Agent inbox: sending instructions to agents
Learn how Dailybot’s agent inbox lets managers and teammates queue instructions, tasks, and context that coding agents receive on their next check-in.
Most agent tooling is built around one direction: the agent reports out, and humans read the summary. That is useful, but it leaves a gap. When someone needs to steer the agent, they often resort to Slack threads, comments in a pull request, or hoping the right person updates a local file before the next run.
Dailybot closes that gap with the agent inbox: a structured way for managers and teammates to send instructions, tasks, or context to coding agents through Dailybot—not only to receive updates from them.
Reverse the default flow
Think of it as reverse communication. Standups and agent reports answer “what happened?” The inbox answers “what should happen next?” and “what do we need you to know?” That symmetry matters for teams where agents run on schedules, in devcontainers, or across time zones. You are not always at the keyboard when the agent starts; the inbox makes sure your intent still arrives at the right moment.
What the agent inbox is
The agent inbox is a queue of messages meant for agents connected to your Dailybot workspace. Anyone with permission to collaborate on that workspace can leave items for a given agent or team context: short directives, links to specs, bug IDs, or narrative context (“we deprioritized the API refactor; focus on the billing bug first”).
Those messages are not lost in chat history. They stay associated with Dailybot’s view of the agent and the team, so the next run has a single place to look for human input.
How delivery works when the agent checks in
When an agent checks in—whether that means running a Dailybot CLI report, reading workspace context during startup, or another supported integration step—it can pull queued inbox items. Ordering is preserved so earlier messages are handled before later ones unless your workflow merges or deduplicates them.
Practically, that means:
- You can post instructions before the agent’s next session; they wait in the inbox.
- Multiple teammates can contribute; the queue reflects the order messages were added (or your team’s rules for prioritization).
- The agent’s outbound report still describes what it did; the inbox supplies what humans want it to do next.
Exact mechanics depend on how your agent is wired (CLI, API, or orchestration around Dailybot), but the model is consistent: queue on the human side, consume on the agent’s next pull.
Use cases teams care about
Assigning tasks. Product or engineering leads drop the next ticket or outcome into the inbox instead of opening a new chat thread the agent might never see.
Context for ongoing work. Paste a decision log, a design link, or constraints (“do not touch the legacy webhook path”) so the agent does not rediscover old assumptions.
Redirecting mid-session. Priorities change. A queued override can steer the next run without requiring someone to edit repo instructions before the agent restarts.
Priority overrides. When something is urgent, an inbox message can flag it so it surfaces at check-in—complementing your normal backlog order.
Completing the communication loop
Agents that only report out leave humans reactive: reading summaries after the fact. The inbox adds the inbound leg: humans instruct, agents execute and report, humans adjust again. That loop matches how real teams work—async, documented, and visible in one place.
If you are already using Dailybot for standups and agent visibility, the inbox is the natural next step: same workspace, clearer steering, less friction between people and autonomous coding workflows.
FAQ
- What is the agent inbox in Dailybot?
- The agent inbox is a human-to-agent channel in Dailybot. Managers and teammates can send instructions, tasks, or context that are queued for coding agents and delivered when those agents run their next check-in or reporting flow, so people can steer work without chasing the same terminal session.
- How are instructions from the inbox delivered to an agent?
- Messages are stored in a queue tied to your workspace and agent identity. When the agent performs its next check-in (or the step where it pulls team context from Dailybot), queued items are surfaced in order. The agent can then incorporate them into its plan for that session, similar to reading assigned work from a shared inbox rather than only pushing updates outbound.
- What are typical use cases for the agent inbox?
- Teams use it to assign the next task, drop context for ongoing work (links, decisions, acceptance criteria), redirect an agent mid-initiative when priorities change, and send priority overrides so urgent work is visible before the next autonomous stretch of coding.