Judgment

“Good judgment comes from experience, and experience comes from bad judgment.” — Rita Mae Brown

Context

This is a strategic pattern. You have requirements, constraints, and a field of tradeoffs. Many decisions in software can’t be resolved by looking up the answer or running a calculation. They require weighing incomplete evidence, anticipating consequences, and choosing a course of action that’s good enough to move forward, even when certainty is impossible.

That capacity is judgment. It operates in the gap between what the rules cover and what the situation demands.

In agentic coding, judgment matters in a specific way: the human must supply it. AI agents can generate options, evaluate criteria, and follow instructions with precision. But deciding which criteria matter, when to deviate from convention, and whether an unexpected result is acceptable? Those calls require human judgment.

Problem

Many of the most consequential decisions in software have no objectively correct answer. Should you refactor now or ship first? Should you use a proven but dated technology or a newer but less battle-tested one? Should you invest in testing this edge case or accept the risk?

These questions can’t be resolved by gathering more data alone. At some point, someone must decide. How do you make good decisions when the information is incomplete and the consequences are uncertain?

Forces

• You want certainty, but many decisions must be made before all the facts are in.
• You want speed, but hasty decisions lead to costly mistakes.
• Rules and frameworks help, but every interesting problem has aspects the rules don’t cover.
• Delegating decisions to an AI agent is tempting, but the agent lacks the context of your business, your users, and your team.
• Experience helps, but past experience can mislead when the situation has changed.

Solution

Develop judgment as a practice, not a talent. Good judgment isn’t a gift some people have and others lack. It’s built through deliberate cycles of deciding, observing consequences, and updating your mental models.

Several habits support better judgment:

Name your assumptions. Before deciding, write down what you believe to be true and what you’re uncertain about. This makes your reasoning visible and auditable, to yourself and to others.

Seek disconfirming evidence. The most common judgment failure is confirmation bias: seeing only the evidence that supports the decision you already prefer. Actively look for reasons your preferred option might be wrong.

Decide at the right altitude. Some decisions are strategic (what to build) and deserve careful deliberation. Others are tactical (which variable name to use) and should be made quickly. Matching effort to importance is itself an act of judgment.

Make decisions reversible when possible. If you can structure a choice so that it is cheap to undo, you reduce the cost of being wrong. This lets you move faster without recklessness.

When working with an AI agent, reserve judgment calls for yourself. Use the agent to generate options, explore consequences, and surface information. But make the final call on decisions that involve values, priorities, or uncertain outcomes.

How It Plays Out

A developer is building a feature and the agent suggests two architectures: one simpler but limiting future extension, the other more flexible but complex today. The agent can lay out the tradeoffs, but it can’t know that the team is under deadline pressure, that the product direction is uncertain, or that the simpler approach fits the team’s current skill level. The developer chooses the simpler path, noting the conditions under which they’d revisit the decision.

Consequences

Good judgment leads to decisions that hold up over time, even when they were made with incomplete information. It builds trust within teams and reduces the cost of uncertainty.

The cost is that judgment takes time to develop and is hard to transfer. You can’t write a checklist for judgment the way you can for acceptance criteria. You also can’t fully automate it, which means that as AI agents take over more execution work, the human’s role shifts toward judgment and taste.

Judgment can also be wrong. The remedy isn’t to avoid judgment but to create conditions where wrong judgments are detected early and corrected cheaply.

Related Patterns

• Uses: Tradeoff — judgment is how you evaluate and choose among tradeoffs.
• Uses: Constraint — judgment operates within the space constraints leave open.
• Complements: Taste — judgment chooses well; taste recognizes quality.
• Enables: Acceptance Criteria — good judgment determines which criteria matter.
• Contrasts with: Constraint — constraints are fixed; judgment is adaptive.