What happens when coding stops being the hard part?
That was one of the questions at the center of a wide-ranging CHM Live conversation on September 23, 2026 when Boris Cherny, creator and head of Claude Code at Anthropic, joined CHM curator David C. Brock to discuss AI agents, the changing role of software engineers, and why he believes curiosity and judgment may matter more than coding expertise in the years ahead.
Along the way, he explained how a coding tool that began as an experiment became a multi-billion-dollar business, why building products with AI feels more like studying a living system than writing traditional software, and why AI safety remains the industry's most important challenge.
The event was made possible by the generous support of Mark and Mary Stevens.
Watch the full event with Boris Cherny to hear his thoughts on AI agents, Claude Code, software engineering, AI safety, and the future of human-computer collaboration.
When Cherny joined Anthropic in 2024, he believed the company's latest models were already capable of much more than existing products allowed people to see.
"The model was ready and no one on the product side was realizing it," he said.
That insight helped lead to Claude Code, which emerged from Anthropic's efforts to understand what increasingly capable models could do when they were allowed to act rather than simply respond.
For Cherny, the difference between a chatbot and an agent is simple: "[An agent] is an AI chatbot, but it's able to use tools." Instead of only generating text, they can read files, write code, search for information, and take action.
Reading files, writing files, running code, searching for information, spawning additional agents, these seemingly basic capabilities transformed AI from something that provided suggestions into something that could perform work.
One of the most memorable parts of the discussion was Cherny's description of what it feels like to build products on top of large language models.
Traditional software engineering, he explained, is built around deterministic systems. Models are different.
"It feels more empirical," he said. "It's more like a social science."
Rather than designing every behavior in advance, developers often discover capabilities through experimentation. They try ideas, watch how models behave, and learn what works. At one point, Cherny compared the process to "communicating with a weird alien creature."
That mindset has shaped the evolution of Claude Code itself, as Anthropic continually adds, removes, and redesigns tools to better match what increasingly capable models can do.
Asked what AI coding tools mean for software engineers, Cherny described a future that looks less like the end of programming and more like its expansion.
He compared the moment to the invention of the printing press. Reading and writing were once specialized skills practiced by a small professional class. Over time, literacy became widespread.
Cherny believes software creation may follow a similar path.
"If I did have to make a prediction," he said, "everyone will be able to write code as well as I can."
The more interesting question, in his view, is what happens after that. If building software becomes easier, what new things will people create? What becomes possible when the ability to turn ideas into working software is available to millions more people?
While much of the discussion focused on possibility, Cherny repeatedly returned to safety.
He described the layers of work Anthropic invests in to understand and manage the behavior of increasingly capable models, from alignment training and behavioral testing to mechanistic interpretability research and real-world monitoring.
"It's no accident that everyone that works closely to model research takes safety very seriously," he said.
For Cherny, safety is not separate from building useful AI systems. It is a necessary part of making sure increasingly powerful technologies ultimately benefit the people using them.
Toward the end of the evening, Brock posed a question that many students and software engineers are asking: if AI can write more and more code, what skills will matter most?
Cherny's answer had little to do with programming languages.
He pointed first to curiosity. The strongest engineers, he said, increasingly look beyond engineering itself, developing interests in design, business, research, and other fields.
He also emphasized judgment, communication, and the ability to operate in ambiguous environments.
In a world where AI can generate code on demand, understanding what should be built may become more valuable than knowing how to implement every detail yourself.
For students, his advice was simple: learn the tools, stay curious, and consider studying something beyond computer science.
The future, he suggested, belongs to people who can connect ideas across domains, not just write code.
