Jul 24, 2026 · 44m · startup-ideas
Most Valuable Skill of 2026: Managing AI Agents
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 veteran operator Ryan Carson explore how knowledge workers can transition into elite engineering managers of autonomous AI agent fleets. The conversation provides an actionable blueprint covering cloud-based software factories, mobile orchestration, autonomous testing loops, and token cost optimization.
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 17.9% of the talking time here. How this is scored →
Greg as informed peer 3.0 Guest teaching 5.0 Guest disagreement 2.0 Greg pushing back 0.9
speaking balance: gold is Greg, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 12:56 Local dev maxis called cavemen
The guest bluntly rejects the conventional developer ethos, claiming anyone developing locally instead of using cloud VMs is a caveman holding back output by ten times.
Hardest push from Greg ▶ 13:15 Host relays local-first counterargumentsThe host confronts the guest's stance by highlighting prevalent industry pushback from local maxis on social media who consider cloud VM users amateurish.
Biggest teaching moment ▶ 32:42 Self-improving agent evaluation architectureThe guest details how to construct an automated evaluation pipeline where daily conversations are graded against a rubric to trigger autonomous bug-fix PRs.
Greg holds their own ▶ 39:17 Mortgage broker model routing analogyThe host reframes the value proposition of independent agent labs by comparing them to mortgage brokers finding the best model rates, earning immediate praise from the guest.
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 Shift to Becoming an Agent Operator | 1 | 4 | 1 | 0 | The host sets up the premise of managing AI agent teams and openly admits he conducts zero percent of his work on mobile. The guest establishes the mindset shift required for founders to act as engineering managers scaling solo operations. | |
| Ryan Carson's Hardware and Software Command Center | 1 | 5 | 1 | 0 | The guest gives a detailed operational walkthrough of his eight-screen setup, voice-to-text integration, and production key separation in 1Password. The host remains in pure listening mode as the guest presents tactical security protocols. | |
| Cloud Agents Versus Local Development Environments | 3 | 6 | 4 | 2 | The guest aggressively dismisses local development proponents as 'cavemen' who hold back output. The host challenges this view by bringing up the pervasive sentiment from local maxis on X who dismiss cloud environments as amateurish. | |
| Operating Cadence and High-Stakes Decision Making | 4 | 5 | 2 | 1 | The guest shows live GitHub data demonstrating 20 to 40 PRs shipped per day and introduces the 25-minute decision cadence. The host draws meaningful parallels between modern agent management and classic pre-AI organizational executive functions. | |
| Architecting Autonomous Testing and Self-Improvement Loops | 3 | 7 | 1 | 1 | The guest educates the audience on multi-tier automations, from browser QA playbooks to automated rubric evaluations of paralegal chats. The host asks sharp questions clarifying the boundary between functional bug QA and UX design review. | |
| Token Economics, Model Routing, and Independent Labs | 5 | 6 | 5 | 2 | The guest passionately criticizes founders who lock their software workflows into single frontier labs like Anthropic or OpenAI. The host intervenes to clarify that indie labs still route to frontier models and articulates the mortgage broker analogy. | |
| Building in Public and Final Takeaways | 4 | 2 | 0 | 0 | The host provides a concrete case study of Sahil Bloom to illustrate the mechanics and compounding benefits of learning in public on X. The guest completely concurs as they wrap up key takeaways. |
Statements from this episode (15)
Carson: Knowledge workers across all roles are becoming AI agent managers
“Essentially you are a manager of agents now.
So no matter what you used to do, whether it was a people manager, an IC, you are going to become a manager of agents now, and you need to be the best in the world.”
Carson does over half of his startup work on his phone
“You know, I do almost 50%, probably more of my work from my phone.”
Carson: Never give AI agents direct production write keys
“It's very important to keep your keys safe, secure, and separated from the agent, right? So I've got all my prod write keys in one password. I do not give my agents a prod write keys. So production writing is dangerous, right? And your agents will do something…”
Carson: Devin is one of the best software factories in tech
“I think it's one of the best software factories in the industry. It's not cheap. But it's good.”
Carson: Managing AI agents makes developers more technical, not less
“The more you become a better agent manager actually the more technical you become.”
Carson runs up to ten cloud coding agents simultaneously without collisions
“So I often have, you know, at least five cloud agents working at once, often 10, and there's no risk that the code is going to collide.”
Carson: Engineers coding locally are cavemen shipping 10x less code
“So I think if you are working locally, I honestly think you are a caveman. Like, I think you're, you are holding yourself back, and you are shipping 10 X less than you could be, and it is not smart. No matter if you think the cool people work locally and all t…”
Carson: The AI agent era requires working more, not less
“And in in order to survive in this new world, you're going to work a lot more, not a lot less.”
Carson averages 22 to 25 pull requests daily using AI agents
“And you can see what, you know, the average here is, is sort of this 22 to 25 PRs a day, right? And sometimes, you know, 40 a day.”
Carson spends $60 per run for automated Devin browser tests
“And I have a it's called end-to-end sign-up test, and that runs three times a week because it is expensive. It's probably 60 bucks in tokens because it's doing a lot of, it's doing a lot of browser testing”
Carson ships three PRs daily from an autonomous AI grading loop
“Every day I have an automation that looks at these chats and then grades them on a rubric. And so you, and again, just talk to your agent about this. You pick the most important part of your app that you want to self-improve and say, here's how you agent judge…”
Carson spent $20,000 on AI tokens in a single month
“So last month I spent probably 20 grand in, in tokens. Which is just too much. Like, it's not viable.”
Carson: AI engineering token costs will hit $5,000 monthly per employee
“I think all of us are in a place where we, we're getting to the spot where it's realizing, okay, for real engineering work per employee, you're looking at probably five grand a month. Like, that's probably where we're gonna shake out here.”
Carson warns against building workflows on frontier models to avoid lock-in
“Like, if you are, if your whole engineering you know, motion is happening inside of clock code or inside of codex, What are you doing? Like, because they are not incentivized to make it reasonable for you long term, right? They're going to lock you into their …”
Carson: AI agents will write, review, and ship 100% of code
“You will be using a software factory. Like, and what I mean is, the agents are going to be writing a hundred percent of your code, reviewing a hundred percent of your code, shipping a hundred percent of your code. Like, that's where we're going.”