Promptless means nobody has to type a request. Nobody opens a chat window, picks a deal, and says what they need. Something happens in the tools your team already uses, like a call ending, and that starts the work.
Engineers have a name for this. LangChain calls these ambient agents: agents that “listen to an event stream and act on it accordingly,” instead of waiting for a person’s message. Microsoft’s Copilot Studio has the everyday version in its event triggers, which “allow your agent to act autonomously in response to the defined event occurring.” So a promptless tool is still an AI agent. It just starts work from events instead of requests.
That matters because a prompt only gets the work someone thought to ask for, on the deal they happened to open. Microsoft put it plainly when it published its 2024 Work Trend Index: “The prompt box is the new blank page.” On a sales team, the deals that slip are rarely the ones a rep is looking at.
Who starts the work is the easy question. The one that tells promptless tools apart is what decides the work once it starts.
What decides the work once something happens?
With a prompted assistant like ChatGPT or Claude, the rep’s request decides the work, even when the task runs on a schedule. Once an event starts the work instead, one of three things decides what that work is.
A fixed rule. One event, one predefined step: when a meeting ends, send the recap template; when a deal sits for 14 days, send the check-in. Strictly speaking, that’s promptless too, since nobody types anything. That’s why the word alone tells you so little. A rule runs the same step whatever else changed on the deal.
The model’s own plan. You give a general agent a goal, like “keep this deal moving,” and it works out the steps each time. That copes with surprises far better than a rule. But the judgment it uses is its own, so two reps’ agents can work the same kind of deal two different ways.
Your team’s playbook. The playbook is how your team works its best deals, written in plain English for the agents: your methodology, your buyers, the moves your top reps make on instinct. From it, Arrows keeps a gameplan for every deal, a path to close built from the playbook and from what’s actually happening on that deal. Every time something happens (a call ends, an email arrives, a meeting moves, a new person shows up on a thread), the agent reads the playbook and decides the best next action for that deal.
That last one is where your process lives, and it’s the one we built Arrows on. A top rep carries a plan like this in their head for their best five deals. Arrows writes it down as a gameplan and keeps it current for all forty.
What does each one do when a buyer hits pause?
Here’s one real moment from our own pipeline, anonymized. A buyer had one discovery call with us on September 17. On September 28, the buyer emailed to say they were pausing until they hire a sales leader.
The first two rows are illustrations of each kind, not of any named product. The last row is what Arrows did.
| What happens after the September 28 email | What the rep does | |
|---|---|---|
| A fixed rule | The next scheduled check-in goes out on time, as if the buyer never wrote. The CRM still shows the old close date. | Finds out later, if at all |
| The model’s own plan | A polite reply and a reminder to check back in a month. Reasonable, but it’s the agent’s habit, not your team’s. | Reviews whatever the agent chose |
| Your team’s playbook | A reply drafted, and a CRM update proposed: push the close date, lower confidence, set a new next step. The next steps turn to the sales leader they plan to hire. | Approves the reply and the CRM update before either goes out |
A week later, Arrows checked public sources for a sales-leader announcement and drafted a re-engagement note with a short walkthrough for the new hire. Nobody prompted any of it. You can read the full three weeks of that deal in our guide to AI sales agents.
Does promptless mean nobody checks the work?
No. “No prompt” is about how the work starts, not about who signs off. LangChain’s own post says ambient agents aren’t necessarily completely autonomous. It names three ways an agent hands work back to a person: notify (flag something and take no action), question (ask when it’s missing information), and review (wait for a person to approve a risky step, like an outbound email).
Arrows works in review mode today. A rep approves every draft and every CRM change before anything goes out.
Add one more question to how you choose an AI sales agent: can your team read what the agent decides by? With a fixed rule, you can read the rule. With a model’s own plan, the reasoning lives in the model. With a playbook, it’s a document your team wrote and can change. When you want to know why the agent did something, you read the playbook. When you want it to do something differently next time, you change it.
