What we're learning in the field.

Field Notes from the Nth AI founding team on enterprise context, semantic models, ontologies, agent engineering, and governed AI.

  • It's 8:55. The board meets at 9:00.

    Shane Heddy · 2026-11-10 · Founders & Field

    Three systems, three answers to the CFO's question, five minutes before the board meets. Shane Heddy on why AI answers aren't trusted yet, what data leaders have already tried, and how to get to one number everyone accepts.

  • All they wanted was AI to draft partner emails.

    Leon Papkoff · 2026-10-21 · Founders & Field

    A request to draft partner emails reveals what enterprise AI needs underneath: a governed data foundation and shared business meaning. Leon Papkoff on moving from a plain-English goal to context that teams can build AI workflows on.

  • Two teams. One dataset. Two buying decisions.

    Shane Heddy · 2026-10-14 · Founders & Field

    One AI pilot became an opportunity for two teams, each with its own budget and questions for the same data. Shane Heddy on why shared data does not mean a shared buying decision, and how to plan for the second use case.

  • The smartest thing on my desk every morning

    Leon Papkoff · 2026-08-22 · Enterprise Context

    Every morning, without fail, it puts the most important decision of my day on my desk, with the number and the confidence already attached.

  • What a confidence score actually means

    Connie Yang · 2026-07-14 · Trust & Governance

    An AI answer with no confidence attached is just an opinion in a nice font. In our last two posts we argued that trusted definitions make the number right, and depth plus reach surface it on time. The piece we have not explained is the score itself: the small number next to a recommendation that tells you how much to trust it. It is the hardest part to get right, and the reason a system either earns a place in the decision or gets switched off within a month.

  • Your AI copilot is confidently wrong. Faster.

    Connie Yang · 2026-07-06 · Trust & Governance

    Every enterprise now has a chat-with-your-data tool, and most executives trust it less than the ticket queue it was supposed to replace. The problem was never the model. It is that AI got bolted on top of data nobody governed, so it inherits every disagreement in the stack and states it with total confidence. The fix is not a smarter copilot. It is a trusted definitions layer underneath it.

  • Harness engineering for the enterprise: tools, sandboxes, and the validator that has a name.

    Connie Yang · 2026-06-24 · Agent Engineering

    The model gets the headlines. The harness decides whether it ships. Harness engineering is the layer one step below the loop: the tools an agent can reach, the sandbox it runs inside, and the checks that grade its work. On a laptop the final check is a linter. In the enterprise it has a name, a role, and a signature. Here is what changes when the harness has to be governed.

  • Loop engineering came for the enterprise. The hard part is the governed loop.

    Connie Yang · 2026-06-23 · Agent Engineering

    In June 2026 the agent world stopped talking about prompts and started talking about loops. The insight is real, and it is the fourth layer of leverage after prompt, context, and harness engineering. But a loop that runs while you sleep on a laptop is not the same as a loop that an auditor can read on Monday. Here is what changes when loop engineering meets the enterprise.

  • What four research talks reveal about where enterprise AI ships next

    Connie Yang · 2026-05-21 · Agent Engineering

    Four talks in one night. RL training is now predictable. Not every model self-improves. Outcome-only evaluation breaks the moment a domain stops being verifiable. And self-improving agents are shipping into production anyway. The throughline is the entire enterprise AI roadmap for the next eighteen months.