How a company of two people and a few agents works

Two of us run HeyBeta, with agents helping with mechanical, repeated tasks: organizing the board, capturing follow-ups from meetings and preparing drafts. We keep judgment, tone, strategy and final approval. We're sharing how it works because we bring the same approach to our clients.

We use three questions to decide what to pass to an agent. Does the task happen at least three times a week? Is it mechanical rather than creative? Can we document its steps in less than a page? Three yeses make it worth exploring. A no means we keep doing it ourselves for now.

What does an agent do, and what do we do?

Our internal operations agent organizes the task board, sorts the week's work and captures follow-ups from meetings. That removes some repetitive work. It doesn't decide priorities or speak for us. We keep responsibility for those decisions.

We keep judgment, tone, strategy and final approval. An agent can save mechanical work; it doesn't replace judgment.

What broke?

Plenty, and that's part of the point. An agent summarizing meetings invented follow-ups nobody had mentioned. Another wrote in a tone that wasn't ours. We learned that a first version needs supervision while we establish where it can be trusted. Teams encounter the same issues when they start.

Why share this?

Each automation we build for ourselves also tests what we propose to a client. If it doesn't work for us, it isn't ready to bring to someone else. Showing what works and what doesn't gives people more to assess than a promise about what AI might do someday.