Agentic Engineering: Avoiding the Pitfalls of Pull Requests
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisis ton rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Behaviors to Avoid in Agentic Engineering
In the field of agentic engineering, certain behaviors are considered anti-patterns, meaning practices to avoid in order to ensure effective and productive collaboration.
Submission of Unreviewed Code: A Common Trap
One of the most problematic behaviors is submitting pull requests (PR) without prior review. This phenomenon is not only frustrating but also imposes unnecessary workload on collaborators.
When you submit a PR containing hundreds, or even thousands, of lines of code generated by an agent, without ensuring they function correctly, you transfer the responsibility of review to others. They could just as well have used an agent to generate that code themselves. What added value do you bring then?
Responsibility for the Initial Review
Before submitting code for review, it is crucial to ensure that it is ready to be examined by others. The initial review is your responsibility and should not be delegated.
A well-prepared pull request in the context of agentic engineering must meet several criteria:
-
The code must function correctly, and you must be convinced of this. Your role is to deliver functional code.
-
The changes should be small enough to allow for effective review without overwhelming the reviewer. It is better to submit several small PRs rather than one large one. Using a coding agent to break the code into separate commits can facilitate Git manipulation management.
-
The PR should include additional context to explain the change. What is the overall goal of this change? Linking the PR to relevant issues or specifications is very helpful here.
Importance of Reviewing PR Descriptions
Agents can write PR descriptions that seem convincing, but it is essential to review them. It is impolite to expect someone else to read a text that you have not validated yourself.
Proof of Review and Testing
Given how easy it is to submit unreviewed code, it is advisable to include proof that you have done this extra work. This can take the form of notes on how you tested the code manually, comments on specific implementation choices, or even screenshots and videos showing the functionality in action. These elements greatly contribute to demonstrating that the reviewer's time will not be wasted exploring the details.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.