Every Dawn agent now knows Dawn. Ask one how to set up your workspace, where generated files end up, or whether something should be a memory page or a connected source, and you will get a real answer in plain language — plus the next step to take in the workspace you are sitting in.
Dawn agents already read your tickets, review your pull requests, run web research, schedule tasks, and reason across half a dozen connected sources. Somewhere between shipping the files browser and the mini-games launcher, someone on the team finally said it out loud: the agents are right there. Why are we still asking people to read the docs?
So we pointed the agents at our own documentation. Turns out they are pretty good at explaining Dawn. Revolutionary stuff, we know.
Why This Should Have Always Been a Thing
A new workspace raises small questions immediately, none of them hard, all of them in the way. Which agent should handle this work? What is the difference between a memory page and a connected source? Where do generated files appear afterwards? How do teammates use the same agent from Slack, Discord, Telegram, Teams, and Dawn Web at once?
Each question is small. Together, they slow onboarding and make the platform feel harder than it needs to be — exactly because the answer always lived somewhere else. Someone left the thread, opened the docs, asked an owner, or guessed. The thread waited.
The thing the agent could obviously have helped with was sitting there the whole time. Asking the agent in the thread keeps you in the work, gets you a workspace-aware answer instead of a generic one, and turns “what is this feature?” into “here is what to do about it next”. You do not have to choose between getting work done and learning the tool. You can do the second one in the middle of the first.
Ask Any Agent How Dawn Works
You can now ask a Dawn agent direct questions about Dawn itself, in the same plain language you would use with a teammate. “How should I set up this workspace for our engineering team?” “What is the difference between a memory page and a connected source?” “When should I run a prompt in the background, and when should I just wait?”
The answer comes back in Dawn terms, not generic chatbot terms. The agent can talk about workspaces, channels, integrations, memory pages, scheduled tasks, the files browser, prompt commands, background tasks, teammate accounts, and the difference between owner, manager, and teammate roles, using the same vocabulary that appears in the product. If you ask “what is the difference between a memory page and a connected source?”, the answer should not just define the two things; it should suggest which one fits the situation you described, and what to do about it.
Setup Help, Right When You Need It
Setup questions rarely arrive at a convenient time. They arrive in the middle of trying to finish something else. A product manager asks an agent to summarise Linear tickets and realises the integration is not connected yet. A support lead drafting a daily customer-escalations review notices this should be a scheduled task. An engineer wants Dawn to remember the team’s release process and is not sure whether that is a memory page, a prompt command, or notes inside a connected doc.
Those are the moments the new behaviour earns its keep. Instead of breaking context to find the answer, the user can ask the agent in the thread — and the agent will lay out the practical choices: connect a source if the information already lives in a tool, write a memory page if the team wants durable instructions the agent should always have on hand, build a prompt command if the workflow should be repeatable, or move the work into a background task if it does not need to block the conversation. The output is not a tour of features. It is a recommendation grounded in what the user just said they were trying to do.
This is also where cross-references between Dawn capabilities pay off. A “summarise customer escalations every morning” question naturally pulls in scheduled tasks, connected sources, and the right shape of prompt command. A “we keep retyping the same review wording” question pulls in prompt commands and the global command pack. The agent can explain how those pieces compose for the team’s specific case, without requiring anyone to read three docs and stitch them together.
Better Onboarding for New Teammates
New users should not have to learn the whole platform before getting value from it — and now, mercifully, they do not have to.
A new teammate joining a workspace can open a thread and ask the agent what it can help with, what to set up first, or how their Slack DMs end up linked to this account. The agent can explain the workspace they have just joined: which channels are available, which integrations are connected, what the default agent does, which prompt commands the team has configured, and how the identity-linking flow makes their DMs in Slack, Telegram, Teams, and Discord work as the same person.
That matters because every team uses Dawn a little differently. Some teams start with PR reviews, some with support triage, some build their workflow around scheduled tasks, some live mostly in Dawn Web while others stay in their channels. There is no single right path through the product, and there does not need to be. The agent can explain the options in the context of what the user is actually trying to do, and skip the rest until it becomes relevant.
Guidance for Owners and Managers
The same behaviour helps the people responsible for the workspace setup. Owners and managers tend to spend the first day translating an operational goal — “we want Dawn to help with release planning” — into a list of platform pieces: which agent, which connected sources, which memory pages, which prompt commands, which scheduled tasks, which channel surfaces, which model.
Asking the agent collapses that work into a conversation. “We want Dawn to help with release planning — what should I configure?” produces a workspace-shaped answer rather than a generic feature checklist. “Which parts of this workflow should be memory pages, commands, or scheduled tasks?” produces a recommendation that respects the actual moving parts of the team’s process. The agent will not pretend it can make every change for you, but it will tell you the next step, who needs to make it, and where in Dawn Web it lives.
Honest About What You Can Actually Do
Product guidance is only useful if it is honest about permissions. Dawn knows the difference between an owner, a manager, and a teammate, and the agent’s answers reflect that.
If a setting requires an owner, the agent says so instead of pretending the teammate can change it. If a connected source is not available to the selected agent, the agent explains what is missing rather than confidently misdescribing what would happen. If a request depends on credentials, integration permissions, or workspace configuration the current user cannot reach, the answer describes the steps and the right person to involve. The same applies to model settings: an agent on a model that does not support hosted web search will not pretend /web-research is available; it will explain why it is not, and what would change that.
The honest answer is sometimes “you cannot do this yourself, but here is the request to send your owner”. That is still a step forward from “Dawn does not know”.
Getting Started
There is nothing to enable. Open any thread in Dawn, with any agent, and ask a question about Dawn itself. Ask how to set up the workspace, what a feature is for, which workflow fits the work you are doing, or what to configure next. The agent will answer in Dawn’s own terms, suggest the concrete next step, and stay honest about what needs an owner.
In hindsight this was the obvious move. We are calling it an update anyway.