Faire doubles PR throughput with Cursor Cloud Agents
Faire selected Cursor as its platform for coding agents, replacing an in-house background agent system.
Faire doubled weekly PR throughput and collapsed an 18-month migration down to a single engineer managing a fleet of agents. Cursor's cloud agents made this possible through scaled parallelization and autonomy. Cloud agents run without the memory and resource constraints of a local machine, and each one gets its own development environment—just like an engineer—to write code, test and verify its work, and ship software.
Cursor is now Faire's recommended platform for agentic development, replacing an in-house background agent system. The team also uses Cursor Automations to kick off more than 2,000 autonomous agent runs per week to save time on repetitive tasks like triaging bug reports in Slack, fixing CI failures, and routing code reviews.
Cloud unlocks scaled agent parallelism
Running multiple agents in parallel on a local machine quickly runs into constraints on local resources: each agent competes for the same compute and managing 10 in-flight tasks across separate terminals becomes its own job. "There are ways to parallelize agents on your local machine, but it's much more complicated," says Luke Bjerring, a principal engineer.
Initially, Faire tried solving this by building an in-house cloud agent system called Samurai that ran on self-hosted infrastructure. But getting the developer experience right requires substantial investment.
"Standing up our own servers is a significant investment. It requires hiring talent, bootstrapping machines, and maintaining complex infrastructure. We'd rather have engineers focused on adding value to our end users," says Bjerring.
Faire went looking for a platform that would give them the flexibility to run cloud agents on managed infrastructure or self-hosted machines. They selected Cursor for that deployment flexibility, close integration with tools like GitHub, agent reliability, a clean interface for managing agents in parallel, and seamless handoffs between local and cloud. Cursor is now Faire's recommended platform for agentic development across the company.
Cursor's cloud offering is a lot better than running local agents with worktrees or 10 remote environments you're shelling into. It's a streamlined UX for managing multiple concurrent agents.
Development environments give agents more autonomy
Parallelization only pays off if agents can operate like engineers: pulling dependencies, hitting internal services, running code, and verifying changes. Without a configured development environment, an agent can write code but can't close the loop on its work.
Faire's development environment makes cloud agent setup challenging. Backend and frontend live in separate repos with their own internal package dependencies, managed through Gradle and Bazel, and gated behind separate AWS credentials.
To simplify this complexity, Faire uses Cursor's agent-led onboarding: Cursor inspects each repo, figures out the required toolchain and dependencies, and produces an environment configuration that the team can edit and version. For workflows that require tighter control, Faire can define and manage development environments via Dockerfile.
We let Cursor onboard itself on every repo in our codebase. That takes a lot of the overhead out of new session starts and lets agents tackle tasks just like an engineer would.
Faire's designers use an internal tool called Playground to translate design systems in Figma into React components in code. With a fully-configured development environment, Cursor can run the Playground server, produce React components, and record video demos for designers to review the agent's work.
Faire is a Slack-heavy company and a lot of engineering work starts as a question or bug report in a channel. Engineers often call @cursor directly from Slack threads, handing off conversation context to a cloud agent that can investigate and return with a PR.
A lot of our work comes from ideas and discussions in Slack. You can see the message, kick off @cursor in the same context, and you get a PR a few minutes later. This helps me avoid jumping between tools and context while the agent does the work.
Programmatic agents with Cursor Automations
Beyond running parallel agents, Faire needed agents that could take on repetitive engineering work autonomously to save the team time. Faire has set up more than 25 Cursor Automations and now executes more than 2,000 autonomous agent runs per week without any manual prompting. The most common use cases are:
- Triage bugs reported in Slack. Automations monitor designated Slack channels for bug reports. When an issue comes in, a cloud agent is kicked off to investigate, open a PR with fixes, and provide a summary of its work.
- PR auto-healing. When CI fails on a PR, an automation kicks off, investigates the failure, pushes fixes, and updates the PR.
- PR routing. An agent labels every PR by author, risk, and size, then routes the PR to a tailored code review workflow.
The concept of automations has been long-lived at Faire, but setting them up was painful and complicated. Cursor Automations makes spinning up always-on agents accessible to every user.
Automating legacy migrations with parallelized agents
When Faire needed to migrate a large, retailer-facing application from MobX to native React state management, the team built an agent coordination system on top of Cursor called Swarm.
First, a scraper finds every detected usage of MobX in the codebase and writes the list to S3. Swarm then reads the list and delegates migration tasks to Cursor cloud agents, each running in its own isolated VM on Cursor's infrastructure. As one agent completes its work and merges its PR, Swarm fires off the next one.
What would have been 18 months of manual work for a full team is now coordinated by a single engineer managing a fleet of cloud agents.
Cursor's value comes from great context management and getting useful proprietary information across the company and codebase. All these human tasks that would take you hours, you can now delegate to an agent. We're saving huge amounts of manual labor.
Standing up build previews in less than a day
Faire's web application is large and complex, and seeing how even a small change behaves often requires spinning up the full app locally. Blair McAlpine, a senior engineer on the platform team, wanted to build a preview tool so that when any developer creates a PR, a sandbox spins up and the team can interact with the changes remotely.
McAlpine used Cursor to plan and execute the build. He started in plan mode and iterated on a step-by-step plan, with each step scoped to a separate PR. He then handed the plan to a cloud agent. The agent ran for two hours and produced five stacked PRs, each implementing a step of the plan.
What McAlpine expected to take weeks took less than a day with cloud agents.
The cloud agent ran in the background while I worked on other things. It took the preview builds from scratch to a working internal tool in less than a day.
Faire is now turning its attention to the next bottleneck. With engineering output approaching 2-3x, Faire is now looking at where that same kind of leverage can unlock momentum across the broader product development process. "As teams get more leverage from coding agents, the constraints are shifting," says Bjerring. "The opportunity now is to help adjacent teams scale their impact as well. That lets us reallocate within our existing footprint and take on more ambitious work."
If you're working on parallelizing your engineering velocity with cloud agents, reach out to start a Cursor trial.