Brendan Foody is cofounder and co-CEO of Mercor, which connects domain experts with AI training and evaluation work. He also coauthored APEX-SWE, a benchmark for software-engineering tasks involving integration and production debugging.[2][5]
Start with APEX-SWE →
X · BrendanFoody Career
Georgetown’s original student-founder record describes Foody starting Seros, a cloud-PC business, in November 2021. It also records a student pitch for the earlier Mercor idea. Mercor’s current founding account dates the company to 2023 and identifies Foody and Adarsh Hiremath as co-CEOs. Its focus is organizing professional judgment into training material and tests for AI systems.[3][2]
Contributions
Evaluate the integration work around a coding task
APEX-SWE contains two kinds of work: integrating systems across cloud services and business applications, and investigating production failures through logs, dashboards and unstructured context. Each category has one hundred tasks. The coauthored paper evaluates frontier models on completing those tasks, and releases a harness and smaller development set. Its analysis emphasizes checking assumptions and verifying facts before acting, connecting coding performance with the surrounding investigation and coordination work.[5]
Turn professional judgment into a repeatable test
Mercor’s enterprise methodology asks experts to define tasks and pass/fail criteria against a real workflow or a simulated environment. The same suite can compare models, tools and context configurations while recording execution cost. This makes the quality threshold explicit before choosing a cheaper configuration. The company distinguishes live production rubrics from offline APEX benchmarks, since they answer different questions about a deployed agent.[6]
Tools, writing & talks
Writing, research & talks
Start here
APEX-SWEarxiv.orgA benchmark covering cloud integration and production observability work, with a public harness and development set.
Read the task definitions and compare integration with observability before interpreting a score. Software-engineering and production-debugging experience helps.
Coauthor[5]- Why experts writing AI evals is creating the fastest-growing companies in history | Brendan Foodyyoutube.com
Mercor's Brendan Foody explains why expert-authored evals and RL environments have become the lab bottleneck, and how a labor marketplace for professionals scaled around that need.
Catalog workOriginal source - Brendan Foody on Expert-Written AI Evaluationslennysnewsletter.com
A chaptered interview about expert data, evaluation and the changing work around model training.
Interview guest[7]