Google DeepMind Veterans Raise $130M to Build AI That Codes Without Humans

Reflection AI is the latest startup aiming to redefine AI-powered coding, but its roots trace back to some of the biggest breakthroughs in artificial intelligence. Co-founders Misha Laskin and Ioannis Antonoglou bring deep expertise from their time at Google DeepMind, where they played pivotal roles in advancing reinforcement learning. Antonoglou, a founding engineer at DeepMind, was instrumental in creating AlphaGo, the first AI to defeat human champions in Go. Laskin contributed to reinforcement learning models that influenced Alphabet’s AI developments, including Gemini.

Today I’m launching @reflection_ai with my friend and co-founder @real_ioannis.

Our team pioneered major advances in RL and LLMs, including AlphaGo and Gemini.

At Reflection, we're building superintelligent autonomous systems. Starting with autonomous coding. pic.twitter.com/Q0SK4NdYU0

— Misha Laskin (@MishaLaskin) March 7, 2025

Now, they are channeling their knowledge into Reflection AI, a company focused on autonomous coding agents where AI systems that don’t just assist developers but operate independently. 

Today I’m launching @reflection_ai with my friend and co-founder @real_ioannis.

Our team pioneered major advances in RL and LLMs, including AlphaGo and Gemini.

At Reflection, we're building superintelligent autonomous systems. Starting with autonomous coding. pic.twitter.com/Q0SK4NdYU0

— Misha Laskin (@MishaLaskin) March 7, 2025

While most AI-powered coding tools function like cruise control, offering suggestions to engineers, Reflection AI aims to be more like a fully autonomous vehicle, managing and executing complex coding tasks without human intervention.

The company remained in stealth mode until now, emerging with a massive funding announcement: $130 million secured across two rounds. A $25 million seed round was led by Sequoia Capital and CRV, followed by a $105 million Series A led by Lightspeed Venture Partners and CRV. Other backers include LinkedIn co-founder Reid Hoffman, Scale AI CEO Alexandr Wang, SV Angel, and Nvidia’s venture capital arm. The latest funding round puts Reflection AI’s valuation at $555 million.

Reflection AI is already attracting paying customers, particularly in industries with large engineering teams, such as financial services and technology. The company’s initial focus is automating repetitive coding tasks like database migrations and code refactoring, freeing engineers to work on more strategic projects. Lightspeed partner Raviraj Jain, who is joining the board, emphasized that Reflection AI is not about replacing developers but enabling them to take on more complex challenges.

Misha Laskin laid out three core principles driving Reflection AI: a team with deep experience in reinforcement learning and large language models, a sharp focus on autonomous coding, and an equal commitment to research and product development. He stressed that superintelligence cannot be developed in isolation, signaling that Reflection AI’s ambitions extend beyond just improving AI-powered coding.

The AI coding space is heating up, with Reflection AI entering a competitive market that includes well-funded players such as Anysphere Inc., Replit Inc., and Poolside Inc. The field has gained momentum following OpenAI’s launch of ChatGPT in 2022, fueling interest in AI-driven automation for software development.

Reflection AI’s vision builds on past AI milestones. Between 2013 and 2020, reinforcement learning paved the way for breakthroughs like Deep Q Networks, AlphaGo, AlphaZero, and MuZero. From 2020 onward, large language models such as PaLM, ChatGPT, GPT-4, and Gemini pushed AI capabilities further. Reflection AI seeks to merge these advancements by scaling the autonomous capabilities of language models through reinforcement learning.

Sequoia partner Stephanie Zhan, one of Reflection AI’s investors, expressed confidence that the company could compete with the world’s leading AI labs. She noted that fully autonomous AI systems could soon handle a wide range of tasks currently performed by humans, hinting at the broader impact Reflection AI aims to achieve. With operations in New York, San Francisco, and London, Reflection AI is actively expanding its team.

As stated by the team, the goal is to build superintelligent autonomous systems, with autonomous coding seen as the foundational challenge of the “root node problem” that, once solved, could enable superintelligence across all computer-based tasks. The research agenda has been framed around a central question: How can a language model achieve the same level of autonomy on a computer as AlphaGo did in Go?