In 2022, in the bustling tech hub of San Francisco, two visionaries—Eric Steinberger and Sebastian De Ro—came together with a bold mission: to fundamentally transform the world of software development through artificial intelligence (AI). Their startup, Magic, is no ordinary AI company. While countless others focus on specific applications of AI, Magic has set its sights on an ambitious goal: automating software development and pushing toward the creation of artificial general intelligence (AGI). But how did two founders, each with distinct but complementary backgrounds, manage to captivate some of Silicon Valley’s top investors and secure a staggering $465 million in funding without yet launching a commercial product?
Eric Steinberger and Sebastian De Ro
Eric Steinberger’s journey toward AI innovation began at a young age. He was fascinated by the potential of artificial intelligence long before it became a tech industry buzzword. His passion for AI led him to pursue computer science at the prestigious Cambridge University, but he soon realized that academia was not the only path to achieving his goals. Steinberger dropped out and went on to work as an AI researcher at Meta, honing his skills in one of the world’s most advanced research environments. However, his ambitions extended beyond corporate AI, and in 2019, he co-founded ClimateScience.org, a volunteer-driven organization dedicated to tackling environmental issues.
Steinberger’s co-founder, Sebastian De Ro, followed a different path but one that ultimately converged with Steinberger’s. De Ro rose to the role of Chief Technology Officer (CTO) at the German business process management firm FireStart, gaining valuable experience in leadership and innovation. The two first connected through their mutual involvement with ClimateScience.org, but it wasn’t long before they realized they shared a much bigger vision: to create AI that could revolutionize the software industry.
Building the Future of AI-Powered Software Development
Magic is not just another AI startup; it’s aiming to be the AI coworker developers never knew they needed. By focusing on automating key aspects of software development—code generation, debugging, and even project planning—Magic’s AI tools function as an automated pair programmer. The idea is simple but powerful: developers working with Magic’s AI could write, review, and manage code faster, more accurately, and with fewer errors.
The crown jewel of Magic’s technology is its LTM (Long-Term Memory) series of models. These models are designed to handle vast amounts of contextual information, which is critical for tasks like code generation and debugging. Their latest innovation, the LTM-2-mini model, boasts an unprecedented 100 million-token context window, meaning it can process roughly 10 million lines of code or the equivalent of 750 novels in a single go. This makes it far more powerful than other commercial models, such as Google’s Gemini flagship, and allows it to synthesize code with a deep understanding of the entire codebase, documentation, and even libraries.
Nat is a great sparring partner, coach and supporter. He has consistently pushed us to be even more ambitious while remaining practical. We are incredibly fortunate to have him as our major backer and now also as a board member at Magic. https://t.co/Gw7jBmK4Z6
— Eric Steinberger (@EricSteinb) February 15, 2024
Imagine you’re a software engineer tasked with creating a custom password strength meter or developing a calculator using a new UI framework. With Magic’s AI, these tasks can now be completed almost autonomously, as the model continuously learns and adapts to the context, providing real-time support and suggestions.
Magic’s Rise in the AI Space: Massive Funding, Big Names, and Bold Goals
Despite having no revenue or commercially available products yet, Magic’s vision has captured the attention of some of the biggest names in tech and venture capital. In August 2024, the company closed a $320 million funding round, raising its total funding to $465 million. Among the investors are heavyweights like Eric Schmidt, former CEO of Google, Alphabet’s CapitalG, Sequoia, Atlassian, and notable individual investors such as Elad Gil, Nat Friedman, and Daniel Gross.
Nat is a great sparring partner, coach and supporter. He has consistently pushed us to be even more ambitious while remaining practical. We are incredibly fortunate to have him as our major backer and now also as a board member at Magic. https://t.co/Gw7jBmK4Z6
— Eric Steinberger (@EricSteinb) February 15, 2024
This influx of capital comes as no surprise when you consider the potential market. The demand for AI-powered coding assistants is set to explode, with the market for such tools projected to be worth $27.17 billion by 2032. And Magic’s valuation reflects this optimism. The company was reportedly seeking to raise over $200 million at a $1.5 billion valuation in July 2024, marking a steep rise from its $500 million valuation just five months earlier in February.
Investors aren’t just betting on Steinberger and De Ro—they’re betting on the potential for AI to fundamentally change software development. As Daniel Gross put it: “If a copilot generates $10 billion of revenue, how much is a colleague worth?”
Magic’s Models and Technology
While the staggering valuation and capital backing are impressive, it’s Magic’s technology that truly sets it apart. The company’s LTM-2-mini model is revolutionary in its ability to handle ultra-long context windows, but Magic isn’t stopping there. The company is already working on larger versions of their models, aiming to push the boundaries even further.
One of the key innovations is HashHop, a novel evaluation method developed by Magic to rigorously test their models. Unlike traditional benchmarks that rely on semantic hints, HashHop forces the model to store and retrieve vast amounts of information from its memory. This eliminates semantic shortcuts and ensures that the model is truly capable of handling complex, multi-step reasoning tasks—a crucial ability for any AI looking to automate software development.
The Road to AGI
While Magic’s current focus is on automating software development tasks, Steinberger and De Ro have their eyes on an even bigger prize: artificial general intelligence (AGI). The company’s approach to achieving AGI combines cutting-edge techniques like frontier-scale pre-training, domain-specific reinforcement learning, and ultra-long context processing. They also employ inference-time computation, which allows their models to adapt and learn in real-time, improving their performance continuously.
To help realize this vision, Magic has brought on Ben Chess, a former lead from OpenAI’s supercomputing team, to help scale their models and enhance their capabilities. The ultimate goal is to create AI systems that can not only automate complex tasks but also solve alignment issues, paving the way for safe and reliable AGI.
As Steinberger said in a recent interview, “We’re building towards AGI and safe AGI in particular… but we’re thinking a lot further than just code completion.”
Building the Infrastructure for AI’s Future
To support their ambitious goals, Magic has partnered with some of the biggest names in tech infrastructure. The company is building its next-generation supercomputers on Google Cloud, leveraging NVIDIA’s GB200 NVL72 systems. This setup is designed to enhance the training and inference efficiency of their models, ensuring that they can scale rapidly to meet the growing demands of developers and enterprises.
Nat is a great sparring partner, coach and supporter. He has consistently pushed us to be even more ambitious while remaining practical. We are incredibly fortunate to have him as our major backer and now also as a board member at Magic. https://t.co/Gw7jBmK4Z6
— Eric Steinberger (@EricSteinb) February 15, 2024
What’s Next for Magic?
As of now, Magic has yet to release a commercial product. But with their massive funding, groundbreaking technology, and relentless focus on AGI, the company is poised to become a major player in the AI space. Investors and industry insiders alike are eagerly watching to see how Magic’s AI tools will impact the future of software development—and whether their long-term bet on AGI will pay off.
For now, Magic’s story is one of boundless ambition, driven by a team that believes AI can not only assist but ultimately outthink and outwork human developers. Whether they succeed in creating the “AI coworker” of the future remains to be seen, but one thing is certain: Magic is a startup to watch closely as it continues to push the limits of what AI can achieve.