Compressed Research
Also known as: research agent, research automation
What is Compressed Research?
Compressed research is an agent design pattern where an AI agent performs the research phase of a business process automatically, while a human still makes the final decision. The name was popularised by Malte Ubl, CTO of Vercel, who identifies it as one of the most practical and highest-ROI agent archetypes businesses can deploy today. Instead of replacing a human workflow end-to-end, compressed research preserves the existing process shape — business event → research → human decision — but collapses the research step from minutes or hours to seconds.
Why Compressed Research Works
The pattern’s power is that it doesn’t change the risk profile of a business process. The human still makes the final call, so there’s no new category of bad outcome. But the human arrives at the decision pre-briefed with relevant context that previously took 15 to 30 minutes to gather manually. Multiply that across 100,000 events per year and the savings are substantial — without requiring a company to re-engineer workflows, change compliance procedures, or retrain staff.
Example: Vercel’s Contact Sales Agent
When a visitor fills out the “Contact Sales” form on vercel.com, the message goes first to an agent. The agent checks the sender’s LinkedIn, researches the company, assesses company size and fit, and either routes the lead to the appropriate sales person with a briefing, or flags that the inquiry is actually a support question (which Malte notes is the case about 75% of the time). A human still takes the meeting — but they arrive with context that would have taken them a quarter of an hour to gather.
Vercel runs the same pattern for abuse reports: an agent gathers signals about a flagged site, and a human makes the final moderation call.
Where to Apply It
Compressed research suits any process with this shape:
- A business event arrives (a form submission, a support ticket, a new lead, an abuse report)
- A human researches the situation using publicly available or internal data
- The human decides what to do
The research step is the candidate for compression. Sales qualification, support triage, compliance review, recruiting screens, due diligence, and customer onboarding all fit naturally.
Related Reading
- AI Agents - The broader category of autonomous software
- Application Layer - Where compressed research sits in the AI stack
- Enterprise AI - Business deployment patterns