Control Without the Sprawl: How Mercor Unlocked Hundreds of Builders with Vercel

Highlights

• Mercor scaled from less than a dozen projects last year to 2,000 projects owned mostly by non-engineering teams.

• Vercel Passport enables hundreds of thousands of outside experts to be provisioned and off-boarded in seconds.

• AI Gateway served 80 million requests and 63 billion tokens in its first 10 days.

• LLM batch grading evaluation speed dropped from 12 hours to two hours after adopting AI Gateway.

• Every API key enforces a span cap, expiry, designated owner, and default zero data retention.

Timeline

1. 00:00 - 02:23

   Scaling Pressures and the Proliferation of Projects

   Domain experts like doctors, lawyers, and engineers supply data to fast-moving Frontier Labs. Because internal product and engineering teams could not keep up with growth demands, business teams began shipping their own solutions. This decentralized expansion increased the exposure surface to 2,000 projects. Requiring individual security reviews for each project would force teams to bypass controls, necessitating a platform-level paved path instead of security gates.

   • Mercor connects specialized domain experts with Frontier Labs.
   • Project counts surged from under a dozen last year to 2,000 current projects.
   • Manual security reviews for every project create untenable bottlenecks.
2. 02:23 - 04:50

   Default Identity and Automated Provisioning with Passport

   Application owners previously wasted time choosing between homegrown auth, library-based solutions, or risky public access. Passport standardizes logins and passkeys with zero friction for internal employees. For external experts, the audiences tool connects with Passport to grant temporary access to tens of thousands of users instantly. Once engagements conclude, purging these groups from the IDP revokes access automatically, satisfying security and engineering teams.

   • Vercel Passport eliminates custom authentication code by providing an IDP layer by default.
   • The audiences tool provisions large collections of external experts in seconds.
   • Removing groups from the IDP upon engagement completion enforces automated off-boarding.
3. 04:50 - 08:12

   Inference Management and Capacity Scaling via AI Gateway

   Around 100,000 experts rely on LLMs for core workflows, quality control, and data generation. Previously, switching models required tedious contract negotiations and rate-limit adjustments with specific inference providers. AI Gateway eliminated this friction, reducing batch grading run times from 12 hours to two hours. Furthermore, request-level attribution and unified billing provide finance teams with clear spend visibility and budget enforcement.

   • AI Gateway unifies open and closed source models under a single endpoint.
   • The platform handled 80 million requests and 63 billion tokens within its first 10 days.
   • Virtual API keys enforce span caps, expirations, owners, and zero data retention.
4. 08:12 - 13:15

   Autonomous Agents and Secure Platform Defaults

   As security, fraud, and internal teams deploy autonomous agents that operate tirelessly toward given goals, security risks multiply. Agents must possess isolated gateway keys, model restrictions, and expiration dates rather than inheriting user permissions. By anchoring security within Vercel's platform defaults and organizational controls, the company scales efficiently while maintaining builder velocity.

   • Autonomous agents require strict scoping, model safe lists, and spend caps to prevent runaway execution.
   • Making the secure path the easiest path prevents builders from creating custom workarounds.
   • Managed user settings prevent accidental public exposure of sensitive organizational data.