Cheap generation raises the bar

When first drafts are cheap, the scarce skill is not producing more options. It is selecting, shaping, refusing, and calibrating.

Agents make average work abundant. They can produce a polished memo, a plausible strategy, a tidy refactor, or a convincing summary on demand. That does not mean the work is good. It means the surface area of mediocre work has become smoother.

Taste is how the individual operator prevents scaled output from becoming scaled mediocrity.

Taste is an explainable standard

Taste is not "I like it." That is preference.

Operational taste is a standard you can explain and enforce. It notices when an output is too generic, too verbose, too clever, too fragile, too disconnected from the user, too confident for the evidence, or too expensive to maintain.

In technical work, taste shows up as consequence-modeling: knowing what will age badly, create operational burden, confuse future maintainers, or fail under real usage. In content and strategy work, it shows up as specificity, courage to cut, source fidelity, and a feel for what the reader actually needs.

The taste review rubric

When reviewing agent output, inspect for:

  • Source fidelity: Are claims grounded in the provided evidence, or did the agent fill gaps with generic knowledge?
  • Specificity: Could this only have been written for this situation, or would it fit any company, team, or person?
  • Consequence awareness: Does it name second-order effects, risks, maintenance cost, or stakeholder impact?
  • Simplicity: Is the solution as small as the problem allows?
  • Fit: Does it match the audience, timing, decision, and operating context?
  • Coherence: Does it preserve prior decisions instead of re-opening settled questions?
  • Cut discipline: What should be removed even though it is well-written?

A five-page agent output that contains two useful paragraphs should become two paragraphs. The operator who cannot cut is not scaling taste. They are scaling accumulation.

Failure mode

The failure mode is not obviously terrible work. That is easy to reject.

The failure mode is plausible, polished, low-specificity work that slowly becomes normal. The team starts accepting generic recommendations. The writing gets smoother and less true. The code gets more elaborate and less understandable. The strategy becomes a collage of reasonable sentences with no hard choice inside.

Taste drift is cultural debt for a team of one.

Taste artifacts

If taste matters, make it visible:

  • keep examples of excellent prior work;
  • keep examples of rejected output and why it failed;
  • define review rubrics before generation;
  • write "do not sound like this" notes;
  • maintain a short list of standards that are never optional.

Agents can imitate a bar only if the operator has one.