AI's Impact on Code Reviews and Development

Graphite emphasizes smaller code changes for efficient code reviews, with a technical team organized around experienced engineers and deep collaboration.

### The Importance of Small, Fast Code Changes and Effective Code Reviews
* 🔍 Greg Foster, CTO of Graphite, discusses the benefits of smaller code changes, citing Google research that shows improvements in **review quality, bug detection, and rollback likelihood**.
* 🛠 Graphite is a platform designed to facilitate **smaller and faster code changes**, built on top of GitHub to help teams collaborate on **code creation, review, and merging**.
* 🏢 Large tech companies like Facebook and Uber have been using more advanced code review workflows, such as **stacked code changes**, for over a decade, but many other companies are still using **traditional pull requests**.
* 🔄 Foster emphasizes the historical trend in software development towards **faster, smaller changes**, from **annual releases** to **continuous integration and continuous deployment**.
* 📊 Foster highlights the importance of keeping code reviews **small and fast**, as they are often the slowest part of the development process, with an **average review time of tens of hours**.
* 🔒 Foster recommends combining code review best practices with other continuous delivery practices, such as trunk-based development and monorepos, to **maximize efficiency**.
* 🔄 When discussing synchronous vs. asynchronous code reviews, Foster suggests that the best approach may vary depending on the specific goals of the review, such as **security, knowledge sharing, and consistency**.
* 🤖 Foster acknowledges the potential for AI and automation to **improve code review processes**, but cautions that at scale, human review is still **essential for security and context**.

### The Importance of Code Reviews and the Future Impact of AI in Software Development
* 💬 "One of the most common mistakes I see in the world of code changes is an engineer creating a **big code change**, putting it up, getting a **mediocre review**, and maybe **failing CI**."
* 🤖 "AI is possibly changing the way we do this part of the workflow, with **over 50% of code being generated** now AI-assisted."
* 🔍 "There is a **meaningful increase in the net volume of code** that the average engineer is producing, potentially putting strain on review, CI, and deployment processes."
* 👥 "The continued increase in AI-generated code is actually putting more emphasis on **human review, CI, deployment, and the human elements** of the development process."
* 🏗 "While AI can assist in some aspects like linting and consistency checking, humans are still essential for reviewing code, especially for **security and catching subtle bugs**."
* 👩 "The future of software development may involve a more nuanced division of responsibilities between AI and humans, with AI assisting in tasks like **summarizing, explaining, and referencing**."
* 🏢 "Graphite, a small startup with 22 employees, has a technical team that is deeply involved in using their own product (**dogfooding**) and heavily relies on a staging environment for **testing before production deployments**."
* 🚀 "Their delivery process includes an **automated pipeline** to deploy changes to production, with a **one-hour lag time** in staging before automatic promotion, unless **manual intervention** is required."

Code Reviews, AI, and Modern Engineering 🔍 — with Greg Foster

Interviews with world-class engineering leaders about writing great software and working well together.


Refactoring

The Impact of Smaller Code Changes on Development

Greg Foster, CTO of Graphite, emphasizes the significant benefits of smaller code changes, as supported by Google's research over the past 15 years.

Graphite, a code change collaboration platform, focuses on keeping code changes small and fast, drawing inspiration from big tech companies like Google and Facebook.

The traditional pull request workflow of larger code changes has stalled progress in the industry, while smaller code changes unlock parallelization and faster development.

The Shift Towards Smaller, More Frequent Code Changes

The traditional approach of large commits and pull requests is being replaced by smaller, more continuous changes in the development process.

This shift is driven by the need for parallelization, faster reviews, and a more unified collaboration process.

Graphite is aiming to support this transition by focusing on code review and offering guidance on best practices for small, fast code changes and trunk-based development.

Monorepos are generally favored for closed-source development, with the belief that smart tooling can effectively manage conflicts.

The popularity of polyrepos in the past was partly due to limitations in tooling, but the improved tooling now provides the benefits of polyrepos within a monorepo.

The decision to use a monorepo or polyrepo may depend on the need for open source code, but starting with a monorepo is advised unless there is a strong reason not to.

The choice between synchronous and asynchronous code reviews should consider the cycle time and the need for deep, contextual review.

The Importance of Code Reviews in Software Development

Code reviews are crucial for ensuring security, knowledge sharing, and consistency in the code base, as well as for identifying and addressing nits and bugs.

To maintain high velocity and keep developers unblocked, code changes should be kept small, and tools like Stack should be utilized for asynchronous collaboration.

Additionally, a human element, such as direct communication after a few back-and-forth comments, is essential for keeping the review process efficient.

Some companies are exploring the possibility of automating trivial code reviews using sophisticated pipelines and AI, but the effectiveness of this approach is still a topic of debate and experimentation.

The Impact of Deviating from Code Review Norms

Deviation from code review norms may not be beneficial, especially as companies grow and become more complex.

Certain types of projects, such as data science or internal tests, may allow for more relaxed code reviews on a case-by-case basis.

The primary goal of code reviews should be to avoid blocking people's progress, and stacking changes can help achieve this, while also reducing the sense of urgency to skip reviews.

AI is significantly increasing the volume of code changes, with over 50% of code being AI-assisted, leading to larger PRs and more strain on the development process.

The rise of AI-generated code is shifting the focus towards the importance of human review, CI, and deployment for catching subtle bugs and ensuring security.

Despite AI's assistance in common checks, human involvement in reading and reviewing code remains crucial, especially for identifying potential security issues.

The Impact of AI on the Software Development Process

AI is likely to enhance the code creation process by helping developers write code faster and more consistently, as well as aiding in summarizing and explaining code changes.

Human involvement will still be crucial in code reviews, considering implications of changes, while AI can assist in summarizing, explaining, and referencing relevant context from the code base.

The role of AI in builds, deployments, and the software release cycle is not yet clear, and it may have a limited impact on these aspects of the development process.

The Evolution and Organization of Graphite's Engineering Team

Graphite is a small startup with about 22 people, primarily based in New York City, with a technical team working on a software collaboration tool.

The team is divided evenly between engineers and those working on product, design, and go-to-market functions, with an emphasis on deep technical knowledge for everyone.

The engineering team is structured with one engineering manager (currently one of the founders) and around 10 individual-contributor engineers, mostly operating at a senior to staff level of experience and responsibility, in order to provide high autonomy and decision-making for the product and engineering workflows.

The Importance of Thoughtful Product Development

Developing a minimum viable product (MVP) depends on the type of product, with the need for a more thoughtful approach when aiming to challenge the status quo.

Graphite's design decision to create a code collaboration platform that syncs back to GitHub has set a high bar for their MVP, aiming to provide a seamless experience for users.

The company's internal delivery process includes extensive dogfooding of the product and a separate AWS staging account for employees to replicate the infrastructure, emphasizing the importance of empathy and thoughtful testing.

Staging deployment should be automated to allow periodic production merges without manual intervention.

A faithful data replication of production and meaningful organic traffic are crucial for a reliable staging environment.

It is important to maintain and utilize the staging environment to avoid creating a higher maintenance burden.