The way we collaborate is about to change, again
A foray into background coding agents
If you’ve used coding agents, you know how powerful they can be. As an engineer, they can truly be a force multiplier, helping you do the work of a small team. As someone who collaborates with engineers, but might not be as technical, they can shatter any sort of ceiling limiting what you’re capable of building.
Once you experience even a small sliver of the possibilities they enable, it’s hard not to turn into a maximalist and try to offload more and more work to coding agents, if nothing else to see how far they (and you) can really go. Especially if you have a Claude Code Max subscription, then the cost of spinning up another agent drops to 0, so why not do it? But, as soon as you start working on more than a couple of tasks at once, the workflow really starts to fall apart.
You’re juggling multiple terminals, trying to keep track of what’s where, your computer fan sounds like it’s about to take flight, and you’re hoping one of the agents hasn’t been stuck for the last 30min waiting for your permission to run a simple command. You become a babysitter and destroy any sort of flow you had, which goes against the point of multi-tasking in the first place.
That’s why we initially built Terragon, a developer tool that lets you orchestrate Claude Code agents in parallel, and in the cloud. Create a task in Terragon and a virtual development environment spins up for an instance of Claude Code to write, test changes, and automatically open a GitHub pull request when work is done, while you get a dashboard to easily track progress and CI status.
Instead of micro-managing agents locally, you turn them into true collaborators where you trust them to work independently, and you check in when work is done. The best part? You can continue the conversation with Terragon from wherever you are: pull down the changes locally using the Terry CLI, or ask Terragon to follow up on the task from the web, GitHub pull request, even your phone.
Me, Sawyer, and Michael have been building this for 2 months, alpha testing for 1 month, and so far:
📈 300+ hours of work offloaded to Terragon
🚀 900+ PRs merged
🔄 We’ve continuously used Terragon to build Terragon (for a small team, we’ve been shipping like mad)
Terragon is now free while in public beta for anyone to use, and I’m so excited for more people to try it out.
One thing I particularly love about Terragon is that it works great for both engineers and non-engineers alike. I’ve shipped (and broken) my fair share of production code before, but I’m not an engineer and don’t live in an IDE, so Terragon’s web and mobile-based workflows are perfect for letting me quickly capture and act on ideas.
We’ve been using Terragon nonstop to build Terragon the past 2 months and it’s honestly been a bit addicting given how powerful it feels. These days, when a Terragon user reports a bug, I tend to create a quick task in Terragon to just start working on it instead of tracking it in Notion to consider later, which is kind of a wild mental shift.
Instead of logging tasks, prioritizing them, discussing them etc., I get Terragon to start building it and then submit a pull request on my behalf, while I bounce to the next thing. The discussion now centers on how to get the PR merged instead of whether to even consider building it. It’s been so empowering to just test out ideas directly in a production codebase, while having the peace of mind that all work is happening in isolated sandboxes, so I can’t really mess things up. This has been a massive unlock for validating ideas without wasting team resources.
The other day, an early Terragon user reached out asking for a Slack integration. Before, I might add this to a backlog, where the individual priority might not be high enough to get to this anytime soon. Now? I fire off a task to Terragon from my phone while checking my email at lunch, then review the work from my laptop when it’s ready to merge.
It’s early, but I’m already seeing big shifts in how I work and collaborate. Pull requests, agent conversations, and build previews have now become key collaboration artifacts I create. I still write docs, whiteboard in FigJam, and design in Figma (I'm actually designing more than I ever have before), but there’s much more fluidity between my work now. This is a small example, but captures a deeper paradigm shift: that agents move the cost of doing so far below the cost of planning to do that the entire product development cycle starts to reorganize itself.
This opens up so many possibilities for what the future of cross-team collaboration looks like, and that’s a future we’re excited to sprint towards by opening up Terragon.
We built it to help people orchestrate Claude Code agents in the cloud, but the long-term vision is much bigger: more agents, more of your team, less friction. Stay tuned.