tomte

← All series

The Tools That Shape Us

A 6-part series · best read from the top, in order.

  1. You Can Outsource Thinking, But Not Understanding

    The more you hand to an agent, the more you get done — and the more your own grasp of the work stays exactly where it was. On the surface that delegates cleanly (the execution of thinking) versus the depth that does not (defining the problem, deciding, understanding), and why Hesse said the same thing a century ago.

  2. Documents That Agents Read and Write: Notion's Blocks and the Problem of Verifiable Truth

    As AI agents start reading and writing documents on our behalf, Notion's block model is being put to the test. Why a markdown-and-git workflow suits agents, why a Notion doc's mutability and lack of verification make it a weak source of truth, and how Notion answers back with MCP, Workers, and an agent hub — plus where the real battle line falls.

  3. The Tools That Shape Us: Notion's Intellectual Lineage and the Dream of 'Software Lego'

    See Notion as just a pretty all-in-one document app and you miss half of it. Trace the lineage that runs from Marshall McLuhan's theory of tools through Douglas Engelbart's intelligence augmentation to Alan Kay's Dynabook, and you find what founder Ivan Zhao was really after: a tool for making tools, and the 'software Lego' behind it. With the Kyoto reboot that saved the company — and the criticism that shadows the ambition.

  4. Smarts You Can't Control Isn't Smarts: From Multi-Agent Back to a Single Agent

    Why walk a multi-agent setup back to a single agent? A year of hand-building a personal AI assistant, then moving it off messengers like Telegram and Slack onto a web app I built myself — read through shared context, the cost of supervising agents, lock-in, and knowledge management, with Cognition's 'Don't Build Multi-Agents' and Anthropic's multi-agent research alongside.

  5. Writing With an Agent: How to Reach Further, and the One Line I Still Have to Write

    Writing with an AI agent lets you go faster and reach further — translation, search, accessibility. Its real uses, the objection that writing is thinking (Paul Graham, Ted Chiang, Cory Doctorow), and the supervision it demands: I can only write what I understand. Authorship is not about who typed it, but who takes responsibility.

  6. Loop Engineering: When You Build the Loop, Not the Prompt — and Where Verification Goes

    What loop engineering is. It means replacing the version of you that kept prompting an agent with an autonomous loop that does the prompting instead. A first-person account of running this pattern — which broke out in June 2026 — not just in coding but across investing, knowledge curation, and document review, the two places it always collapses (the verifier and the stop), and why human verification never disappears; it only moves.