One of the fastest ways an AI agent earns your trust is when it sounds like it already knows you. Your business. Your audience. Your tone. Your way of working.
Until now, most agents had to relearn that context every time they ran. Today, that changes.
We just shipped Custom Instructions, a new preference that lets you give supported agents persistent context about your business, so every output feels unmistakably yours.
Custom Instructions let you provide lasting, per-agent context about your business and communication style.
Instead of repeating the same background and preferences every time you run an agent, you set them once in the agent’s settings. From that point on, the agent automatically tailors its outputs to match.
Custom Instructions are a free‑text field where you can describe things like:
Once saved, these instructions are applied automatically every time the agent runs.
Agents use Custom Instructions to adjust how they respond—not just what they say. Specifically, they influence:
The result is output that feels aligned with how you normally communicate, without extra prompting or cleanup.
Custom Instructions are currently available on select agents, including:
More supported agents are coming, but these two already show the impact immediately.
Custom Instructions are saved per agent and applied automatically every time that agent runs. You can update them at any time, but you only need to set them once for the agent to start using them.
For Meeting Prep and Meeting Follow‑up, the impact is especially visible.
These agents weave your Custom Instructions into every section of the output:
The agent doesn’t just know what happened—it knows how you would think and talk about it.
Getting started takes less than a minute:
From that point on, every run of that agent uses your instructions automatically.
Custom Instructions are about more than tone control. They’re about moving from prompting to partnership. When an agent understands your business context and how you like to communicate, you stop managing outputs—and start trusting them.
This is a foundational step toward agents that don’t just execute tasks, but work the way a great teammate would.