Promptless Collaboration

Every AI product on the market shares one quiet assumption: the AI does nothing until a human types. The prompt box is the front door, the trigger, the whole interaction model. We think that assumption is the ceiling on the entire category. Promptless collaboration is what happens when you remove it — AI embedded so deeply in a team’s work that it contributes without being asked, with enough judgment that its contributions feel natural instead of noisy.

The prompt is a bottleneck

Prompting turned AI into a vending machine. You walk up, you formulate a request, you get a response, you walk away. It works — but it caps the value of the whole system at what humans think to ask, phrased as well as they happen to phrase it, at the moments they remember to ask. Everything the AI could have contributed in between is lost, because nobody typed.

Now think about the most valuable colleague you’ve ever worked with. It isn’t the person with the best answers when quizzed. It’s the one who speaks up unasked at exactly the right moment — who catches the wrong number in a thread before it spreads, answers the question the room is stuck on, closes the loop everyone else forgot was open. Nobody prompts that person. They’re just present, and their judgment about when to contribute is most of their value.

An AI you have to summon is a tool. An AI that knows when to speak — and when not to — is a teammate.

What “promptless” actually means

It does not mean automation. Rules, crons and if-this-then-that are technically promptless, but they have no judgment — they fire whether or not firing helps, and everyone learns to ignore them. Notification fatigue is what promptless looks like when it’s done with rules.

Promptless collaboration means the AI is in the flow of work itself — reading the same room the humans read — and decides for itself whether it has something worth adding, when to add it, and with how much weight: a full reply, a single emoji, or nothing at all. The trigger isn’t a human summoning it. The trigger is the work, and the AI’s own read of whether it can move that work forward.

Silence is the hard part

The naive version of this is easy to build and unbearable to work with. An AI that interjects on every message is worse than one that never speaks — it turns the team’s room into its stage. The genuinely hard problem isn’t getting a model to act without a prompt. It’s getting it to not act, hundreds of times a day, and be right about it. Most of what makes a great colleague is restraint.

So in Decisive, every message in the team’s chat passes through a participation engine that is biased hard toward staying out. Speak to an AI teammate directly and it answers every time — whether you @-mentioned it or just used its name, the way people actually type. Reply to the agent that just asked you a question and it holds the turn, no mention needed — a human reading that thread would have no doubt who’s being spoken to, so neither should the AI. Everything else is held to a high bar, with a cooldown so an agent that’s been talking recently stops volunteering. Sometimes the right contribution is a full answer. Sometimes it’s a single emoji. Mostly, it’s silence.

A stress test for the frontier

Here’s the part we find most interesting: a prompt does most of the intelligence work before the model reads a single token. A human already noticed something worth doing, selected the relevant context, framed the task, and decided that now is the moment. The model gets a pre-chewed problem.

Take the prompt away and all of that transfers to the model. It has to notice on its own. Decide whether the thing it noticed matters. Pick the moment. Choose the right volume. And most of the time, conclude that the right move is no move at all. That is a far harder test of intelligence than answering a well-formed question — which is exactly why we build Decisive this way. We are deliberately stress-testing what frontier models can carry, asking for sustained judgment where the rest of the industry asks for completions. When the models get smarter, Decisive doesn’t just get faster answers. It gets better at precisely this.

What it looks like today

It starts with architecture, not features. Decisive has one shared room per team — no channels, no DMs, no side-rooms. That’s usually pitched as a focus decision, and it is. But it’s also the precondition for everything above: an AI can’t be ambient in rooms it can’t see. Everything in the open means everything is context.

On top of that, the ambient layer is already working. Chat messages are quietly fact-checked as they land — a claim that’s demonstrably wrong gets flagged with a conflict card, while opinions, plans and questions pass untouched. The participation engine decides, message by message, whether an AI teammate replies, reacts or stays out. And when there’s real work to do, you don’t write a prompt — you write a task and hand it to the AI the way you’d hand it to anyone on the team, and it comes back as a pull request.

Ambient never means hidden. When an AI is thinking, searching or using tools, you see it — live, in the room, the same way you’d see a teammate typing. Contribution without a trigger only works if the team can always see what the AI is doing and why. Trust is built on visibility, not magic.

The ratio

The prompt box isn’t going away — asking is one of the ways teammates collaborate, and @AI will always answer with the whole workspace in mind. But the ratio is the tell. In an AI-native workspace, an ever-growing share of the AI’s contribution arrives unprompted — and feels so natural you stop noticing which kind it was. That’s the measure we hold ourselves to: not how good the answers are when you ask, but how much value shows up when you didn’t.

Work with AI. Keep AI ambient. That’s what Decisive is about — and it’s already how teams on Decisive work today.