Although Pull Requests (PRs), a.k.a. Merge Requests, have become the standard practice in modern software development, that may turn out to be an anti-pattern for engineering teams, as exposed in the recent heavy AI activities. The original intention of PRs was designed for foreign contributions, widely pioneered in Linux kernel development as patch submission, and later formalized in GitHub’s core concept.
Intention Declaration and Intention Reviews are an obvious replacement, which ideally fits rapid AI workflows. Instead of looking at the results after intensive work, the control focus shifts to preliminary analysis and endorsement of planned intensive work.
What Is an Intention Declaration and Review?
Nothing is new under the sun. That is an industrial and leadership practice when team members preliminarily analyze a problem or a task, and declare their intention, so their colleagues and supervisors can promptly respond with known features, issues, gaps, and suggestions. Most importantly, it enables a clear mind and grows abstract vision skills. One way or another, it is included in open source project contribution guidelines.
The Review part means pre-approval is needed before further work is done. The balance between just declarative and essential approval should be defined in the project risk management policy and the cost of going the wrong way.
Both declaration and review can be skipped for obvious small changes along with any following PR.
Assumed Benefits of PRs Compared to IDs/IRs
Grok was kind enough to provide the key benefits of PRs for us to compare below.
1. Improvement of Code Quality. Design is improved based on intention review, as the actual code is better controlled by linting, static analysis, and unit tests. Use of PRs for that is pedantic hell.
2. Knowledge Sharing and Team Learning. This is absolutely unrelated to PRs, as those do not involve the whole team and only fragment discussions. Proper retrospective meetings in the full group reviewing and discussing exact commits and related issues are much more productive.
3. Better Collaboration. No change in that compared to IDs/IRs, but there is much less drama with IRs in case of discarding the work done.
4. Early Detection of Issues. PRs are MUCH more delayed in detection of issues compared to IDs/IRs and automated testing.
5. Enhanced Traceability and Accountability. That’s what task trackers are for, but not PRs.
6. Risk Reduction. Proper prior design plan, automated tests, and actual post-integration code review of factual changes are the real practices of risk reduction. PRs tend to shift accountability to the reviewer due to the common misunderstanding of PRs’ concept.
7. Documentation and Context. Absolutely unrelated. Most teams have only source code in PRs. Proper Agile Documentation is the answer.
8. Workflow and Process Enforcement. This makes sense only when the team is not trusted and is not empowered. Most of the other concerns are implementable by tools and automation.
9. Psychological and Cultural Benefits. Review benefits are still present with IDs/IRs, but the drama part of defending significant coding effort done gets eliminated.
10. Scalability and Maintainability. AI agents have proven that PRs are not scalable. Delayed design review feedback does not aid maintainability either.
11. Security and Compliance Advantages. Security and compliance must be based on actual deliverables, but not on any intermediate state. Many aspects are better handled with tools in CI.
12. Productivity and Velocity Gains (Long Term). As exposed by AI agents, neither of the goals is fulfilled by PRs. The same problem exists in biological teams, but less visible. Only top performers, who do not rely on reviewers’ competence, really understand that.
Conclusions
Intention Declarations cover the purpose of Pull Requests, but eliminate associated roadblocks and bottlenecks of team performance. Declaration of Intentions is basically what Agile Documentation is, with not much extra effort needed for maintaining a living knowledge base.
It also opens the route to canonical Continuous Integration in a modern AI agentic contribution model, eliminating waste of branching and synchronization during the review period. Post-commit peer reviews eliminate any remaining concern.
Pre-commit peer code review flow should be considered a punishment and distrust for low-quality work. Of course, new team members and AI agents require prior alignment with the project and are supposed to use PRs/MRs, but only for the onboarding period.
Connect with me on LinkedIn.