When one agent isn’t enough
Directing a single coding agent has become routine. Conductor extends that workflow to a fleet of them: a GUI where developers spin up multiple agents in parallel, each working on an isolated branch. The developer reviews the results, merges what’s good, and sends the rest back for another iteration. The tool is model-agnostic, so it works with Claude Code, Codex, and other agents, and is in use by engineering teams at Notion, Linear, Ramp, and Life360.
The early version had a hard ceiling: everything ran locally. CPU usage climbed, fans spun up, and closing the laptop meant the work stopped. Conductor’s co-founder and CEO, Charlie Holtz, previously spent time at Vercel and understood the constraint that local environments place on how developers work.
Removing the hardware ceiling
Users pushed for remote execution almost immediately. They wanted more agents running for longer periods, and local hardware was the bottleneck. For a team of six, moving from a local process to a cloud-native one is a major architectural shift. And when the user base includes engineers from Linear and Ramp, the choice of infrastructure provider carries weight: those engineers are deciding whether to run their code on hardware they don’t control.
Conductor evaluated several sandbox providers, looking for fast spin-up, snapshot support, and longevity in addition to the quality of the support relationship.
“There are so many sandbox providers now, and we needed to know that ours were going to be fast and reliable, and that we would get great support. Vercel checked all of those boxes, and they have incredible DX for our engineering team.”
— Charlie Holtz, Co-founder and CEO
Cloud Workspaces on Vercel Sandboxes
Conductor’s remote execution layer, Cloud Workspaces, is built on Vercel Sandboxes. When a developer opens a new workspace, agents start on a remote server rather than on the local machine. The interface is unchanged from the local version.
“Our users can’t tell the difference between local and cloud because Vercel Sandboxes are super fast.”
— Charlie Holtz, Co-founder and CEO
Holtz reported that Notion’s team described the experience as feeling “like an octopus, where each one of your arms is working on something separately.”
Staying ahead of the model churn
Conductor’s model-agnostic stance is deliberate. Agent harnesses and model capabilities shift frequently, and Conductor keeps the same cloud layer underneath regardless of which tool is currently in favor. As Holtz put it: “The vibes shift every 3 months. Right now GPT-5.5 has the vibes on its side, but that’s going change in another few months. Being agnostic to the providers is the key, and we get that with Vercel.”
The bet is on the interface itself: a single place to direct agents, review their output, and decide what to merge. With the sandbox layer in place, that interface is no longer tied to the hardware the developer happens to be carrying.
