# Measuring Productivity and Quality Honestly — Coding Agents & AI-Assisted Development

Source: https://www.geekswithgeeks.com/en/coding-agents/m-value

> Look beyond lines of code at lead time, rework and defects.

## Faster typing is not faster delivery

It is easy to feel faster with an agent and hard to prove it. Lines of code or number of PRs reward volume, not value, and can hide extra review, rework and bugs. Better measures: **lead time** from task start to merged change, **review time and iterations**, **rework** (how much is rewritten after merge), **defect and incident rates**, **test coverage and CI stability**, and developer-reported **cognitive load and satisfaction**. Compare against a **baseline** (same kinds of tasks before adoption, or a control group), and look at *which* tasks benefit: boilerplate, tests, migrations and exploration often gain a lot, while novel design and subtle debugging may not. Watch for hidden costs: reviewer overload from large diffs, skill atrophy in juniors, and growth of code nobody fully understands.

## Measure before and after

Without a baseline you cannot tell whether the agent helped or only felt faster.

**Quiz:** Which measure better reflects real value than "lines of code written"?

- [ ] Number of prompts sent
- [ ] Number of characters typed
- [x] Lead time to a merged change together with defect and rework rates
- [ ] Time spent in the chat window

*Answer:* Lead time to a merged change together with defect and rework rates. Value shows up in delivery speed and quality, not in output volume.