Against the Personal Agents Theory of Everything

Author: Nathan Baschez URL: https://www.gettheleverage.com/p/against-the-personal-agents-theory One-line: Personal agents are good, but shared “factories” — AI systems that are specific, scrutinized, and deployed team-wide — are the inevitable next paradigm because they unlock specialization and gains from trade.

Key claims

• The “factory” is the next AI paradigm after personal agents. A factory is defined as a shared AI system designed to handle specific types of recurring tasks within an organization — distinct from a personal agent in three key attributes: it is Shared (multiple people use it), Specific (optimized for certain tasks, not general-purpose), and Scrutinized (visible to owners who can observe and fix failures).
• 99% of companies today operate in “DumboCo” mode — each employee tinkering with their own agent across 5–10 task types, averaging about 3.7 hours/month per task type on the learning curve. BigBrain concentrates those hours: 1–2 tasks per person at 18.3 hours/month, yielding roughly a 5× deeper optimization per task.
• The economic logic is Adam Smith’s specialization argument: factories enable gains from trade at the team level — the CFO who is an AI nerd can optimize the “monthly close” factory once and have the whole team benefit, without anyone else needing to understand the sausage.
• Skills are not factories because they lack a shared runtime environment (different configurations on different machines) and are not scrutinized — skill runs aren’t easily observable by the skill owner, so the feedback loop is broken.
• Vendor factory markets are maturing: Sierra, Fin, and Decagon sell customer-support factories; Cognition and Cursor sell software factories; Amplitude is pivoting to a “product factory” (autonomously identify, build, and ship improvements from metrics); XBOW, Novee, and Snyk are building cybersecurity factories.
• Factories benefit from evals — a checklist of tasks the factory should handle, plus a rubric for judging success — which are much easier to create and use than most builders think, and enable systematic testing when updating factory prompts.
• The minimum viable factory is a Notion database: status column (todo/doing/done), an agent triggered on card creation/status change, instructions hosted in Notion for versioning and collaboration.

Notable quotes

“Factory (noun) — a shared AI system designed to handle specific types of tasks within an organization.”

“It is as simple as Adam Smith: factories are key to specialization and gains from trade — the root of all productivity and prosperity.”

“BigBrain has the same budget of 5 to 50 hours of ‘AI tinkering time’ per person but concentrates it on just 1–2 tasks per person. This yields 18.3 hours of optimization time per task per month, which means going 5x deeper into the learning curve on each task.”

“Skills can be sort of factory-like, in that they help agents perform specific tasks, and the prompts are shared. But the actual runtime environment is not shared, so people running a skill with different agents on different machines may run into configuration issues or other hiccups. And, crucially, skill runs are not easily scrutinized by skill owners. So the feedback loop is broken.”

“For example, the HR team could have a factory to answer questions about benefits, and to the user it just feels like they asked their AI a question and quickly got a good answer.”

“Factories — or whatever they end up being called — are AI systems that are shared, specific, and scrutinized. They are inevitable because they are economically rational. They enable specialization and gains from trade.”

How it connects

• Artificial Intelligence — the broader paradigm shift within which personal agents, factories, and the future of work sit
• Adam Smith — the economic theory of specialization and gains from trade underpins the entire factory argument
• Division of Labor — the factory model is essentially a rediscovery of the division of labor inside organizations
• Agentic AI — the current moment of personal agents that factories build on top of
• MCP — the Model Context Protocol that allows agents to operate tools like Lovable; the intro example involves Claude operating Lovable via MCP
• Notion — the recommended minimum viable factory substrate (database + agent trigger + versioned prompts)