Getting real work done · for software engineers
Finishing real engineering work with AI
Generated code is the easy part. Engineering work counts when it is merged, deployed, and working for users. Many engineers produce code quickly with AI and then spend days on review rounds, missed edge cases, and rework.
Getting real work done means using AI to reach a finished change. That change meets the requirement, passes review, includes tests and documentation, and behaves well in production.
Where it comes up
A design doc for a new service
The doc must let reviewers approve or reject the approach. AI can structure it, list alternatives, and find gaps. You supply the requirements, the constraints, and the choice between alternatives.
A complete pull request
Code, tests, migration, updated documentation, and a description that explains why the change exists. AI can draft each part. You make sure the parts agree with each other and with the ticket.
A runbook for an on-call alert
An engineer woken at night must be able to follow it. AI can turn your notes into clear steps. You confirm each command and dashboard link works before the runbook is published.
What it looks like at each band
E1 · Foundational
Produces first versions of code with AI, then finishes them mostly by hand. Pull requests sometimes need several review rounds for gaps AI introduced. Final quality depends more on manual effort than on how AI was used.
E2 · Proficient
Uses AI through to a change that passes review with ordinary feedback on routine work. Includes tests and a clear description. Knows which parts of a change to write personally, such as security-sensitive logic.
E3 · Distinguished
Delivers complete work on complex jobs, such as a cross-service change or a design doc with real alternatives. Uses AI for the parts it does well and personal judgment for architecture and trade-offs. Reviewers find few gaps.
E4 · Exceptional
Consistently delivers work that raises the team’s standard, including under incident pressure. Is deliberate about where AI helps and where it does not. Can show others how a finished change was built and checked.
Common mistakes
Stopping when the code compiles
Compiling code is not finished work. Judge the change against the ticket, the tests, the documentation, and how it behaves when a dependency fails.
Letting AI choose the architecture
Ask for a design without your own requirements and you get a generic answer that fits no real system. Use AI to test and extend your design, not to replace your judgment.
A pull request description nobody wrote
A generated description often lists what changed and leaves out why. Reviewers need the reason, the risk, and how you tested it. Write those parts yourself.
A worked example
An incident write-up for a two-hour outage caused by an expired certificate, due to the team the next morning.
Paste the incident chat channel into an AI tool, ask for a write-up, and send the result to the team.
The draft may state a root cause nobody confirmed, give times that do not match the alert logs, and propose generic follow-ups.
Pull the alert times, deploy history, and certificate details into a short evidence list. Ask AI to draft the timeline and impact sections in the team’s template from that list only. Write the root cause and the follow-up actions yourself, each with an owner. Check every time and number against the evidence, then read the document once as a reviewer would.
The facts come from logs and the engineer’s analysis. AI structures the document and saves writing time. The team receives a write-up it can act on and trust.
The full study guide for this skill, for every role, is Getting real work done. The other five skills for software engineers are on the AI skills for software engineers page. A one-page summary is in the AI-assisted work checklist.
See where you stand
The free check grades getting real work done and the other five skills in about fifteen minutes. Nobody sees the grade but you.
Then build the skills and prove it. The three steps for software engineers list the courses and templates that fit the work.