Featured runs 100,000 monthly AI media pitches on Vercel with three engineers

Three engineers behind 100,000 monthly pitches

Featured, a platform that matches subject-matter experts with journalist requests, podcasts and awards, says its software delivers more than 100,000 media pitches every month and connects a user with a journalist or publisher roughly every six seconds. Behind that volume, according to a case study published on Vercel's blog, is an engineering team of just three people supporting three brands and more than 100,000 users.

The company also owns Help A Reporter Out (HARO) and Connectively, and has sent over 100 million HARO emails in the past year, the post states. Founder Brett Farmiloe previously spent a decade running Markitors, a marketing agency with 500 small-business clients, where he observed that companies had genuine expertise but no practical route to journalists. PR, in his words, has historically been "about who has access to what" — the premise of Featured is to shift that to who actually has the knowledge.

The infrastructure that could not keep pace

The case study describes a familiar pre-migration picture: hosting on AWS Elastic Beanstalk, AI features built on bespoke provider integrations, and long-running multi-step jobs on separate orchestration infrastructure. Each layer functioned, but together they consumed operational attention a three-person team could not spare.

The trigger for change was a hard deadline. Featured had to launch 374 Sanity-based sites simultaneously, and every Elastic Beanstalk deployment required manually creating an environment, migrating the URL and tearing the old one down. Multiplied across nearly four hundred sites, the process was unworkable. After the move to Vercel collapsed each launch into a one-click deployment, the team went further and migrated the entire application and its agents to the platform.

One abstraction over 17 models

Featured now routes its model traffic across 17 models at any given time through Vercel's AI SDK and AI Gateway. According to the case study, this gives the team a single standardized calling interface with a schema, so switching providers becomes a configuration change rather than a code rewrite. When a new model is released, the team can immediately test it against their own use cases without maintaining per-provider rate-limit or API logic.

Workflow SDK replaces custom orchestration

The workloads that cannot fit inside a request cycle — continuous media monitoring, qualifying opportunities, and combining those signals into pitches — previously ran on a separate orchestration system. Featured moved them onto Vercel's Workflow SDK, consolidating the pipeline that produces its monthly pitch volume onto the same platform as the rest of the application.

Conversation replaced the dashboard

As users increasingly chose to converse rather than navigate menus, Featured made chat its primary interface. The company built it with eve, Vercel's open-source agent framework, whose useEveAgent hook supplied session durability, streaming, tool calling and approval prompts without custom wiring, with model calls routed through AI Gateway so the right model can be selected per task. What the team expected to be a months-long rebuild, they describe as roughly a week of replacement and refactoring — the product is now a question box rather than a dashboard of buttons.

Looking ahead, Featured aims to become the agent layer for public relations, analogous to what has emerged in legal and finance. The team's advice to other builders in the agentic era: treat your data as the moat, and rely on proven abstractions instead of hand-rolling security and infrastructure.

Why it matters

This is a useful data point for small cloud and platform teams facing the same trade-off Featured did: every hour spent operating infrastructure is an hour not spent on product. The case shows how consolidating compute, model access and long-running jobs behind managed abstractions let a three-person team run a high-volume, multi-brand workload that would traditionally demand a larger operations function.

Two caveats are worth keeping in mind. First, this is a vendor-published case study, so the numbers and framing come from Vercel and its customer rather than independent verification. Second, Featured's workload — many small sites, bursty AI calls, asynchronous background jobs — fits a serverless-style platform well; teams with different constraints may not see the same leverage. Still, as a scaling pattern of few engineers, one platform, and standardized AI plumbing, it is a template others can evaluate.

tags