Mira Murati became one of the most recognizable technical leaders in artificial intelligence while serving as OpenAI’s chief technology officer during the launches of ChatGPT, DALL-E, and GPT-4. In September 2024, she left one of the most influential companies in tech without announcing her next move.

Five months later, she introduced Thinking Machines Lab. The company entered an already crowded AI market with an unusually ambitious premise: the next useful model may not be the one that knows the most, but the one people and businesses can adapt to the way they actually work.

From Product Leadership to the CEO Seat

Murati was born in Albania and studied mechanical engineering at Dartmouth College. Before joining OpenAI in 2018, she worked on the Tesla Model X and led product and engineering at Leap Motion, a company that developed motion-tracking technology.

At OpenAI, she moved from applied AI and partnerships into product and research leadership, becoming CTO in 2022. Her role placed her at the intersection of research and products used by millions of people—a position that made her one of the company’s most visible public voices.

She also served as interim CEO for three days during OpenAI’s leadership crisis in November 2023, after the board removed Sam Altman and before he returned. Murati remained CTO until September 2024, when she said she was leaving to create “the time and space to do my own exploration.”

Building Thinking Machines Lab

Murati launched Thinking Machines Lab in February 2025 with researchers and engineers from companies including OpenAI, Meta, and Mistral. The founding group included prominent OpenAI alumni such as John Schulman, Barrett Zoph, and Lilian Weng.

Investors treated the team as a serious competitor before it had released a product. The company raised a reported $2 billion seed round at a $12 billion valuation, one of the largest seed rounds in technology history. A later funding effort at a much higher valuation was widely reported but did not close as initially expected.

That scale creates attention, but also pressure. A famous founding team and a large balance sheet can buy computing power and time; neither guarantees that a new model will solve a problem customers are willing to pay for.

The Bet on AI You Can Adapt

Thinking Machines Lab’s first public product was Tinker, a platform that helps developers fine-tune open models without managing a large distributed training system themselves. It addresses a practical obstacle for teams that want specialized AI: customization usually requires infrastructure and machine-learning expertise that most businesses do not have.

In July 2026, the company released Inkling, its first open-weight model. Unlike a finished chatbot, Inkling is meant to be downloaded, modified, and adapted for specific applications. That makes it more of a foundation for developers than a product a small business would deploy on its own.

The distinction matters. General-purpose AI is useful for broad tasks, but a customer-support agent, scheduling assistant, or internal operations tool needs to understand the rules and context of one business. The commercial opportunity is moving from access to a powerful model toward making that model reliable inside a real workflow.

Why Interaction May Matter More Than Another Benchmark

The lab’s longer-term research focuses on what it calls interaction models. Most conversational AI waits for a person to finish speaking, processes the full turn, and then responds. Thinking Machines is experimenting with systems that continuously process audio, video, and text in 200-millisecond chunks.

That approach could make an AI system better at the behaviors people expect in natural collaboration: recognizing an interruption, responding to a visual cue, speaking at the right moment, or adjusting while a task is underway. The goal is not simply a faster answer. It is a model that can participate in an interaction instead of treating every exchange as a separate prompt.

For a business, this could eventually change customer service, training, sales conversations, and accessibility. It could also create new risks. A model that acts in real time needs clear permissions, reliable business data, and a safe way to hand control back to a person when context is missing.

What Business Leaders Should Watch

Thinking Machines Lab is still proving that its research can become durable products. Business owners do not need to choose an Inkling model—or rebuild their systems around interaction models—to learn from its direction.

Watch three shifts:

  • Customization over novelty: The value of AI will increasingly depend on how well it fits your processes, not how impressive a generic demo looks.
  • Interaction over single prompts: AI interfaces are moving toward continuous conversations and actions, which raises the standard for context, latency, and safeguards.
  • Infrastructure behind the experience: A smooth assistant still depends on accurate data, clear workflows, integrations, and human escalation.

Murati’s new company is a reminder that the AI market is not settled. The next phase may be less about asking which model wins and more about deciding which systems can be shaped around the work people actually need to do.

If your team is exploring that question now, our AI Solutions practice helps businesses map useful workflows before choosing tools. You can also schedule a short conversation to discuss where AI could remove friction without adding another disconnected system.

This is the ninth installment of the Women Who Built Technology series. Start with the complete series or read the previous entry on Gladys West.