The single-player era is ending

The pattern

Every great work tool of the last twenty years won the same way. Word processing was solitary until documents became something a team could stand inside together. Design was solitary until the canvas became a place, not a file. Code review, whiteboards, spreadsheets: the tool that let the team in beat the tool that didn't, every single time.

The pattern is not about features. It's about where the work lives. When work lives where everyone can see it, teams trust it, improve it, and ship it faster. When it lives in one person's window, it becomes that person's problem.

The line agents crossed

In 2026, AI agents crossed a line that changes everything: their runs got longer than one person's attention. An agent that answers in four seconds can live in a chat window. An agent that works for four hours, or four days, cannot. It outlasts your meeting. It outlasts your workday. It outlasts your timezone.

Work at that scale was never a solo activity. It pulls in reviewers, approvers, the teammate covering the weekend, the manager who needs to know if it's on track. And today, every one of those people is locked out.

The lonely box

Because AI is still single-player. You type into a box only you can see. Your company's most consequential new work is scattered across a thousand private windows, and the only sharing mechanism is a read-only transcript nobody else can touch.

Nobody can see the runs. Nobody can take over. Nobody shares the review. The person who started the run carries it alone, and everyone else finds out when it ships.

The missing seat

Here's the strange part: the industry knows. Every serious agent platform published a protocol in the last two years, with states, activities, steering. And every one of them models exactly one human. There is no second observer, no role, no hand-off in any of them. The seat next to you at the agent's table simply does not exist in the spec.

Some are answering with a shared canvas: everyone's hands on the work. We think that's half right, and the wrong half matters. When everyone's hands are on it and something breaks, whose hands were they? Visibility without ownership isn't collaboration. It's a crowd.

What we're building

Handbell is a table. Your team's agents, runs, and people, in one live surface. Anyone can drop in and watch. One person steers at a time. Every action carries a name. Hand-offs are passed and accepted, like work between professionals, because that's what this is.

Many eyes. One owner. That's our whole disagreement with the field, and we think it's the difference between a demo and a company you can run on this.

The wager

We're named after an instrument that is nearly impossible to play alone. Each ringer holds a few bells; the music exists only when the group plays together. The opposite instrument is the carillon: one player, alone in a tower, commanding everything. The last two decades of software are a graveyard of carillons.

AI will have its multiplayer moment. Our wager is that it looks less like a louder solo instrument and more like a band: individual responsibility, precise timing, collective work. If your team is already living the single-player problem, we're building this with founding teams right now.