Stop Researching Prospects Manually. Build an Agent That Does It in 30 Seconds.
Cold email response rates collapse when you're generic. We built a Runtype agent that searches the web with Exa, finds real personalization hooks, and drafts an email in 30 seconds. Here's how it works.
The best cold emails reference something real. Not "I admire your company." Something specific: a blog post published last Tuesday, a product launch from last month, a job posting that signals a strategic bet.
The problem is that finding those specifics takes 45 minutes per prospect. Open LinkedIn, Google News, their engineering blog, their changelog, maybe their Twitter. Synthesize it. Figure out which detail is actually worth mentioning. Then write an email short enough to get read.
Most people skip it. They send the generic version. Response rates suffer.
This template deploys a Runtype agent that does the research and drafts the email in about 30 seconds. Here's how it works.
The core idea
The agent has one job: before writing anything, search the web.
Give it a company name and what you're selling. It runs 2–3 targeted Exa searches using neural retrieval, finding recent, relevant content rather than relying on cached knowledge. It reads the results, surfaces 3–6 real signals (a blog post, a product announcement, a hiring pattern, a statement from their CEO), ranks the strongest personalization hooks, and drafts a cold email that opens with something specific from their world.
The email comes with the research attached. You can see exactly what the agent found and why it chose the hook it used. No black box, no fabricated facts.
Why Exa specifically
There's a temptation to just skip the search step and ask the model directly: "What do you know about Acme Corp?" The problem is that training data is months stale by the time you're using it. Anything that happened recently isn't in the model's memory. A product launch, a funding round, a strategic pivot, a blog post that went viral.
Exa is designed for exactly this. It indexes the web with neural retrieval, which means it finds semantically relevant content rather than just keyword matches. Ask it about a company's recent announcements and it surfaces their actual press releases, engineering blogs, and product pages, instead of a Wikipedia summary from two years ago.
In Runtype, Exa is a built-in tool. You wire it into an agent the same way you'd add any capability. No API keys to manage, no separate account to set up. The agent calls it, reads the results, and reasons from real evidence.
What the output looks like
Here's an actual exchange:
User: Research Vercel. I'm selling a developer analytics platform that helps teams measure AI feature adoption.
Agent:
Research Summary
- Vercel shipped v0 4.0 in April 2026, adding multi-model prompt chaining — AI is core to their platform direction
- Published "Measuring AI in production: what we wish we'd tracked from day one" last week — this is a live internal problem, not theoretical
- 14 open roles including two "AI Platform Engineer" positions — the team is scaling and needs instrumentation
- CEO Guillermo Rauch spoke at Vercel Ship 2026 about not knowing whether developers actually use AI suggestions
Top Personalization Hooks
- Their own blog post on measuring AI in production — they're already thinking about this problem, you're not creating new need
- Guillermo's talk cited "do developers actually reach for AI suggestions?" as an open question — your platform answers it directly
- New AI Platform Engineer hires signal they need to measure before they scale further
Email Draft
Saw your post on measuring AI in production — specifically the part about not knowing which AI features developers actually reach for. We built a platform for exactly that: adoption funnels, session replay, and model usage dashboards scoped to AI features. Happy to show you what it looks like on a codebase Vercel's size. 15 minutes?
That email is 63 words. It references a specific blog post published in the last week. It connects the prospect's stated problem directly to a specific capability of what's being sold. It asks for 15 minutes, not a demo.
That kind of specificity doesn't come from templates. It comes from research. The agent does the research.
How to build this on Runtype
The architecture is simple. One agent, one built-in tool, one chat surface.
The agent gets a system prompt that defines the research playbook: which search queries to run, how to rank the signals, and how to write the email. The prompt instructs it to never fabricate facts. If Exa doesn't find anything recent, the agent says so rather than inventing details.
The Exa tool is added to the agent as builtin:exa. Runtype's platform key handles authentication. The agent calls it during its tool-use loop, reads the results, and continues reasoning.
The Persona chat surface ships with the template. Drop the widget into your CRM sidebar, your sales team's internal portal, or any page with a script tag. Every rep gets the same agent. No one has to remember to do the research.
Deploy the template, and the Persona widget is live. Share the URL with your team. You're done.
This is the gap Runtype closes
Sales teams already know this. The best reps research every prospect before reaching out. The blocker has never been the model. It's been the research. The model is only as good as what it knows about the person it's talking to.
An agent doesn't close deals. It does the research so your reps spend their time on the 15-minute call, not the 45-minute tab sprint before it. Every rep on the team gets the same research quality on every prospect, every time. No senior researchers required.
That's the pattern Runtype is built for: the structure between AI generation and the real thing you're trying to ship. Here, the real thing is a cold email that actually converts. The agent handles the part that was slowing everyone down.
What's next
The template ships as a standalone chat widget, but it's designed to extend. A second capability could post the research summary and email draft directly to a HubSpot contact or Salesforce lead. No copy-paste; research lives in the CRM automatically. Add Firecrawl alongside Exa to scrape the prospect's pricing page or job board. Point the widget at specific verticals with a tuned system prompt.
The research playbook is just a system prompt. You can tune it to your ICP, your sales motion, and which signals matter most for your team.