Meet Loopway

• web
• github

Working on and in Loopway has been my core focus for the last two weeks, and it's been pretty great. Over that time my bots and I have closed 178 tickets, including three full system refactors. I've reviewed all of them.

The Idea

The basic idea behind Loopway is that we already have the communication tools we need to build software and work together, but our agents can’t react to them. Yes, they can pull data from them, but only when we prompt them to. Loopway is a notification system for our agents, so they find out about feedback as it happens. My hope is that with notifications, our agents react to what’s happening rather than waiting for us to tell them. Even as a team of one, this shift matters: it got me thinking about bots as teammates. The shape of our tools shapes how we work with them and shapes our imaginations.

Stuck at Level 2

The idea for Loopway came while I was building a toy survey tool to better understand Jev. I had to take the survey many times, and while I did, I found myself reflecting on where I sit in the four levels of AI automation laid out by Notion. The honest truth is that I sit closer to level 2 than level 3. That is, I use AI to do tasks faster, not to add capacity to my team.

Part of why I get stuck at level 2 is that I don't take the time to define the roles I need AI to fill. Part of the reason is that the standard tools present as a single bot. So in Loopway, bots are the core entity you create. Each bot has a role (its base prompt), the types of messages it listens to, and logic for how it groups messages into "threads" (agent sessions). By starting with bots, my goal is to put forward the idea that AI has roles in an organization and that those roles should be defined. My hope is that this defaults teams into thinking of bots as teammates and pushes us all a bit closer to level 3.

So. Who Is My Team?

Loopway has been built by three bots and me.

• loopway-epic: the planner. I work with this bot to cut tickets for loopway-dev on larger projects. When plans are unclear, it tells me.
• loopway-dev: the core implementer. It sees tickets ready to work and does them. Sometimes I cut tickets directly for it, but most of the time I'm just giving it code feedback.
• loopway-review: I don't talk to this bot much. Its goal is to let me focus on functional requirements in PR review rather than edge cases and basic security rules.

This set of three bots has worked pretty well for me. Using Loopway has shifted my perspective toward seeing different bots I can shape and work with, rather than the singular "claude" or "codex" way of thinking I had before.

Still the Bottleneck

That success has pulled me deeper and deeper into delivering Loopway rather than freeing up my time. It has put me in the role of PM/Staff Eng rather than EM or Sr Eng, which is nice. But my three bots still need more from me than I can give. I am still the bottleneck almost all of the time, and I am still working too low-level. If I'm going to get the full impact of AI into my workflow, I need to think even bigger.

Week Three

As I go into week three, my focus is on bots that aren't execution-focused. All of my current bots assume I have the ideas. All of them need direction. Now I'm building bots that aren't here to ship code: bots that track what I'm learning in discovery calls, research topics for me, and prepare briefs for me to read and review. Because Loopway is so early (and just one of many product ideas), I expect to still be in the weeds. But I need to take bigger swings than my current setup lets me, and finding ways to make that happen is my focus.