Sometimes you don't need a workflow. You need one agent to go deep on a topic and come back with a structured report.
The setup
The fastest path: go to Create, describe the research you need, and Svortie builds the agent for you — picks the right tools, model, and settings automatically.
Or go to Agents and pick the Research Agent template. It comes pre-configured with web search, a structured research methodology, and sensible defaults:
- Model: Haiku (cost-effective, fast)
- Max iterations: 30 (enough for thorough research)
- Timeout: 600 seconds (10 minutes, gives it room to work)
- Web access: built in via the template's tools — no MCP server needed
The template pre-loads a system prompt that handles source verification, cross-referencing, and structured output. You can customize it, but it works well out of the box.
Running a research task
Set the task to something specific:
Research the current state of the AI agent platform market. Cover: major players (funding, team size, pricing model), market size estimates, key differentiators between platforms, and emerging trends. Write a comprehensive report to
/home/daytona/output/ai_agent_market_report.md.
Hit Run. The agent searches, reads, cross-references, and writes. Typical research runs take 2-5 minutes. Download the report from artifacts.
Prompting for better research
Be specific about scope. "Research AI agents" is too broad. "Research AI agent platforms that target non-technical users, launched after 2024, with public pricing" is focused enough to produce useful results.
Define the output format. "Write a report" is vague. "Write a report with sections for Market Overview, Player Profiles (table format), Pricing Comparison, and Trend Analysis" tells the agent exactly what you want.
Set boundaries. "Focus on US-based companies" or "exclude open source projects" or "only include companies with disclosed funding" keeps the research on track.
Ask for sources. Always ask for URLs. It lets you verify claims and dig deeper on anything interesting.
Attaching a playbook
For research you run repeatedly (industry updates, competitive scans, market tracking), attach a research playbook with:
- Source hierarchy (primary > secondary > tertiary)
- Output format rules
- Confidence level requirements
- What to include vs. skip
The agent follows the same methodology every time, making results comparable across runs.
When to use a workflow instead
If you find yourself wanting the research agent to also analyze and also format a presentation, split it into a workflow. Research, then analyze, then format. Each step is simpler and the overall output is better.
One agent is for focused research with structured output. Workflows are for multi-phase projects where different skills are needed at each stage.
Try it
Go to Create and describe a research topic, or pick the Research Agent template at Agents. Give it a topic you've been meaning to research. The report shows up in your artifacts in a few minutes.