Every engineering team has a list of things they should do but do not.

Write tests first. Document architectural decisions. Review every commit. Generate runbooks from incidents.

We know these practices work. We skip them anyway.

The Economics of Impatience

The reason is economic, not technical.

Good engineering practices share a pattern:

  • High upfront cost
  • Distant payoff
  • Paid by humans under deadline pressure

When the deadline is tomorrow and the benefit is next quarter, the discipline loses.

We called this pragmatism. It was actually a tax on human patience.

What Changed

AI has infinite patience.

Writing comprehensive tests before code is psychologically brutal for humans. For a model, it is just another input format.

The cost of rigor dropped to near zero. The practices we abandoned are suddenly affordable.

The Short List

Here are the disciplines worth reconsidering:

Test-Driven Development A failing test is the clearest specification you can give an agent. TDD went from “thing we should do” to “best way to constrain AI behavior.”

Documented Decisions Architecture decision records were always right and rarely written. Now agents can generate them from code history. And existing ADRs make agents dramatically more effective.

Continuous Review Human attention for code review is finite and expensive. AI review on every commit is neither.

Living Documentation Documentation used to rot because nobody read it. Now agents read it on every task. Stale docs degrade your tools immediately, not eventually.

The Pattern

Every abandoned practice has the same structure:

  1. We knew it was correct
  2. The upfront cost was too high
  3. The payoff was too distant
  4. Human impatience killed it

AI removes the constraint.

The Catch

This only works with discipline.

AI as autocomplete makes you write more code, faster, with less correctness. AI with rigorous practices makes you write better code, verified, with guardrails.

The difference is whether you use structure to constrain the tool or let the tool generate without boundaries.

A 2025 study found experienced developers were 19% slower with AI assistance. The 2026 follow-up found that gap closing as developers learned disciplined patterns.

The tool is not automatically better. Disciplined use of the tool is better.

What To Do

Look at your team’s “we should really do this but we don’t” list.

Ask one question for each item: Was this abandoned because it was wrong, or because it was expensive in human patience?

The wrong ones, leave abandoned. The expensive-in-patience ones just got cheap.

The Inversion

For twenty years we said: “These practices are correct, but we cannot afford them.”

Now we can afford them.

The question is whether we remember they were correct.