KM4BI KM4BI

You’re Not Bad at Prompting. You’re Bad at Defining the Work.

Nate B. Jones argues that the key skill is no longer isolated prompting, but framing AI-agent work like a senior-partner assignment.

Nate B. Jones argues that “prompt engineering” is no longer the central conversation. It still matters, but it has become table stakes. The real leverage now comes from defining the work: asking better questions, stating a thesis, setting boundaries, and giving the agent the right evidence to reason across.

What changes with stronger agents

The core mental-model shift is to stop treating advanced agents like junior assistants that need every micro-step. For complex knowledge work, Nate suggests treating them more like senior partners: capable of exploration, synthesis, and pushback if the human gives them a clear direction.

Three practical principles

  1. Give the agent a center of intent, like the middle of a flashlight beam: a thesis or angle, without closing down exploration.
  2. Ask questions that force the agent to reason about what “good” looks like, instead of relying only on narrow eval-style criteria.
  3. Explicitly connect hard data, files, transcripts, and softer opinions so the AI does not over-index on one artifact.

Why teams should care

This is especially relevant for product, marketing, and strategy work where the source material is messy and multi-format. Output quality depends less on magic wording and more on framing: which data to inspect, which hypothesis to test, which areas to ignore, and what kind of synthesis would actually support a decision.

The broader point is that working well with AI increasingly resembles good management: delegate the problem with context, direction, and standards, then ask for a strong thesis rather than a flat recap.

Source