marketers? Yeah, we don't need any.”
Lemkin: Running AI agents permanently in dev environments preserves infinite context
“If you run them in the dev environment permanently, like we're doing, you get an extra agent that has an infinite context window and infinite history.”
Lemkin: SaaStr's custom agent outperforms Intercom Fin and website chatbots
“This is fin or whatever intercom to the millionth degree that we built because it knows every so many details in real time for a hundred people and gets them all right. It's better than any chatbot on any website.”
Lemkin: SaaStr automated sending 5,000 individual parking passes via AI agent
“There's 5000 plus parking passes, each of which are individual. To each attending. And of which there's three different categories. One day, three days. And so this was weeks of man time before. And you're surprised you were able to get the agent to automate i…”
Lemkin: Salesforce serves as the connective glue across SaaStr's AI agents
“Right now, Salesforce is our connective glue, right? We've talked about this many times, but it'd be even better since they're all built on Replit, if Replit could be the glue for the agents, right?”
Lemkin: 99% of successful GTM agents are optimized around Salesforce
“In GTM, I think, 99% of the agents that are working today, that are successfully are optimized around Salesforce.”
Lemkin: Platforms managing hybrid human-agent teams will win; Marketo will lose
“I think Marketo is the loser here, but I think in a lot of cases, the leaders that are, that can manage a human agent interface are going to win because we need apps to manage hybrid teams.”
Lemkin: SaaStr replaced a $4,000 annual newsletter tool with an AI agent
“Most important thing is we also got rid of a piece of software we've used for eight years. It was 4000 dollars to do this. So Amelia would manually use this company called Bee. Which would not really hook up into Marketo, but it's a nice little newsletter buil…”
Lemkin: AI will kill single-function point solutions, not public SaaS leaders
“Is the SaaS apocalypse going to kill the leaders in B to B that are public? No, there's too much workflow. They're too rich. They're too complicated, but these point solutions like this B newsletter builder, we don't need it.”