That was Go champion Fan Hui’s reaction to Move 37—a moment that left the world in awe. It was the move that defined the 2016 showdown between Go legend Lee Sedol and DeepMind’s AI agent, AlphaGo. In that historic match, AlphaGo dominated, winning four out of five games, marking a pivotal milestone for artificial intelligence: that AI could reach superhuman levels of capability.

Ioannis Antonoglou, Reflection’s cofounder and Deepmind founding engineer behind AlphaGo, recalls the moment vividly: “Move 37 was spectacular … a brilliant, unconventional move that showed the system had a deep understanding of the game. It had creativity. It could think of things that people hadn’t thought of before.”

Ioannis continued pushing the frontier. He extended AlphaGo’s capabilities with AlphaZero—an AI that learned entirely from self-play, mastering the game without human guidance. Then, he led the development of MuZero, an even greater breakthrough: an AI that conquered Go, chess, Shogi, Atari and more—all without even being told the rules of the game.

These weren’t just incremental advances. They were seismic shifts in AI that showcased the significance of reinforcement learning in building superhuman agentic systems. Even today, AlphaGo, AlphaZero and MuZero remain amongst the most powerful AI agents ever created.

From AlphaGo to Reflection AI

At DeepMind, Ioannis went on to play a key leadership role in post-training efforts for Gemini, where he met Reflection cofounder Misha Laskin, who was leading reward model development. Together, they continued to push the boundaries of two of the most critical fields in AI: reinforcement learning and large language models.

Their shared vision for superintelligence, and their elite domain expertise in building both the most capable agents and the most powerful LLMs, led to the birth of Reflection AI.

The vision of Reflection is to build superintelligent agents to conduct all knowledge work. Crucially, Reflection is built on two core beliefs:

1. That autonomous coding is a crucial step toward superintelligence.
2. That real-world evaluations—not just benchmarks—matter most.

The company’s first focus? Building autonomous coding agents. 

Autonomous coding agents

We’ve seen the incredible benefit of coding assistants in enabling developers to work with 10x speed and productivity. In parallel to this, we believe that the next leap forward is autonomous coding agents that handle workloads autonomously.

Reflection’s autonomous coding agents integrate directly into an organization’s codebase and engineering workflows, and autonomously tackle well-scoped engineering tasks entirely end-to-end. They read, write, test and deploy code, handling all the infrastructure on your behalf, relieving the burden of entire engineering workloads from developers and teams.

The reality is, rather than writing new code, most engineers spend the majority of their time buried in refactoring, debugging and fixing bugs; code reviews, standups and sprint planning; updating documentation and learning new tools, languages and frameworks; and managing infrastructure, devops and testing. Reflection’s autonomous coding agents start with these sets of tasks: important, yet tedious, time-consuming work that often sits in an engineering team’s backlog.

Crucially, Reflection is centered on reliability. To start, Reflection’s agents are not all-encompassing software engineers. Rather, they are agents that engineering teams and leaders design to handle specific sets of tasks and workloads, with trusted performance:

• Refactor legacy code safely, without breaking dependencies
• Generate and run test cases automatically
• Optimize memory usage across pull requests
• Find and fix security vulnerabilities before they cause problems
• And more!

As the team advances model intelligence to increase its scope of capabilities, Reflection’s agents take on more responsibilities. Imagine autonomous coding agents working tirelessly in the background, handling workloads that slow teams down. These agents understand your entire codebase, no matter how large or complex. They execute autonomously: planning, writing, testing and deploying code independently. They work continuously, spotting and solving problems proactively, and they improve over time, learning from real-world use.

These are superintelligent coding agents, deployable instantly and in parallel.

It’s a fundamental shift in how software is built and maintained. It’s one massive leap towards a future where developers become directors of autonomous coding agents who build and maintain code on our behalf … and long term, a future where we all become directors of superintelligent agents that conduct knowledge work on our behalf.

The elite team building this future

In just one year, Reflection has assembled a world-class team—made up of the very researchers and engineers who pioneered some of AI’s greatest breakthroughs:

• Deep Q Networks, AlphaGo, AlphaZero and MuZero—landmark achievements in reinforcement learning.
• PaLM, Character AI, ChatGPT and Gemini—cutting-edge advances in large language models.

But what makes Reflection truly unique isn’t just its elite technical expertise—it’s the tight integration of research and product, combined with a customer-first approach to AI development.

They’re not just building frontier intelligence; they’re building the future of AI-driven software engineering with real-world impact.

Join us

Legendary companies are built on solving the most ambitious problems of our lifetime. As CEO Misha puts it:

“We’re working on the root node problem of our lifetime.”

Reflection AI is on a mission to achieve superintelligence, starting with autonomous coding agents that will redefine software development. We have deep conviction in their potential and are honored to have been their first and long-term partner in this journey.

On a personal note—working with Misha and Ioannis has been one of the greatest adventures of our lives.

We’re hiring world-class researchers, engineers and product builders in San Francisco, New York, London and Paris.

Come build the future with us!