Harvey AI recently published a deep dive into Spectre, their internal platform for running coding agents in the cloud. It's one of the most important articles written about AI-assisted development this year, and most people missed it.
Here's the short version: Harvey's engineers were using Cursor and Claude Code on their laptops. It worked great for individual tasks. Then they hit a wall.
The wall every team hits
When one engineer runs a coding agent locally, the output is invisible to everyone else. There's no shared context, no visibility into what the agent is doing, and no way to collaborate on the output. Code gets written in isolation, reviewed in isolation, and merged without the team ever seeing the agent's reasoning.
Harvey described it perfectly: "local coding agents break down at the organizational boundary."
This isn't a Harvey-specific problem. Every company running more than 3 engineers with AI agents will discover the same thing. The agent is powerful individually. But it creates a coordination gap at the team level.
What Harvey built
Spectre is Harvey's answer. It's an internal platform that moves agent execution from individual laptops into shared, cloud-based infrastructure. The key design decisions:
Cloud sandboxes, not local machines. Every agent run happens in an isolated sandbox with explicit security boundaries. No more "the agent had access to my entire desktop."
Durable run records. Every agent session is recorded as a durable object. The team can see what the agent did, what context it used, what decisions it made. Think of it as an audit trail for AI-generated code.
Collaboration surfaces. Agent work shows up in Slack, on the web, and in PRs. Other engineers can see what's happening, comment, and redirect the agent before it goes down the wrong path.
Same runtime for interactive and automated work. The platform doesn't distinguish between a human asking the agent to do something and a scheduled job. Same infrastructure, same visibility, same review process.
Why this matters for you
Harvey is an $8B company with a dedicated platform engineering team. They could afford to build Spectre from scratch. Most companies can't.
But the problem Spectre solves isn't optional. If your team is using AI coding agents, you will hit the same wall Harvey hit. The question is whether you build the infrastructure yourself or use something that already exists.
The architecture is converging. Harvey built Spectre. Stripe built their own internal agent platform. Every company at scale is discovering that local agents don't scale to teams.
The patterns are the same everywhere: cloud execution, shared visibility, structured review, and team collaboration on top of individual agent runs.
This is what Lightsprint is building. Not as an internal tool for one company, but as a product for every team.
The takeaway
If Harvey, with thousands of engineers and billions in funding, decided that local coding agents weren't enough, your team will too. The only question is when.
The good news: you don't need to build Spectre. You can start using cloud-based agent infrastructure today.