Vibeleaderboard

Benedict Brady

Viber rank914

Benedict Brady is a member of technical staff at The Bot Company and the creator of Optimization Arena, now Paradigm Puzzles. He writes about evaluating agents, automated research and the engineering practices needed for long-running software tasks.[1][2][3]

Start with Learnings from Optimization Arena →
Benedict Brady

Education

Harvard University

BA, Statistics

[4]

Links

Personal website X GitHub LinkedIn

Member of Technical Staff · The Bot Company

Career

Brady previously worked in quantitative research at Cumberland, part of DRW, and helped build the DeFi trading firm Ergonia. He founded Meridian, an interface for financial markets, before joining The Bot Company in August 2026 to work on machine learning and robotics. His former Meridian biography records a Harvard statistics degree; his current personal site gives the later career chronology.[1][4][5]

Contributions

Optimization Arena gives participants a reproducible environment and a numerical objective. Brady’s account emphasizes that strong challenge design needs an elegant verifier, enough realism to be useful and few opportunities to overfit accidental details. His experience running the competitions connects agent orchestration with the design of an experiment and its measurement. The project moved to Paradigm as Paradigm Puzzles in July 2026.[2][5]

Ideas

In Notes on the Software Factory, Brady argues for giving agents development tools, tests and feedback rather than stopping at text generation. He discusses critics, memory, integration tests and the bottlenecks that limit longer tasks. The essay is a practical account of his workflow and its remaining weaknesses, including business context, continual learning and always-on operation.[3]

Tools, writing & talks

Writing, research & talks

  • Start here

    A practical account of designing verifiable challenges, evaluating submissions and building agent research workflows.

    Read the explanation of how a challenge is scored, then compare it with the discussion of overfitting and verifier design. It is a useful introduction to evaluating an agent by reproducible outcomes instead of how convincing its answer sounds.

    Author and challenge organizer[2]
  • Notes on tools, critics, memory and testing for agents carrying out longer software-development tasks.

    Author[3]

Apps

References

Latest source check 2026-10-30

  1. [1] Owned website and career chronology

    www.benedict.dev · Checked 2026-10-30

  2. [2] Learnings from Optimization Arena

    www.benedict.dev · Checked 2026-10-30

  3. [3] Notes on the Software Factory

    www.benedict.dev · Checked 2026-10-30

  4. [4] Former Meridian official biography

    meridian.app · Checked 2026-10-30

  5. [5] Owned professional X account

    x.com · Checked 2026-10-30

  6. [6] Owned GitHub profile

    github.com · Checked 2026-10-30

  7. [7] Shape Packing Challenge original repository

    github.com · Checked 2026-10-30

Benedict Brady | VibeLeaderboard