Product development used to be a clean relay: the PM writes requirements, the designer draws the interface, and the engineer turns the spec into a product. Generative AI is rewriting how that relay collaborates. Today, PMs can build clickable flows, designers can test interactions with near-real data, and engineers can compare several prototypes before formal development even begins.
A recent, widely discussed observation comes from Boris Cherny, the lead of Anthropic's Claude Code. His key point: what AI really changes is not the job titles themselves, but the "write a document, then hand it off for the next function to implement" workflow. When more people can produce working results directly, the bottleneck shifts from "who can build it" to "what is worth building, how to ship it safely, and who cleans up and maintains it over the long run."
That shift closely mirrors what we have long experienced while helping enterprises with digital transformation. Below we borrow this framework to discuss what it means in practice for small and mid-sized businesses and real-world system delivery.
## What AI changes first isn't titles, it's the handoff artifact
Figma CPO Yuhki Yamashita shares a similar view: once AI lets more people "make products," speed can create a false sense of progress. A team's real edge is no longer just shipping faster, but knowing what is worth shipping.
We feel this deeply. In many projects, the screen looks reasonable and the demo runs beautifully, but once it connects to an enterprise's existing processes and data, a critical step simply doesn't work. Validating a flow early with near-real data is far more valuable than producing a few more good-looking prototypes.
## Five work archetypes: which stage does each advance?
Cherny's five archetypes don't map to specific departments; they describe which state a person is best at moving a product from and to
- Prototyper: proposes and tests many ideas; most experiments never ship. The job is to narrow the search space fast.
- Builder: turns concepts into real products or infrastructure, adding tests, permissions, performance, and deployment.
- Sweeper: simplifies interfaces and code and removes features no longer needed, controlling product and system complexity.
- Grower: continuously tunes existing products to improve product-market fit, retention, and use cases.
- Maintainer: ensures the security, reliability, speed, and cost efficiency of mature systems so products stay stable at scale.
In practice, one person often spans two or three archetypes: an engineer can be both Builder and Maintainer, and a designer can be both Prototyper and Sweeper. The real value of this framework is that it answers a question titles never did: "Which state is this person best at moving a product from, and to?"
## As building gets easy, cleanup and maintenance grow scarce
Anthropic's research also warns: engineers now produce far more with AI, but if a team simply stuffs all of that output into the product, the outcome is usually disastrous — excessive technical debt, unmaintainable architecture, and a pile of useless features nobody cares about.
This is the point we most want to stress. Coding can increasingly be delegated to AI in large volumes, but direction, architectural cleanup, and long-term governance remain the last stronghold of human expertise. In other words, in the age of AI, the value of the "Sweeper" and the "Maintainer" will surpass, not fall below, that of the "Builder."
A reminder for business owners adopting AI and new systems: check whether your team's KPIs are punishing the people willing to spend time deleting redundant features and tidying architecture. If so, you may be accelerating an invisible pile of debt.
## For managers, the point isn't renaming roles, it's finding the gaps
Companies don't need to rush to abolish engineering, product, or design departments; these titles still carry professional standards, decision accountability, security permissions, and career systems. The more practical move is to ask three questions during hiring, staffing, and project reviews
- Does the team only reward adding features?
- Who has the authority to stop the wrong direction?
- Who is clearly responsible for cleanup and long-term maintenance?
AI lowers the barrier to producing the first version, not the full difficulty of building a good product. This is exactly how we position ourselves: as an enterprise's system integration and digital transformation partner, we care not only about "getting the system built," but about delivering security, maintainability, and long-term operability together, in line with governance standards such as ISO 27001. Because what truly determines how far a system can go is rarely the day it launches, but every day after.