AI coding agents are starting to leave the laptop behind.
Tools that started as running only inside terminals and IDEs are moving into persistent cloud environments, where they can operate for longer periods, run in parallel, and continue working after developers close their machines.
The latest example is Conductor, an AI coding startup that recently raised a $22 million Series A round of funding. The company gained traction with a Mac app that serves as an interface for managing coding agents locally across workspaces. In early May, the company announced Conductor Cloud, which moves those agents into hosted environments.
The shift mirrors broader changes across the AI coding market. Anthropic recently launched Claude Managed Agents, a service that lets businesses run long-lived agents on Anthropic’s infrastructure and adds remote-control capabilities for Claude Code sessions via web and mobile interfaces. Mistral has similarly started pushing its Vibe coding agent into the cloud.
Elsewhere, open-source AI coding startup Roo Code recently announced it was shutting down its VS Code extension and broader IDE tooling in favor of Roomote, a cloud-based autonomous coding agent designed to operate across platforms such as Slack, GitHub, and Linear.
So Conductor’s latest move is very much on-trend.
Into the cloud: Solving the “interface challenge”
Founded out of San Francisco in 2024 by Charlie Holtz and Jackson De Campos, Conductor gained traction with a Mac app that lets developers run multiple coding agents — including Anthropic’s Claude Code and OpenAI Codex — in parallel across isolated copies of a codebase, review their work, and merge the results back together.
The company’s new cloud offering, available by invitation as part of an early access program, extends those agent sessions into hosted environments that continue running remotely after a developer disconnects locally.
“I think to get to the next level, where you run more than three to five [agents], it’s an interface challenge.”
In a recent video interview with Y Combinator partner Aaron Epstein, Holtz says he views the current crop of AI coding tools as an orchestration problem, particularly as developers begin running multiple agents simultaneously.
“In my head, I can only really manage like three to five agents at once,” Holtz explains. “I think we’ve proven that you can run more than one coding agent at a time, and it will still be productive. But I think to get to the next level, where you run more than three to five, it’s an interface challenge.”
Conductor Cloud lets developers run coding agents across separate hosted workspaces tied to different tasks and repositories. Developers can then inspect the code changes generated by those remote agents directly in the Conductor interface via a side-panel diff view.
It’s worth noting that hosted workspaces potentially change the economics around products like Conductor. Much of the AI coding boom so far has revolved around local clients and developer tooling layered on top of foundation model APIs. Cloud-hosted workspaces introduce the possibility of companies charging not just for coordination software but also for the infrastructure on which those agents execute.
Conductor hasn’t detailed pricing for Conductor Cloud, though the hosted service will likely sit alongside its existing local-first product, which already includes an enterprise incarnation that the company says counts users based at companies such as Spotify, Square, Ramp, Linear, and Notion.
Terminal decline?
AI coding tools are already changing how quickly software teams can build and ship products, and models are only getting stronger. Tasks that once required carefully scoped MVPs and narrow product ambitions can now be tackled more aggressively as coding agents take on larger chunks of implementation work.
Holtz, for his part, argues that the current generation of tools still represents a relatively early stage of that shift.
“The models are going to get 10, or 100 times smarter. They’re going to be able to run for longer without you needing to intervene.”
“One thing we feel confident about is that the models are going to get 10 or 100 times smarter,” he says. “They’re going to be able to run for longer without you needing to intervene. They’re going to start feeling more like a human coworker would, although it’s like they will have an alien brain that’s very different to ours.”
And that, in part, is why Holtz views persistent cloud execution as important to the future of AI coding. If developers are eventually supervising fleets of longer-running agents rather than directly steering every step themselves, keeping those systems tied to a single laptop session doesn’t make sense.
“I think that’s one reason we’re really excited about [Conductor] cloud – it’s that the agents are going to be able to run for much longer,” he says.
While cloud-hosted agents are becoming a bigger part of AI coding, few companies appear ready to abandon local development environments entirely.
Amp, the AI coding startup spun out of Sourcegraph, recently rebuilt its CLI with support for remote control, plugins, and longer-running agent sessions. The local environment doesn’t disappear; more, it becomes a place developers use to monitor and steer agents running elsewhere.
Atlassian is moving in a similar direction, expanding access to its suite of enterprise tools through a new CLI designed for AI coding agents. The idea is to let agents navigate those systems directly — querying tickets, pull requests, and project data — rather than relying on developers to manually feed that context into prompts.
And so the emerging picture across the industry is coding agents becoming persistent systems that move between laptops, terminals, browsers, and hosted infrastructure – with developers spending less time writing every line themselves and more time supervising agents across different environments.
YOUTUBE.COM/THENEWSTACK
Tech moves fast, don’t miss an episode. Subscribe to our YouTube
channel to stream all our podcasts, interviews, demos, and more.
