Key Takeaways

  • GitHub pull requests create an artificial bottleneck that cancels out the speed gains of AI coding agents.
  • Amp engineers rarely use branches; they review artifacts like automated video walkthroughs and live dev servers, then push straight to production main.
  • Code review shifts from reading raw syntax to inspecting frame-by-frame UI jitter and performance profiles.
  • SOC 2 compliance does not require blocking human code reviews on GitHub, only proven, repeatable controls.

The Pull Request Bottleneck

AI coding agents can generate clean, working features in two minutes. The problem is what happens next: the code sits in a queue waiting twenty-four hours for a human to open GitHub and read through fifty lines of diff.

Quinn Slack, co-founder of Sourcegraph and founder of Amp, points out that human review schedules break the math of agent workflows. As Slack puts it: “And if it's going to take a day anyway for a teammate to review it, you actually don't really care about making something go from 30 minutes of friction to 0 minutes of friction.” (play from 33:50)

The entire traditional review flow was designed around human typing speed and human bug patterns. You wrote code line by line, so someone else read it line by line to spot typos, logic errors, and bad variable names. But when an agent writes the implementation, line-by-line inspection is both the slowest and least effective way to test if the software actually works.

“But if you think about code review as making a change, going and creating a GitHub pull request, having one of your colleagues usually go on the web and read through the code manually, read everything, and then go back and forth, that, we're seeing that die,” (play from 32:57) Slack says.

Reviewing Artifacts Instead of Syntax

If developers stop reading raw diffs, what do they review? Slack argues the answer is artifact verification. Instead of guessing how code behaves by looking at text, teams verify the tangible outputs produced by the software.

“So if you think of code review more expansively, you know, you're reviewing the output, you want to see the agent showing you a video,” (play from 34:56) Slack explains. “What is it like? You want to know that the agent, in some cases if it's in the UI, has inspected every single frame of the video to make sure that there's no jitter during the loading of something, for example.” (play from 35:04)

At Amp, the team runs this workflow directly. Developers interact with agents across persistent threads. The agent runs the code on a live dev server portal, captures video evidence of the user interface, checks for visual defects, and runs test suites. The human evaluates the live output rather than parsing the syntax.

Once the output passes inspection, the ceremony ends. There is no long-lived branch, no waiting for a thumbs-up emoji on GitHub, and no round-robin review pinging. Slack notes: “We review it within Amp, and we have the agent review it as well, and then when it looks good, we've tried it out, we've looked at videos, we've played around with the dev server, then we just instruct the agent to push to main with that Ship button.” (play from 38:17)

The Compliance Myth

Most engineering teams cling to pull requests because they believe their auditors demand them. Enterprise founders often assume enterprise compliance requires two human approvals on every pull request.

Slack rejects that premise outright. Amp maintains enterprise security standards without blocking PR queues. “And things like SOC 2, they do not literally mandate that you must have blocking code review on GitHub,” (play from 36:58) Slack says. “What they require is a set of controls that you've shown to be effective, and you need to follow those.” (play from 37:04)

Automated CI gates, agent-verified test runs, recorded dev server interactions, and auditable deployment trails satisfy auditors while letting engineers push code twenty times a day.

What to Do With This

Take your team's next internal feature or bug fix and ban GitHub pull requests for it. Require the engineer to use an agent to spin up a live preview URL and generate a short screen recording of the change working in the browser. Have the reviewer approve the visual artifact and test output, then push the change directly to main.