Running My Day From a Terminal Window

Brad Feld built something called CompanyOS. He wrote about it on his blog – the philosophy behind it, the architecture, how it works. I’d encourage you to read those posts for the technical backstory. Brad set me up on it yesterday, and what I want to write about here is what it actually feels like to use it. One day in and I’ve already shifted a surprising amount of my daily business operations into it. It’s been mindblowing.

What CompanyOS Is

The short version: CompanyOS is a git repo full of markdown files that teach Claude Code how to run your business. There are “skills” – markdown documents that describe how to draft emails, manage calendars, do research, handle customer communications. There are “rules” that encode your company’s context – who works here, what systems we use, how we communicate. And there are agent definitions that let Claude dispatch specialized sub-tasks when the work gets complex.

Brad’s framing is “running a company on markdown files” and that’s exactly right. There’s no app to install, no platform to configure, no orchestration layer or workflow engine sitting in the middle. It’s plain text files in a repo, a smart model that reads them, and MCP connections to the systems you already use – Gmail, Google Calendar, Slack, Asana, and others. That’s the whole stack.

iTerm2 as Mission Control

Here’s the part that surprised me. My primary work interface is now a terminal window. I have iTerm2 open with Claude Code running, and from that single window I can draft and send emails, message colleagues on Slack, look up my calendar, research topics on the web, analyze documents, manage tasks – all without opening a browser tab or switching between apps.

I realize how that sounds. A terminal window as a business operating system is not an obvious pitch. But the thing I didn’t anticipate is how much cognitive overhead disappears when you stop context-switching between eight different applications. I’m not logging into Gmail to draft an email, then switching to Slack to message someone, then opening a PDF viewer to review a document, then jumping to Asana to check my task list. It all happens in one place, in one conversation, with one system that has the full context of what I’m working on.

Asana as the Task Controller

One of the first things I set up was a dedicated CompanyOS project in Asana. The workflow is simple: I load up Asana with the tasks I need done – draft this email, research that topic, follow up on this thread. Then inside Claude Code, I tell it to go grab the next task from the CompanyOS project. It reads the Asana item, figures out what needs to happen, executes it – drafts the email, does the research, whatever the task requires – and marks the Asana item as complete when it’s done.

This turns out to be a powerful pattern. I can batch-process work. I queue up a dozen tasks in Asana during the morning, then let Claude work through them one by one while I focus on things that actually require my direct attention – calls, meetings, thinking through strategy. When I come back, the tasks are done and sitting in my Gmail drafts waiting for a final review before sending.

What I Actually Did on Day One

Here’s a sampling of the things I ran through CompanyOS today, to give you a sense of the range:

I needed apartment research for my daughter who is looking at places in two different cities. Claude did the research, compiled the results with links to each property, summarized pricing and amenities, and saved the whole thing as a Gmail draft. When I later asked it to filter those results for pet-friendly buildings, it did a second round of research and sent the updated findings as properly threaded reply emails to the original messages.

I had a legal email to draft – notifying outside counsel about an ownership change in an LLC entity. Claude found the existing email thread with the law firm, drafted a reply with the correct details, and saved it as a draft in the right conversation thread. No copying and pasting between systems.

Someone sent a fee proposal for something I’m working on (weeks ago in email). I described it generally, Claude found the email, figured out that the document was actually sent in a link, found the document in my downloads folder, extracted the document, and gave me a structured summary of the entire fee schedule. I had a meeting with them later that afternoon and walked in fully briefed.

I messaged Brad on Slack about a small bug I’d found. I checked my calendar to confirm a meeting time. I pulled up and summarized coaching documents that had been sent as Word attachments. Each of these took seconds to minutes, all from the same terminal window.

I kicked off a few research threads, wrote detailed responses to a few emails, analyzed some company information. Again, all from a terminal window. Oh, and I wrote two blog posts (including this one). And by “I wrote” I mean that I had Claude write the drafts, which I then gave feedback on; I uploaded them to my GitHub as a draft, where I made some edits, and then published it. I’m working on having it auto-post to LinkedIn and X as well (that should be set up later today).

What I’ve Learned So Far

A few things stand out after day one.

First – the voice profile matters more than I expected. One of the setup steps is building a profile of how you actually write, based on your sent emails and other documentation. Claude uses this to draft messages that sound like you, not like a robot. The emails it produces are close enough to my natural voice that I often send them with minimal edits. This is a big deal. If every AI-drafted email reads like it was written by ChatGPT, you’ll never trust the system enough to use it.

Second – skills compose in ways I didn’t anticipate. The same system that knows how to draft emails also knows how to do web research, analyze PDFs, and manage Asana tasks. When a task requires combining these capabilities – find an email, extract its attachment, summarize it, then draft a reply – it just works. There’s no integration step because everything runs through the same model with the same context.

Third – and this is the big one – the value is compound. No single capability here is revolutionary on its own. Email drafting, web research, document analysis, task management – these all exist as standalone tools. The revolution is having them all in one place with shared context and zero switching cost. It’s the difference between having a dozen specialized tools and having one system that understands your entire workflow.

Where This Goes

It’s early. One day in. But the trajectory is clear – my terminal is already my most-used business tool, which is a sentence I would not have predicted writing six months ago. Brad built CompanyOS and wrote about its philosophy – I’m one day into experiencing how mind-blowing it is. The interesting question now isn’t whether AI can handle routine business operations. It’s what happens when all of your business context – your communications, your decisions, your company knowledge – lives in one place and compounds over time.