Lab notes

Lab notes

Shorter pieces: a concept explained, a distinction made, a comparison drawn. Where an engineering article argues a design, a lab note explains the thing the design is about.

  1. Designing human-in-the-loop AI systems

    Adding a human is easy. Adding one whose attention is actually worth something is the design problem.

    5 min
  2. How AI agents will interact with business software

    There are four routes into a business system, and they are strictly ordered by preference. Most disappointment comes from taking the wrong one.

    5 min
  3. MCP and the future of AI tool integration

    MCP standardises how a tool is described and invoked. It does not standardise who may invoke it - and in a multi-tenant product that is most of the work.

    5 min
  4. Shared context across voice, chat and email

    Synchronous channels differ by seconds. An email thread differs by weeks, and almost every assumption about context changes with it.

    5 min
  5. Why AI agents need observability

    An agent can return 200 on every request and fail every conversation. Nothing in a standard dashboard will show you that.

    4 min
  6. From prototype to production AI agent

    The demo is roughly ten per cent of the work, and it is not the ten per cent that determines whether the thing survives contact with customers.

    6 min
  7. The architecture of an omnichannel AI agent

    Draw the line between adapter and agent in the wrong place and you have not built one agent - you have built four with a shared logo.

    5 min
  8. How computer use agents work

    The agent sees a page or a screen, proposes an action, and checks what changed. The hard part is everything around that.

    5 min
  9. Voice AI and traditional IVR

    An IVR is a decision tree the caller walks. A voice agent is a loop that walks itself. Both properties have consequences.

    5 min
  10. How AI agents manage context

    The context window is assembled fresh every turn. What goes in it is a design decision, not a side effect.

    5 min
  11. Building AI agents that can use tools

    Tool calling looks like a model capability. In practice it is an interface design problem with a model at one end.

    5 min
  12. From chatbots to autonomous agents

    Each generation moved one decision from the author to the system. That is the whole story, and it explains what breaks next.

    5 min
  13. Why latency matters in voice agents

    In text, a slow answer is a slow answer. On a phone call, it is a signal - and callers read it as confusion or a dropped line.

    5 min
  14. How realtime voice AI works

    Six components, one hard constraint: none of them may wait for the previous one to finish.

    6 min
  15. What makes an AI agent an agent?

    The word has been stretched to cover almost anything with a model in it. Here is the distinction we find load-bearing.

    5 min