OpenAI is facing an unprecedented engineering challenge: AI agents are shipping code so rapidly that traditional performance monitoring can't keep up. As ChatGPT scales to 900 million weekly active users, the company's infrastructure teams are deploying always-on AI agents to automatically detect performance problems and optimize systems in real time, according to Martin Spier, who leads ChatGPT's performance engineering team.

The core problem is simple but urgent. Two major shifts are happening simultaneously at OpenAI and across the AI industry. First, user bases are growing exponentially faster than before; companies are reaching millions of users in days rather than months. Second, agentic development workflows, where AI agents write and deploy code with minimal human oversight, are dramatically increasing the volume of code changes shipping to production.

Why Is ChatGPT's Scale Creating Such Pressure on Performance?

ChatGPT's growth trajectory illustrates the scale of the challenge. The service reached one million users in just five days after launching as a research preview in late 2022. That explosive adoption forced OpenAI to rapidly transform ChatGPT from an experimental tool into a production system capable of supporting hundreds of millions of concurrent users. By February 2026, ChatGPT had grown to 900 million weekly active users, representing nearly 11 percent of the global human population.

This growth wasn't smooth or predictable. Major feature launches created viral spikes in demand. When OpenAI released image generation capabilities, over 130 million users generated more than 700 million images in just seven days, creating sudden, massive surges in compute demand that stressed infrastructure teams scrambling to maintain service reliability.

Beyond raw user growth, the nature of how ChatGPT is being developed has fundamentally changed. Agentic coding tools are automating the process of writing, testing, and deploying code. This means more logic is reaching production systems without a human engineer reviewing every detail of what's being shipped. As Spier explained, this creates a critical gap in visibility.

"Previously, we always had this assumption that there was a human that understood all the changes that were going out. It's like someone actually coded that and architected it and know what's going on before they decide to push something out. That is not entirely true anymore," stated Martin Spier, who leads ChatGPT Performance at OpenAI.

Martin Spier, ChatGPT Performance Lead at OpenAI

How Is OpenAI Automating Performance Optimization?

To handle this acceleration, OpenAI is deploying automated systems that continuously monitor and optimize ChatGPT's performance without waiting for human intervention. These always-on AI agents handle three critical functions:

• Automated Profiling: Continuously analyze how code is performing across millions of requests in real time, identifying bottlenecks and inefficiencies without manual investigation.
• Regression Detection: Automatically catch performance degradation when new code is deployed, alerting teams to problems before users experience slowdowns.
• Continuous Optimization: Proactively optimize systems and infrastructure to maintain speed and reliability as code volume and user load increase.

This approach addresses a fundamental shift in how modern AI infrastructure operates. Traditional performance engineering relied on humans reviewing metrics and making optimization decisions. At ChatGPT's scale, with code shipping faster than ever, that model breaks down. Automated agents can detect and respond to performance issues in milliseconds, far faster than any human team.

The challenge extends beyond just performance metrics. Spier noted that the pressure affects every layer of infrastructure, from GPU allocation across global data centers to network connectivity between regions, data replication strategies, and the architectural decisions about where computing resources should be located.

What Does This Mean for the Future of AI Development?

OpenAI's approach signals a broader shift in how AI companies will need to operate as models become more powerful and development accelerates. The traditional separation between development, testing, and production is collapsing. Code is shipping faster, systems are scaling larger, and the humans overseeing the process have less visibility into every change.

This creates both opportunity and risk. Faster development cycles mean new capabilities reach users more quickly. But the reduced human oversight means automated systems must be extraordinarily reliable at catching problems. OpenAI's investment in always-on AI agents for performance monitoring suggests that the company views automation as the only viable path forward as agentic development becomes the norm.

For other companies building large-scale AI systems, the lesson is clear: traditional performance engineering practices won't scale. As development velocity increases and human oversight becomes impractical, infrastructure teams will need to adopt similar automated approaches to maintain reliability and speed. The future of AI infrastructure isn't just about building bigger systems; it's about building systems smart enough to optimize themselves.