The race to AGI
What is driving progress toward more capable AI, what remains difficult, and how to distinguish meaningful advances from hype.
Company
Media & press
Expert commentary, research, and resources for journalists.
Featured expert
Dr. Nazneen Rajani
Cofounder & CEO, Collinear AI
Dr. Nazneen Rajani is cofounder and CEO of Collinear AI, which develops training data, evaluation tools, and realistic environments to help frontier labs rapidly improve their AI models. Her research spans model evaluation, alignment, and agent reliability. Previously, she was a research lead at Hugging Face and a senior research scientist at Salesforce Research. She holds a PhD in computer science from UT Austin.
Dr. Nazneen Rajani is cofounder and CEO of Collinear AI, which develops training data, evaluation tools, and realistic environments to help frontier labs rapidly improve their AI models. Her work examines how models learn from feedback, how their capabilities are measured, and why AI agents fail when faced with the complexity of real work.
Before founding Collinear, she was a research lead at Hugging Face and a senior research scientist at Salesforce Research. She is a coauthor of research including YC-Bench and TraitBasis, spanning long-term planning and simulations of human behavior. She holds a PhD in computer science from UT Austin and was named to MIT Technology Review's 2024 Innovators Under 35 list.
Areas of expertise
Available for expert commentary on:
For interviews, expert commentary, and media requests, contact press@collinear.ai.
Simulated worlds and AI learning
How realistic environments, human behavior, and reinforcement learning shape the next generation of AI.
AI agents in the real world
What it takes for agents to plan, use tools, adapt, and reliably complete complex work.
Measuring and improving AI
What benchmarks reveal, what they miss, and how training data and feedback translate into better capabilities.
Cybersecurity, coding, and computer-use agents
How AI performs practical work, from finding and fixing vulnerabilities to writing software and operating applications, and where its capabilities still fall short.
Selected research
From research to real capability.
2026
Cybersecurity benchmarkCWE-bench
Testing coding agents on their ability to find and fix vulnerabilities in real codebases.
Explore benchmark2026
Agent evaluationYC-Bench
Benchmarking AI agents for long-term planning and consistent execution.
Read paper2025, revised 2026
Human behavior simulationTraitBasis
Impatient Users Confuse AI Agents: high-fidelity simulations of human traits for testing agents.
Read paperIn the conversation
Coverage & interviews.
Quanta MagazineApril 30, 2025
Expert commentary
When ChatGPT Broke an Entire Field: An Oral History
The Free Press JournalJanuary 6, 2025
Interview
“AI Is a Tool, Not a Threat,” Says Nazneen Rajani
MIT Technology Review2024
Recognition
Innovators Under 35: Nazneen Rajani
The New York TimesSeptember 25, 2023
Expert commentary
The Secret Ingredient of ChatGPT Is Human Advice
Portraits
Full-resolution photograph and square profile image.
Company logos
Full-color, black, and white logos for light and dark backgrounds.
Bios & company information
Short and extended biographies, company description, and media contact.
For photo-use questions, contact press@collinear.ai.
About Collinear AI
Data for faster model improvement.
Collinear AI provides data to frontier labs for faster model improvement. The company develops challenging tasks, realistic environments, and rigorous evaluations that help frontier AI teams identify capability gaps, generate useful training data, and measure progress. Its work spans cybersecurity, software engineering, computer use, and complex agent behavior, combining off-the-shelf task libraries with research expertise to accelerate model improvement cycles.