A paper proving that pairing is faster than pull requests.
Download the latest draft of the paper.
We have long discussed how pair programming, among several benefits, accelerates knowledge sharing, ensures collective code ownership, and increases overall developer productivity. These benefits outweigh the previously assumed decrease in velocity and high cost of to having two people do the work of one. We have shown that pairing does not negatively impact velocity, and in fact improves velocity significantly even in the short term. We complete more features faster by pairing because the code reviews happen in real time, rather than asynchronously.
When pairing, dependencies between stories in a backlog have less of an impact on team velocity. The impact of dependencies is much higher in a pull request workflow, meaning stories must be highly parallelizable to achieve the same productivity.
Still not convinced? Try the simulator.