Mar 4, 2026 · 55m · saastr
The Top 5 Issues Managing Multiple AI Agents in Production
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 SaaStr webinar and podcast episode, Jason Lemkin and Amelia Ibarra analyze the practical challenges, technical fragmentation, and managerial overhead of operating roughly 30 production AI agents and vibe-coded applications. They provide actionable frameworks for organizational succession, compliance audits, hiring operators, and evaluating genuine financial ROI to help SaaS leaders scale autonomous workflows sustainably.
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 29.3% of the talking time here. How this is scored →
Jason as informed peer 4.9 Guest teaching 1.2 Guest disagreement 0.4 Jason pushing back 0.4
speaking balance: gold is Jason, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 30:20 Amelia Confronts Her Own Agent's Demands
Amelia describes pushing back against her internal agent when it ordered her to immediately deploy campaigns at midnight despite human fatigue.
Hardest push from Jason ▶ 43:50 Jason Explains Orchestration Layer ObstaclesJason dismisses the naive audience question about building an in-house orchestration layer, asserting point solutions lack complete API exposure.
Biggest teaching moment ▶ 28:10 The 12,000-Line Codebase Reality CheckAmelia reveals her vibe-coded app reached 12,000 lines of code and required complex Zapier, Clerk, and Replit integrations, demonstrating unexpectedly high architectural complexity.
Jason holds their own ▶ 48:50 Jason's Warning on Startups with Weak ROIJason authoritatively breaks down unit economics and customer demand, warning founders that incremental productivity gains cannot compete against agents delivering immediate six-figure pipeline.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Jason as informed peer | Guest teaching | Guest disagreement | Jason pushing back | Why |
|---|---|---|---|---|---|---|
| Scaling AI Agents and the Chief Agent Officer Role | 4 | 1 | 1 | 0 | This introductory teaser and ad block sets up the premise of managing AI agent fleets. Jason Lemkin comments on organizational necessity for a Chief Agent Officer to ensure continuity when scaling past an N=1 setup. | |
| Overview of the Top Five Issues Managing Production Agents | 4 | 2 | 0 | 0 | Amelia outlines the top five issues encountered running nearly thirty production agents and vibe-coded tools. Jason prompts her to describe her specific daily routine and segmentation strategy across different interfaces. | |
| Context Switching and Daily Agent Maintenance Management Routine | 5 | 1 | 1 | 1 | Amelia details the grueling daily maintenance and context switching required across point solutions like 10K, Artisan, Qualified, and Agentforce. Jason notes that despite industry hype, no true orchestration agent exists to unify these separate dashboards. | |
| The Two Week Blackout Period When Onboarding New Agents | 3 | 1 | 0 | 0 | Amelia explains that onboarding any new agent triggers an unavoidable two-week operational blackout where other tools temporarily idle. She highlights trade-offs using Monaco as a concrete example of worthwile onboarding friction. | |
| The AI Agent Succession Planning Crisis and Operational Risks | 6 | 2 | 1 | 1 | Amelia shares comical yet alarming outputs from agents when asked what would happen if she got hit by a bus. Jason reinforces that relying on a single human operator creates severe existential enterprise risk. | |
| AI Agents as Unforgiving Truth Tellers and Daily Accountability | 3 | 1 | 1 | 0 | Amelia reflects on the psychological strain of dealing with 24/7 AI agents that relentlessly track targets and roast her for lagging behind. The dynamic remains collaborative and humorous as she shares Claude transcripts. | |
| Security Audits Compliance and Maintenance for Vibe Coded Apps | 7 | 2 | 0 | 1 | Jason steps in with strong domain guidance on security and compliance, pointing out that custom vibe-coded apps and startup agents are inherently less secure than foundational platforms like Salesforce. He advises regular automated audit prompts and starting with non-sensitive data. | |
| How Managing 24/7 Agents Alters Human Management Expectations | 2 | 0 | 0 | 0 | Amelia delivers a monologue explaining how managing always-on, hyper-informed agents degrades human management patience and highlights vendor laziness during RFP processes. | |
| Live QA Orchestration Feasibility Hiring Criteria and True ROI | 8 | 1 | 0 | 1 | Jason provides definitive perspective on measuring agent ROI strictly by revenue generation and human replacement. He issues a direct challenge to SaaS founders that mediocre ROI is dead in an era where top agentic tools face unserviceable demand. | |
| Live QA Autonomous Marketing Gaps and SaaStr Annual Preview | 7 | 1 | 0 | 0 | Jason and Amelia address the persistent gap in autonomous marketing agents compared to sales tooling. Jason expresses frustration with current point solutions that fail to provide comprehensive full-funnel marketing execution. |
Statements from this episode (23)
Ibarra: SaaStr runs nearly 30 production AI agents and apps
“We have all 20, actually almost 30, I think when I counted it this morning. Agents and Vibe Coded tools and apps in productions.”
Ibarra: Context switching becomes extremely difficult with 20+ AI agents
“I would say issue number one, and this is maybe force ranked, is context switching becomes extremely difficult when you have 20 plus agents.”
Ibarra: Every new AI agent requires a two-week blackout period
“Even with the help we get, like, I feel like every new agent costs us at least two weeks of what I call either a blackout period and somewhat chaotic of, like, I kind of have to ignore the other agents anytime I bring on a new agent”
Lemkin: No master orchestration product exists for multi-vendor AI agents
“We haven't found anyone that can integrate Asian force artisan qualified Monaco. Momentum. Attention. Attentive. Like we don't, there's nothing out there today, even with Tentay, which we built ourselves that at some magical API or other level can integrate ev…”
Ibarra: Salesforce will integrate Qualified and Momentum by Dreamforce
“I will say the benefit of Salesforce. Buying qualified momentum is like eventually by probably Dreamforce, right? Those two things will be integrated into one. So we have like two less dashboards and I'll probably have one centralized one by October”
Ibarra: AI agents idle and waste money without daily human management
“You can't, you, it just can't afford to wait a whole week, or like you're, or you're probably just wasting, honestly, you're probably wasting your money if you don't spend the time to do it daily, because a lot of these are waiting for you to give it inputs, r…”
Ibarra: Teams can only realistically add 1 to 1.5 agents monthly
“But I'd say in that math, you can really only add one slash one and a half agents per month max. Like, Otherwise, you're running in place because you can't really keep up with your current agents to add the new agent, so it doesn't actually, maybe it doesn't m…”
SaaStr's Monaco AI agent booked six meetings from 64 outreaches
“We got it up and running in basically a week and a half, and you can see here, like, in its first week on, you know, it reached out to 64 people, it booked six meetings instantly, Some of those are like tier one accounts for us, which is pretty dang good.”
Ibarra: No AI agent currently exists that can manage other agents
“There is no, like, again, crazy good agent that can manage all the other agents on its own today as of February. There's, that doesn't exist, right? Like, I wish there were, I don't know if, I think there will be, but there's not one today.”
Lemkin: SaaStr would cease using AI agents if Ibarra left
“If you left for some reason, we would have to cease using our agents. There's no plan B.”
Lemkin: Running AI agents with one operator is an existential risk
“You need to recruit somebody else on their team and divide and conquer and make sure you have two. You gotta have two, right? I mean, one's hard enough. There's a lot of, in the early days of AI agents, when we talk to folks, there's very few people at a lot o…”
Ibarra: AI agents act as relentless truth-tellers with complete operational visibility
“What does the agent care? The agent's just gonna tell me the truth. The agent has all the data more than I probably look at it, right? Like I look at it a lot. But it's got all, like, it knows where I'm behind, it knows my calendar, it knows all these things”
Ibarra: Continuous unvarnished feedback from AI agents can become demoralizing
“Because you can't keep up with them, sometimes I feel like it does cross a little bit of a line where I'm like, okay, dude, like, today just, like, sugarcoat it to me like you would a person. Today, I don't need the hard, like, I can't deal with hard truths ev…”
Ibarra: SaaStr has fixed under 10% of SaaStr.ai security audit issues
“We haven't even fixed, I would say, even 10% of the issues that came up in the audit because it became super fragile once we started, like, futzing with the code”
Lemkin: All third-party AI agents are less secure than Salesforce
“The honest truth is no agents, no third party agents are more secure than Salesforce. They're all less secure is the reality of it.”
Ibarra: Marketing agencies may not exist in a year
“Maybe marketing agencies won't exist in a year because of this”
Ibarra: Managing AI agents degrades your patience for managing human teams
“I do feel like now that I've flipped and switched to managing more agents, I think I've probably become slightly worse in managing the people that are, you know, that work with SaaS or especially leading up to SaaS Radial.”
Lemkin: Replit gives candidates $1,000 in credits to build interview projects
“In fact, what they do at replet itself now is they give you a massive amount of credits. I think they give you like a thousand dollars worth of credits in the interview and they say, build anything you want with a thousand bucks, come back to us and then we'll…”
Lemkin: AI outbound startup Monaco is booked two months for demos
“Like Monaco, we just rolled it out for outbound that we talked about. They're booked for two months, including weekends for demos.”
Lemkin: Artisan, Monaco, and Qualified AI agents each replaced two human roles
“Or take over all of like two people did for like each of Artisan and Monaco and qualified. They replaced like two, two people each, right? That's high ROI.”
Lemkin: Software offering mere productivity gains with mediocre ROI is dead
“Mediocre ROI is dead. Is just dead. Products that just give you productivity bounces that no one's going to buy that stuff today.”
Ibarra: Current AI agents cannot run end-to-end email marketing at scale
“These agents can't do that. I have not seen an agent successfully do email marketing at scale. The moment I find one, I'm using it that actually works. I've not found one that actually works. Like, not just writing all of these, just write email copy. They don…”