What Mechanize does

Mechanize, Inc. is a San Francisco-based startup founded in April 2025 that builds reinforcement learning (RL) environments and automated evaluation benchmarks for training and testing frontier AI coding agents. Its core product is virtual environments that simulate real software-engineering work — including building features, deploying applications, and debugging code in unfamiliar codebases — paired with automated grading systems that score model performance using a combination of procedural tests (unit, integration, end-to-end) and LLM-based rubric evaluation. These environments are sold directly to frontier AI labs such as OpenAI, Anthropic, and Google DeepMind for use in reinforcement learning training and capability measurement. While its current focus is software engineering, the company has stated a long-term mission of enabling the full automation of valuable work across the economy.

The company operates a custom-enterprise B2B sales motion, with pricing not publicly disclosed but structured around multi-year contracts to a narrow set of frontier AI research organizations. Revenue mechanics are anchored on the company's prediction that AI labs will pay $500–$2,000+ per high-quality RL task as compute costs rise, positioning Mechanize to scale revenue alongside growing demand for realistic coding-agent training data. The company employs approximately 35 people in-person in San Francisco, with each RL task owned end-to-end by a single specialist engineer who handles ideation, grading implementation, and quality assurance.

Mechanize's go-to-market is supplemented by content marketing through technical essays and a podcast series targeting the AI research community. The company is led by Tamay Besiroglu (CEO), Ege Erdil (CTO), and Matthew Barnett (President), and is backed by a roster of prominent tech and AI investors including Nat Friedman, Daniel Gross, Patrick Collison, Adam D'Angelo, Marco Mascorro, Dwarkesh Patel, Sholto Douglas, Devendra Chaplot, Alex Atallah, and Marcus Abramovitch. In April 2026, Mechanize raised $9.1M in seed funding at a $500M post-money valuation, and it publishes the public GBA Eval benchmark as a demonstration of its evaluation methodology.

Name

Mechanize

Legal name

Mechanize, Inc.

Website

https://mechanize.work

Company type

Private

Founded year

2025

Operating status

Operating

Headcount range

11–50 employees

Short description

Mechanize, Inc. builds reinforcement learning environments and automated evaluations for frontier AI coding agents, selling custom environments to frontier AI labs like OpenAI, Anthropic, and Google DeepMind for model training and capability measurement.

Ownership category

akta.pro rank

Product category

AI Training Infrastructure

NAICS

Custom Computer Programming Services (541511)

SIC

Services-Prepackaged Software (7372), Services-Computer Programming Services (7371)

akta.pro primary industry

RLHF, Human Feedback & Evaluation Data Tools (HDAAALAJ)

akta.pro secondary industries

Safety, Alignment & Content Moderation (Guardrails, Red Teaming) (HDAAACAI), Safety & Alignment Evaluation (red-teaming, harmful capability testing) (HDAAAMAL)

Keywords

• Reinforcement learning environments
• AI coding agents
• Model evaluation
• RL infrastructure
• AI training data

Headquarters

HQ city

San Francisco

HQ country

United States

HQ region

North America

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Personnel, Technology or R&D, Infrastructure

Revenue model

1. RL Environments and Evals: Mechanize builds and sells reinforcement learning environments and evaluation benchmarks to frontier AI labs for training and testing coding agents. AI labs use these environments to train models via reinforcement learning or to measure capabilities. Revenue comes from selling these custom-built environments to AI research organizations.

Pricing tiers

Go-to-market motion2 records

Distribution channels1 record

Marketing channels4 records

Core offering

Mechanize builds reinforcement learning (RL) environments and evaluations for frontier coding agents. The company develops realistic software engineering task simulations and benchmark systems that allow AI labs to train, evaluate, and improve autonomous coding systems at scale.

Product overview

Mechanize builds environments and evaluations (evals) for frontier coding agents. The core offering consists of custom RL environments that simulate real software engineering work—where AI agents build features, deploy applications, and debug code—and automated evaluation systems (graders) that score agent performance using procedural tests and LLM-based rubrics. These environments and evals are bundled together and sold to frontier AI labs for model training and capability measurement. The company also offers a public benchmark called GBA Eval as a demonstration of their evaluation methodology.

Differentiator

Problem solved

Functional benefit

Products and services

• RL Environments for Coding Agents
• Evaluations and Benchmarks for Coding Agents Evaluation and benchmarking systems that measure the performance of frontier AI coding agents on software engineering tasks.

Quantifiable outcome

• GBA Eval benchmark shows Claude Opus 4.8 can write a Game Boy Advance emulator from scratch in 24 hours
• +1 more outcomes

Named customers3 records

Segments2 records

Ideal customer profiles1 record

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability6 records

Feature4 records

Scale indicators5 records

Recent moves5 records

Expansion highlights2 records

Direct peers

• Surge AI: Surge AI is the most directly comparable peer — it provides RLHF data labeling, RL environments, and expert human feedback for training frontier AI models, including RL environments for coding agents. Overlaps almost perfectly with Mechanize's core offering of RL environments and evals for AI labs.
• Mercor: Mercor is an expert contractor marketplace that supplies domain experts (engineers, lawyers, etc.) for AI training data and RL tasks. It is explicitly cited as a competitor in Mechanize's competitive landscape and competes for the same expert-task-creation work for frontier AI labs.

Broad incumbents

• Scale AI: Scale AI is the largest AI training data provider, offering data labeling, RLHF services, and evaluation products to frontier AI labs. While Scale is much broader and serves many model types, it directly competes with Mechanize on RL data and evaluation services for coding agents.
• Appen: Appen is a large incumbent data annotation and AI training data company serving enterprise and AI lab customers. While broader and less specialized than Mechanize, it competes in the same training data and RL data supply category.

Emerging players

• Labelbox: Labelbox provides a training data platform for AI teams, including annotation workflows, evaluation tools, and RLHF pipelines. It is comparable as an enabling platform for AI training data and evals, though it serves a broader set of use cases than coding-agent-specific RL environments.
• Snorkel AI: Snorkel AI builds programmatic data labeling and AI data development platforms, including RL data tooling. It is comparable as a peer in the AI training data tooling space, though its focus is broader than Mechanize's coding-agent-specific RL environments.
• Mostly AI: Mostly AI specializes in synthetic data generation for AI training. While focused on synthetic tabular/structured data rather than RL environments, it competes in the broader synthetic training data category relevant to AI lab training pipelines.
• Synthesis AI: Synthesis AI provides synthetic data for computer vision and ML training, with overlap in the synthetic training data category relevant to AI labs building RL training pipelines. It is thematically comparable as synthetic training data infrastructure.
• Galileo (AI evaluation): Galileo provides AI evaluation, observability, and quality measurement tooling for LLM applications. It overlaps with Mechanize's eval product line for measuring AI model performance, though Galileo's focus is broader application evaluation rather than RL environment evals.
• Inworld AI: Inworld AI builds simulated environments for training AI agents (originally NPCs, increasingly coding/enterprise agents). It is comparable as a peer building training environments for agentic AI, though its primary historical focus has been character simulation rather than software engineering tasks.

Market position

Strengths4 records

Weaknesses5 records

Competitive moat3 records

Key risks6 records

Key highlights7 records

Customer concentration

• Xx.com/MechanizeWork
• LinkedInlinkedin.com/company/mechanize-inc

Revenue estimate

Valuation estimate

Number of profiles

Profiles3 records

Funding overview

Funding rounds2 records

Investors

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Mechanize