Field study 02 · Recurring intelligence
A daily brief that arrives already researched and filed
A scheduled workflow asks an agent for fresh analysis, saves one canonical document, and posts the useful summary back to the team.
The situation
Teams often ask for the same synthesis every morning: market movement, competitor updates, customer signals, policy news, or project health. The research changes; the shape of the work does not. Running it by hand creates inconsistency. Running it as one opaque prompt produces an answer with nowhere durable to live.
A Dana4 workflow separates the job into observable steps. Research is an agent task. Publication is an explicit document step. Distribution is a message step. Each run is visible, and the final artifact has a deterministic key so a retry does not create a second copy.
- An invited agent produces the dated analysis and a concise summary.
- A document step writes the canonical brief to the workspace.
- A message step posts the summary and document link to the run channel.
- Search and topic navigation make previous briefs available to future work.
Why the explicit document step matters
Agent steps should produce structured run output, not a pile of sidebar documents named after their capability. The workflow decides which output becomes a user-facing artifact.
For a daily brief, that means one document with a stable key such as
market-brief-{{run.date}}. If the worker retries the document step for the same run, it resolves
to the same logical artifact. Intermediate research remains inspectable in the run without
cluttering the workspace.
A practical brief contract
Ask the research step for a predictable result:
- a descriptive title;
- the date and scope;
- three to five findings;
- source or evidence notes;
- what changed since the previous brief;
- uncertainties or missing data;
- a two-sentence channel summary.
The next step can reference the structured agent output with
{{research.output}}. Dana4 materializes the value only after the declared dependency completes,
so the downstream task receives the actual result rather than the template expression.
Good fits
| Team | Recurring brief |
|---|---|
| Product | Customer feedback themes and competitor releases |
| Investment | Market, portfolio, and regulatory signals |
| Operations | Exceptions, blockers, and decisions due today |
| Marketing | Category news, campaign performance, and cultural moments |
| Research | New papers, claims, evidence quality, and open questions |
The workflow should narrow the source universe and audience. “Everything important today” is not a brief; it is an unbounded research problem.
Human authority without a human bottleneck
Not every recurring brief needs a review gate. A low-risk internal digest can run end to end and let a person correct the resulting document if needed. Higher-stakes outputs should stop before external publication and make ownership explicit.
The current reliable pattern is to automate research, filing, and internal distribution while keeping consequential decisions outside the automatic path.
What to look for
- Exactly one final document is created per logical run.
- The channel announcement contains rendered content, not a
{{step...}}placeholder. - A retry does not duplicate the task or document.
- The run shows which agent produced the research and what input it received.
- A teammate can find yesterday’s brief without knowing its path.
Try it
- Invite an agent with a general research capability into the workspace.
- Create a three-step workflow: agent,
document_create, andmessage. - Give the document a deterministic key that includes
{{run.date}}. - Reference the agent output from both downstream steps.
- Run it manually once before attaching a schedule.