GitHub’s head, Jay Parikh, no longer uses pull requests for every project. In a Latent Space podcast interview released on November 7, he said that for personal projects used to test an idea, he may turn the idea directly into code without creating a pull request or managing issues. His point was not that code review should disappear, but that GitHub’s collaboration process need not be mandatory for every kind of work.
The question has become more pressing as coding agents take on more development tasks. According to GitHub figures cited by InfoQ, agents drafted the first version of more than 40% of pull requests on the platform by September 2026, up from 17% in March. The number of agent-initiated pull requests grew roughly sixfold over that period.
Different projects, different controls
Parikh said GitHub could make a feature configurable if some users no longer find it useful or do not want it to be a required step. A personal experiment may benefit from a direct workflow; a large enterprise may still need code review, access controls and a record of changes. The distinction is between allowing teams to choose a process and removing safeguards altogether.
Parikh also described how GitHub wants to help developers choose models for different tasks. Copilot Harness connects models and tools and orchestrates development work, while Copilot Fusion is intended to select a model based on the task. That choice involves trade-offs: using more models can increase calls, time and AI credit consumption, and developers would retain a choice of models and tools.
GitHub presented a comparison from the Artificial Analysis Coding Agent Index in which Fusion Max scored within two points of Opus 5.5 Max. GitHub said Fusion Max cost about half as much per task and took about 40% less time. These are the cited benchmark figures, not a guarantee of the same result for every development task.
GitHub is also working on continuity between devices. According to the company, Session continuity lets developers follow the same cloud agent session through the GitHub Copilot app, VS Code, the command line and the web, while the agent continues working in the cloud.
Faster code generation does not remove maintenance
Parikh said GitHub is strengthening code review and security scanning as agents generate more code. At GitHub Universe, the company introduced Copilot code review Max to help find more serious and less obvious problems. A separate demonstration showed a system detecting and blocking a potential secret leak during a repository push.
The podcast also explored whether AI could make it practical for a team to replace open-source dependencies with its own code. The host posed a hypothetical project with 2,000 dependencies that kept just 20. Parikh cautioned that the team would then have to maintain the functionality previously supplied by the other 1,980 components. Rewriting code does not eliminate work such as fixing vulnerabilities, managing compatibility and assessing whether changes are reliable. Nor does owning a component eliminate its security risks. Parikh said he would rather work with the open-source community on software security than have every team move toward a closed approach.
Measuring value beyond token spending
Parikh said GitHub is building tools to help organizations see AI spending, consider alternative models for particular tasks and set usage limits. He agreed with the podcast host that spending alone does not show whether coding agents are worthwhile. He suggested looking at outcomes including revenue, cost savings, user engagement, retention and customer satisfaction; the host noted that a developer’s more immediate result may be delivering a feature.
InfoQ also notes that practical delivery depends on whether agent-generated code passes review, needs rework and runs reliably. More generated code does not necessarily mean a comparable gain in team productivity.
The shift is affecting GitHub’s infrastructure, too. Parikh said the platform is adjusting to increased automated development activity. In an article published during GitHub Universe 2026, he said GitHub expected to complete its migration to a multi-region Azure deployment in early 2027 while continuing to add computing, storage and network capacity. He said it remained unclear whether a developer might eventually manage dozens, hundreds or thousands of agents.