20VC: Codex vs Claude Code vs Cursor: Who Wins, Who Loses | Will All Coding Be Automated - Do We Need PMs | The Real Bottleneck to AGI | The Three Phases of Agents and What You Need to Know with Alex Embiricos, Head of Codex at OpenAI

OpenAI's Alex Embiricos believes his company has already won the AI coding wars, and his reasoning reveals a strategy that most competitors fundamentally misunderstand. While the market obsesses over feature parity between Codex, Claude Code, and Cursor, Embiricos argues that OpenAI's real competitive moat isn't what these tools can do today , it's exclusive access to tomorrow's models.

The Head of Codex at OpenAI makes a bold claim about sustainable competitive advantage in AI tooling. "We have the massive, like, capability advantage of training our own models to be good in our harness and building our harness to be good at the new models, and, like, no one else has early access to those," Embiricos explains. This isn't just about having better models. It's about the flywheel effect of co-evolving the model and the interface, something third-party developers building on external APIs simply cannot replicate.

This dynamic reshapes how we should think about competition in AI tools entirely. When Anthropic ships Claude Code or when Cursor integrates multiple model providers, they're competing with today's OpenAI. But Embiricos and his team are building for the models that don't exist yet. Every interface decision, every UX pattern, every performance optimization gets baked into the training process for future model generations. The result is a widening capability gap disguised as mere product iteration.

Embiricos reveals something even more counterintuitive about OpenAI's competitive strategy: they actually benefit when rivals improve. "Because we're playing such a long game for us, if the competition gets better, we learn. It's actually helpful for us," he says. This isn't corporate doublespeak. When you control the underlying model architecture and training process, competitive pressure becomes free R&D. Every breakthrough in AI coding interfaces can be absorbed and amplified through OpenAI's unique position in the stack.

The numbers back up this confidence. Codex usage has grown 20x since August, driven by a strategic pivot that most observers missed. The platform initially launched as cloud-based AI agents , autonomous systems working in parallel on coding tasks. That vision flopped. Instead, growth exploded when OpenAI focused on interactive coding, the more mundane but immediately useful real-time assistance that developers actually wanted. This pivot illustrates a crucial principle: in AI tools, distribution beats demonstration every time.

But the broader automation question looms larger than any single coding tool. Embiricos offers a framework for which SaaS companies survive the coming AI wave: "Does this SaaS company own a relationship with a human on the other end of things? And if it does, then I suspect it's not going away. Or does the SaaS company own some really important system of record?" These two factors , human relationships and system ownership , matter more than the sophistication of the underlying workflows.

This framework explains why pure-play automation tools remain vulnerable while platforms maintaining human touchpoints continue growing. The insight cuts against the common assumption that better AI automatically displaces existing software. Instead, Embiricos suggests that AI amplifies the importance of human relationships and data ownership rather than eliminating them.

The performance game matters more than most realize. OpenAI recently shipped models that run 40% faster via API and 25% faster in Codex itself. Speed isn't just user experience , it's the foundation of entirely new interaction patterns. When AI responses approach real-time, developers start using these tools differently, more frequently, for smaller tasks. This usage pattern change compounds into behavioral moats that slower competitors cannot easily replicate.

For startup founders and product leaders, the implications are clear. First, if you're building on top of third-party AI APIs, you're playing a fundamentally different game than companies with model access. Sustainable differentiation requires owning some part of the stack that cannot be easily replicated. Second, human relationships and system-of-record businesses are becoming more valuable, not less, as AI automation increases. Third, in AI tooling, performance improvements unlock new usage patterns that create behavioral moats beyond feature differentiation.

The coding automation race is really three races running simultaneously: the model capability race, the interface design race, and the performance optimization race. OpenAI believes they're winning all three because they control the foundational layer. Whether that advantage holds depends not on today's feature comparison charts, but on how effectively they can maintain their unique position as both model provider and application builder in an increasingly competitive landscape.