3 AI Agents That Actually Replaced Human Jobs | E2272
This episode explores three AI agents that are actually replacing human jobs and discusses the economic implications of agentic AI. The hosts and guests—including founders building Claude-based agents like Claw Chief and Sidecast—demonstrate how AI can autonomously perform real work (from chief-of-staff duties to podcast production to business ideation), while debating whether widespread job displacement will be offset by new opportunities or represent a net deflationary shock to labor markets.
Key takeaways
- • Anthropic's decision to end Claude subscriptions for third-party tools forces builders to adopt pay-as-you-go pricing; using Opus 4.6 in production agents now costs $100–$200/day, making builders choose between expensive proprietary models or open-source alternatives running locally.
- • Detail-oriented checklist discipline is what separates functional AI agents from broken ones—apply the same rigor you'd use to onboard humans (SOP, checklists, approval gates) to your agent workflows to prevent costly failures.
- • Memory and persistence are the critical missing pieces for agent reliability; agents running on ephemeral cloud instances (like EC2) lack continuity and forget context, so running them on dedicated hardware or local systems with persistent knowledge bases dramatically improves performance.
- • Autonomous multi-agent swarms can identify and solve real business problems at scale—tools like Henry scan Reddit, X, and other platforms to find unfulfilled customer needs, then autonomously build and launch businesses to solve them, with revenue-sharing or approval gates to control risk.
- • $2,000/month subscription tiers for frontier AI models are coming this year as compute costs force providers to price fairly; founders building serious agent infrastructure should budget $3,000–$6,000/month per production agent and assume pricing will only rise as demand increases.
- • Local open-source models (6 months behind frontier models) are becoming viable alternatives if you're willing to run inference locally on Mac Studios or DGX hardware; intelligent routing systems can load-balance queries across multiple models and failover gracefully, reducing dependence on expensive APIs.
Recommendations (8)
"This is why I talk about the Checklist Manifesto all the time. Make checklists, have your team do it, force them to print them out."
This Week in Startups · ▶ 3:32
"I just bought a new MacBook Pro because I had like a four-year-old Air. When you put 48 gigs of RAM on these things and this new M5 processor, it is like scary how much faster."
This Week in Startups · ▶ 5:59
"I've been using Hermes agent pretty significantly over the last few weeks. I do think it is a very strong tool. It's worth using side by side with OpenClaw."
Alex Finn · ▶ 1:08:43
More from these creators
AI Rebuilt Every YC W26 Startup. Should Founders Be Scared? | E2271
Venture Roundtable: SpaceX IPO, AI's PR Crisis, and the Defense Tech Bubble | E2270
The 5-Step Framework for AI Agents That Improve While You Sleep | E2269
The $60 billion resource hiding in space, and the start trying to mine it | E2268
This Bittensor Subnet Could Cut Drug Discovery Costs in HALF | E2267
32 Million Dollar Fraud? Delve Scandal | E2266