In an era when AI agents write code directly, what capabilities do developers really need?

Phil Chen (필 첸), a founder who previously worked at OpenAI, posted an answer to that question on social media platform X (Twitter).

Chen has worked at Scale AI, Google DeepMind and OpenAI, among others. He shared his views based on experience across settings ranging from a 15-person startup over 6 years to companies with more than 100,000 employees, and from personally thinking through hiring and careers.

The first things he cited were time, relationships and reputation.

"Access to capital has never been higher. But trust and reputation built with good people are still scarce," he said. "Rather than chasing opportunities to make money quickly, you should focus on meaningful problems, do good work, and make sure people walking in the same direction know that."

The second is the ability to find problems.

"In interviews, it makes no sense to ask LeetCode problems in an era when people no longer write code by hand," he said. "What hiring interviews look for is the ability to adapt quickly to an unfamiliar environment, pick problems worth solving, and execute. Agents handle well-defined complex problems. The most important person is the one who judges which problems matter and allocates resources."

The third is choosing the most ambitious form of problem.

"With AI, anyone can now build simple systems. Real value arises only when you focus extremely on truly difficult problems," he said. "The same goes for choosing a company. You need to see whether it is solving problems in the most ambitious way, and whether it is a team that can actually solve them."

The fourth is the ability to execute and complete the last mile. "Most outputs produced by AI agents are sloppy," he said. "Value is created by the person who goes one step further. A habit of spending a bit more time on completeness, clean structure, scalability and creativity makes a decisive difference."

The fifth is that you need to be in a position to see good opportunities.

Chen said he turned down job offers from Anthropic and Cursor in 2023 and chose DeepMind, then moved to OpenAI in 2024. "The opportunity to join Cursor came thanks to my reputation among acquaintances, and Anthropic came because I spent time on problems that team was interested in," he said. "Standing in a position to see opportunities comes first, and then actually choosing and executing comes next."

He also added advice on entering the AI research field.

"Computing resources can be obtained from cloud providers in the form of credits. You can start by using models yourself and evaluating them," he said. "A researcher is not a job but an attitude. You can work like a researcher even if you are not inside a frontier lab."