The AI workspace my former team and many other teams still use

The problem

The company's standard AI tool was Gemini, and the setup we had was very restrictive. It couldn't save files to your computer, control it or run a browser for you, and behind the company firewall it couldn't really search the web. Claude's desktop app wasn't available either.

I found a way to get access keys to Claude Code, Claude's coding tool, for myself and my team. That was fine for me, because I'm comfortable typing commands into a text-only window. Almost no one else on my team was going to be comfortable working that way.

What I built

Claude Code can run headless, meaning in the background with no window of its own. I built a browser interface on top of it that runs on each person's own computer and works like the Claude desktop app. I built it for myself first and called it Winterface, the Wes interface, with a winter theme to match. It ended up doing almost everything Claude's own desktop app does today, and at the time, a few things it didn't.

The basics

Several chats at once, a choice of model, attachments and images, archive search, a panel for background tasks, and work that didn't disappear when you closed the browser.

Project memory

Key decisions, what got built and anything else that would help later on were saved to memory as the work happened, automatically or with little effort. Every new session picked up where the last one left off, so each one got faster and wiser.

Standing guidance

Every session came loaded with rules, including a guard against sycophancy, the habit of telling people what they want to hear. Other rules shaped the output, what it assumed and how often it checked its work.

Publishing

We got our own internal web server, something I hadn't seen another business group do. Anyone could publish to it just by asking their Claude: things they'd made, and small web apps for team operations, automation and sharing information.

Shared projects

When someone turned on sharing, the server became the master copy. People could get the latest version, send their changes back and work on the same project at the same time, with Claude helping sort out any conflicts. Sharing carried the full chat transcripts too, so anyone could ask why something was done, even if it never made the final record. That made status and reporting possible without a meeting, which is how the project hub worked.

Version control and a roadmap

People who wanted it had version control, a saved history of every change they could go back to. A built-in roadmap let them track tasks and formally close them out.

Documentation and updates

It came with its own documentation and an internal download site, and it updated itself.

What had to be handled carefully

• Keep it familiar. It had to look and work like Claude's own desktop app, so it was easy to understand and use. People didn't need the extra overhead.
• Fit it to how people worked. It wasn't just the tool. I had to think through which workflows people needed, like publishing and working together, and customize the tool to support them. That meant understanding how people were using it and what they were using it on, and keeping a feedback loop so it kept adapting to their needs.
• Don't break what people rely on. Once people depended on it, the hardest part was version control, testing and keeping updates backward compatible, so each new version still worked with what people had already built. Any time I cut corners on planning, testing or compatibility, an update broke someone's workflow. That cost the tool credibility and cut into how much people used it and what they got from it. Keeping it stable took the same work and best practices as any software. A product manager, a developer and a tester working on it together would have made it tighter.

The value

Business users who would never have opened a command line got fully capable AI. They used it to build content and applications, work with their files, analyze information and automate routine work, plus the many other business uses I've written about and built. That's what got my team working with powerful AI, and it brought huge gains in productivity and quality.

Other teams heard about it and wanted to use it, and we opened access to anyone who did. Several dozen people used it, and it's still the interface they use.