We're building teamwork for the agent era

Henry Poydar, founder here. A few years ago, a startup I co-founded was acquired by a public company. My team was then handed a risky job: replace a heavily used, ten-year-old piece of infrastructure without disrupting customers.

The team grew to dozens of people across engineering, operations, product, marketing, and management, spread across several time zones. Meeting requests and messages arrived around the clock. Only a small fraction needed my attention, but I had to read everything to find it.

I had plenty of data and dashboards yet almost no useful context. I couldn't easily tell what people intended to do, whether the work matched the plan, or where I could help. We missed deadlines, morale suffered, and longer hours didn't fix it.

So I asked everyone to answer a short automated email each morning: What are you planning to do today, and why?

That small shared loop changed the project. Declared intent gave us something concrete to coordinate around. We caught problems before the work went too far, hit our targets, and felt in control again. And what's more, it gave the team agency: they got to declare what they would be doing, instead of just getting handed work tickets from above.

That idea -- an async check-in -- became Status Hero, which ran the daily coordination loop for teams around the world as remote and distributed work spread. Customers with larger organizations soon showed us the next problem: coordination overhead rose exponentially across five teams, ten teams, or twenty, resulting in an endless morass of meetings and missed deadlines.

So we built the successor to Status Hero, Steady, to run the loop not just within teams, but across whole organizations with complex cross-team dependencies. And we wrote down the practice behind it as Continuous Coordination: declare intent, focus on outcomes, and keep everyone current, automatically. Every turn adds to a shared record of what the team meant to do and what it actually did -- without the grind of endless meetings.

Over the past year our customers started running AI agents alongside their people, and the same failure showed up everywhere: speed and output way up, but most of the work was misaligned and useless, with an expensive AI bill to boot.

I recognized the pattern. It was my project going sideways again: plenty of output, no shared context, and everyone (and everyone's agent) consuming everything to find the part that mattered, at the cost of getting anything meaningful done. Could we fix this the same way: by having agents declare intent and report progress like everyone else?

So we put agents in the same loop as people. In Steady, agents join teams as first-class members. They check in, update goals, and consume real-time context alongside everyone else. Coding agents like Claude Code pull the team’s current goals, plans, and blockers before they begin, and can flag work that is drifting from the goal.

It worked, and it worked fast, because of what Steady had been building all along: a system of record for teamwork. Every turn of the loop writes down what someone intended and what actually happened, side by side. People use that record to sharpen the next plan. Agents can use all of it, every session, so they start work knowing not just the goal but how the team has actually been working toward it. People see what agents are doing. Agents understand what the team is trying to accomplish. And every turn, human or agent, makes the record, and the next round of intent, better.

That is what we mean when we call Steady the human-agent teamwork OS.

We also open-sourced OpenRoutines, a framework for running autonomous agents with those teammate habits built in: keeping track of what they did, what’s next, and when they need a person.

See how agent teammates work in Steady