With the emergence of AI coding agents, the role of engineering, product and design (EPD) organisations at software companies is changing.

Harrison Chase (해리슨 체이스), CEO of LangChain, a company behind an AI app development framework, recently shared his views on how coding agents are changing EPD organisations and processes and what may come next in a post on the company blog.

He said that before the Claude era, EPD workflows were simple. Someone came up with an idea and the product team wrote a PRD, or product requirements document. The design team made a mock-up and the engineering team implemented it in code. Because making software took a lot of time and effort, roles had to be divided, and the PRD was the starting point.

Coding agents have changed that flow. Throw out an idea and working software comes out. Bottlenecks have also shifted. Chase said, "Now anyone can write code. EPD roles move toward reviewing and judging it. As the cost of generating code falls, the number of prototypes has grown. That means there is more to review."

Even when using coding agents, text is not entirely unnecessary.

He said, "When someone brings an idea and a prototype, there is no way to know whether this code is what was intended or something added by mistake. Documents to convey intent are still needed. The process of PRD, then mock-up, then code has disappeared, but writing that contains product requirements remains essential material alongside the prototype. The PRD of the future may take the form of a structured prompt."

He said that in the era of AI coding agents, the value of generalists with a broad sense of product, engineering and design is rising within EPD organisations.

He said, "In the past, you needed someone else’s help at the implementation stage. Now you can talk to an agent. You can get more done on your own." He added, "Coding agents are not optional. Instead of writing a spec document and waiting, PMs can build prototypes themselves to test ideas. Designers can do iterative work in code rather than staying only in Figma. Engineers can spend time on system design rather than implementation. If you do not use coding agents, you will be replaced by someone who does."

Product sense is also an important factor. He said, "People decide what to tell the agent to build. If you point it in the wrong direction, it only increases the review burden. Regardless of role, product sense has become a basic requirement."

System-level thinking has also become more important. He said that as implementation costs fall, the ability to judge what is right from the start becomes a point of differentiation.

He said, "Engineers need a clear picture of service, API and database design. Product owners must see through what users say to what they actually need. Designers must be able to explain why something is easy to use."

Chase said he expects two roles to become more important within EPD organisations as AI spreads.

One is the builder. In Chase’s view, this is someone with product-centric thinking who can handle coding agents and has a basic sense of design. A builder can take a small feature from idea to deployment alone, within guidelines such as test environments and component libraries.

The other is the reviewer. Large and complex features require deep review, and Chase said reviewers need strong system-level thinking in their own domain and must be able to process a growing number of review requests quickly.

He said, "If you are an engineer, you need to choose either to become a reviewer skilled in system design or to build product and design sense and become a builder. If you work in product or design, you need to either sharpen the framework of thinking in your own domain to become a reviewer, or build coding skills and move toward being a builder."