Research & Report Generation

CUA Research Workflows

CUAs are particularly powerful for research tasks that require browsing multiple sources, extracting information, and synthesizing it — without those sources having APIs.

The Research Pipeline

Input: Research brief ("Research Series B fundraising trends in AI infrastructure Q1 2025")

Agent workflow:

1. Search for recent funding announcements
2. Navigate to Crunchbase for deal data
3. Read TechCrunch and VentureBeat for context
4. Extract: company, amount, investors, valuation, date
5. Identify patterns across 10-15 deals
6. Generate structured report

Output: Formatted research report with citations

Instruction Design for Research Tasks

Research [topic]. For each source you visit:

1. Note the URL and date accessed
2. Extract the specific data points listed below
3. Note the publication date of the information
4. Flag any conflicting data between sources

Data to extract: [specific fields]

After visiting at least 5 sources, synthesize findings into a report with:

• Executive summary (3 bullets)
• Key findings (numbered list)
• Data table
• Source list

Handling Paywalls and Login Walls

CUAs encounter paywalls frequently. Your options:

1. Skip and note: "If content is behind a paywall, record the headline and source URL but don't attempt to access it."
2. Provide credentials: For subscriptions you have, provide login credentials in secure variable storage.
3. Use alternatives: "If the primary source is paywalled, find an alternative source covering the same topic."

Quality Control

Research CUAs can hallucinate or misread content. Build verification steps:

"After extracting each data point, verify it by re-reading the relevant section of the source. If you're not confident in a figure, mark it with [VERIFY] in the output."

Scheduled Research Agents

The most powerful deployment is scheduled: a CUA that runs weekly, researches your key topics, and delivers a briefing. This turns CUA research from a one-time task into an intelligence system.