Nov 1, 2024 · 44m · startup-ideas
Fire your team and hire AI employees?
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
Host Greg Isenberg and Lindy founder Florent Crivello present a hands-on masterclass on building, delegating, and scaling an autonomous workforce of AI employees across email, Slack, calendar, and recruitment workflows.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Greg holds 19.5% of the talking time here. How this is scored →
Greg as informed peer 3.6 Guest teaching 4.3 Guest disagreement 0.1 Greg pushing back 0.2
speaking balance: gold is Greg, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 1:17 Challenging narrow views of LLMs as mere copywriters
Flo rejects the mainstream perception that language models are only useful for simple copywriting, asserting they can replace human labor across thousands of parallel tasks.
Hardest push from Greg ▶ 11:21 Probing the mechanism behind time-sensitivity detectionGreg stops the demo to press Flo on exactly how the agent determines whether an incoming email is urgent.
Biggest teaching moment ▶ 5:19 Differentiating brittle step-based zaps from contextual agentsFlo breaks down why traditional automation tools like Zapier break when parsing unstructured HTML data, whereas LLM agents understand persistent context across multi-step tasks.
Greg holds their own ▶ 32:58 Synthesizing product features into a scalable recruiting business modelGreg leverages his entrepreneurial background to translate the sourcing demo into a full standalone business concept of an autonomous niche recruiting agency.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Greg as informed peer | Guest teaching | Guest disagreement | Greg pushing back | Why |
|---|---|---|---|---|---|---|
| The Paradigm Shift: From Text Generation to AI Workers | 3 | 5 | 1 | 1 | Greg sets the stage and brings up the common Twitter critique comparing agent workflows to Zapier automations. Flo explains the architectural difference, demonstrating stateful context and inline prompt logic in Lindy. | |
| Ad Break: The Startup Idea Bank | 1 | 3 | 0 | 0 | Following Greg's ad read for the Startup Idea Bank, Flo continues explaining the second key differentiator of agents: managing long-lived tasks across persistent Slack threads. | |
| Live Build: Constructing a Time-Sensitive Email Alert Agent | 3 | 4 | 0 | 1 | Greg asks practical onboarding questions regarding how a beginner should start and how the LLM determines urgency. Flo demonstrates a live build of a time-sensitive email filter. | |
| Upgrading the Agent: Interactive Slack Commands and Calendar Booking | 4 | 4 | 0 | 0 | Flo extends the agent to handle interactive Slack commands and calendar bookings. Greg compares the workflow to delegating to an Executive Assistant and appreciates the authentic unedited debugging process. | |
| Agent-to-Agent Delegation: Setting Up a Proposal Writer | 4 | 5 | 0 | 0 | Greg suggests applying the 'jobs to be done' framework to map out workflows. Flo illustrates advanced agent-to-agent delegation, describing it as object-oriented programming for AI agents. | |
| Ad Break: Boring Marketing | 4 | 3 | 0 | 0 | Greg provides an ad read and discusses agency proposal bottlenecks based on his experience. Flo diagnoses an internal token/display latency issue with the proposal generation agent. | |
| Live Use Case: Automated Executive Meeting Scheduler | 5 | 5 | 0 | 0 | Flo showcases meeting scheduling with automated calendar rescheduling, followed by an autonomous designer sourcing agent that navigates web portfolios and finds contact information. | |
| Autonomous Recruiting Businesses and Multi-Agent Sourcing | 4 | 5 | 0 | 0 | Greg recognizes the opportunity to build autonomous recruiting businesses. Flo connects the lead generator Lindy to an outreach recruiter agent in a live multi-agent chaining demo. | |
| Proposal Completion and Infinite Cloud Scaling for Support | 4 | 5 | 0 | 0 | Flo reviews the completed proposal document and explains how cloud-based parallel execution lets businesses scale support elastically during demand spikes like Black Friday. |
Statements from this episode (11)
Crivello: AI Employees Are a Thousand Times Better Than Humans
“I just think it's a thousand times better than a human. And it's just, it's gonna make your business more efficient. It's gonna make you grow faster. It's easier to spin up. You don't have to manage it. It scales infinitely.”
Crivello: Future Businesses Will Deploy 5,000 AI Employees Over a Weekend
“I think that the business of the future is basically you can set up a team of 5000 AI employees in like in a weekend.”
Crivello: Zapier workflows are brittle compared to context-aware AI agents
“In Zapier, each step is an island. It's isolated. Here, it's an agent. It's an agent that is aware of the entire context of everything it did, and so if you want to do this with Zapier, you're going to have to write, like, an HTML parser. You may not be able t…”
Crivello: Lindy agent steps retain context from all prior steps
“It's like every step is aware of what's the Kamala saying. It's like a, it's a way of everything that came before it and the context in which it exists. It's aware of every other step that existed before. We can see the output of every other step that existed …”
Crivello: AI agents differ from automations by autonomously sequencing actions
“The difference between these steps and an AI agent is like here, these steps, I am telling Lindy when to do what. With any AI agent, I'm telling Lindy, you figure it out. And so I can give her a prompt for like some guidance about what to figure out and so for…”
Crivello: Multi-agent AI architectures function like object-oriented programming
“It's really like, it's like object-oriented programming for agents.”
Crivello: AI agents need explicit exit conditions to avoid loops
“Cause sometimes AI agents get stuck in the loop. They go on like wild goose chases. Like don't try for too long. And so I gave it an off ramp here. We call those exit conditions.”
Crivello: LLMs easily defeat deliberate email obfuscation techniques
“Sometimes it's people, you know, they spell out like dot com, you know, D O T C O M in order for like, buts not to see them while jokes on them, like now it's AI all over it. Right. And so even if it's like that, like sometimes they write like, oh, my email is…”
Crivello: A new recruiting startup is being built on Lindy
“Just this morning, I was on a call with, there is a company right now that's being formed that is creating a recruiting company on top of Lindy.”
Crivello: Future service businesses will be one founder and AI armies
“But yeah, I mean, in the limit, it's going to be one dude. Well, there happened to be two, but it's going to be one dude and an army of AI recruiters. And that company can scale infinitely with compute, basically. It's just with LLMs.”
Crivello: Lindy scales AI agents elastically like serverless cloud computing
“With Lindy, he's just like, hey, if you receive an influx of 500 tickets in five minutes, you're going to have 500 Lindy. It's just like cloud, it's like a serverless cloud computing. It's just like boom, boom, boom, boom, boom. You just see, like, a bunch of …”