AI Product Builder? Is this realistic?

I recently came across a new type of role advertised in career postings, the one of an "AI Product Builder": a seasoned individual who works as a hybrid between a product manager and developer. The premise is founded on the advancement of code harnesses and the desire to reduce the idea -> ship cycle. By having a single person own the requirements, and the implementation, that person is able to iterate and experiment faster. Is this really possible in mid-2026 and can a conventional product manager evolve into such a role? As always the answer is that it depends. Certainly code harnesses have contributed a lot to making agents more focused and reliable. The recent advancements have been in the following areas: better prompt management (especially in the platform owners tools, e.g. Claude and Codex, that have the best knowledge on how to properly steer their models) context management. That's where a lot of the investment went. For a given task, what is the minimum but sufficient amount of information that needs to be in context. This includes: offsetting deterministic processes to "tools" (APIs, data search, use of MCP) factual and episodic memory (often managed via various markdown files). Skills roughly fall in this category too as they can be seen as "remembering how to do something" semantics (ontology, graph databases, ...) context compaction better workflow definitions (sub-agents, execution loops and graphs) assurance chain of thought validation review by agents ("judges") traditional quality gates (unit tests, code inspection tool, ...) human in the loop (pull requests) (there are other important advancements such as security but they are less important here) From there, we can infer what factors will increase the quality of the output: documentation of the existing code base (so that context injection is optimized; and also to minimize costs by having agents not re-discover the world every time; some of it can be bootstrapped by asking agents to document th