Refresh GTM brain

Refresh builds simulated coding and computer-use environments for AI labs and enterprises to train and evaluate agents.

category

Agent reinforcement learning environments

market profile

  • text: Frontier labs need realistic software tasks and objective rewards to train and compare agents.
  • text: Enterprises can evaluate agents against their own workflows before trusting them with production work.

personas

  • title: Head of post-training at a frontier AI lab
    pains:
    • Models plateau on benchmarks that fail to represent hard real work
    • Producing verifiable, diverse environments requires scarce engineering and domain expertise
  • title: Enterprise AI agent engineering lead
    pains:
    • Cannot reliably tell whether agents complete company-specific workflows
    • Live systems make agent experimentation risky and hard to repeat
  • title: AI evaluation researcher
    pains:
    • Subjective grading masks actual task failures
    • Needs reproducible comparisons across models

icp

  • text: Frontier model developers with dedicated evaluation and reinforcement-learning teams are primary buyers.
  • text: Enterprises building agents for valuable multi-application workflows are an adjacent segment.

message pillars

  • text: Train in realistic software worlds.
  • text: Evaluate success using objective signals.
  • text: Focus tasks on real frontier model gaps.

objections

  • objection: Are simulated tasks representative of production?
    response: Review fidelity of cloned applications and task-specific verifiers, then pilot on workflows that matter to the buyer.
  • objection: Can you accommodate our proprietary workflow?
    response: Discuss a bespoke dataset and agree a measurable task set and evaluation criteria.

competitors

  • name: Aviro
    why: Builds coding and computer-use reinforcement learning environments and post-training datasets.
  • name: Toloka
    why: Provides expert data, custom evaluations and RL environments.

influencers

  • name: Christopher Settles
    url: https://www.chrissettles.com/about
    why: Refresh co-founder with a public technical profile and frontier evaluation experience.

keywords

  • reinforcement learning environments for agents
  • computer use training environments
  • coding agent evals
  • verifiable rewards AI
  • agent benchmark enterprise workflows
  • synthetic software environments

voice

  • Technical and research-oriented
  • Precise about measurement
  • Ambitious

signals

  • text: A lab is hiring for post-training, agent evaluations or reinforcement learning.
  • text: An enterprise is piloting computer-use agents across applications and needs repeatable evaluation.

stack

  • Harbor
  • computer-1
  • Blender

hiring

  • GTM Engineer
  • Research Engineer, Benchmarking
  • Ex-Founder

pricing

model: unknown

motion

sales_led

channels

  • research publications
  • technical benchmarks
  • founder network

confidence

medium

Founders

  • name: Christopher Settles
    title: Founder/CEO
    role: CEO
    bio: Co-founder and CEO @ Refresh | ex-ML @ Uber, CS @ UIUC.
  • name: Erik Quintanilla
    title: Founder/CTO
    role: CTO
    bio: Co-founder and CTO @ Refresh | ex- @Amazon @CapitalOne Previous focus has been in vision data, voice data, tabular data in the form of data pipelines. Web scale scraping previously.

Back to YC companies