
What happened
A Zenn essay proposes "Co-generative Systems Design(共生成的システム設計)", arguing that once AI writes specs, designs, tests and operations, humans can no longer approve every change, so the procedures and rules around change become design objects.
Why it matters
It reframes the core design artifact from the finished program to the environment and procedures through which humans and AI keep changing the system, with the essay noting that approving every generated design by hand may hit limits in both speed and scale as model capability rises.
Summaries like this, in your inbox every morning.
The essay builds its argument by borrowing from architecture and urban design rather than treating software design as a purely technical exercise. It notes that software design already drew heavily from Christopher Alexander's pattern language, and that Eric Evans's domain-driven design shares the concern with growing a shared language between experts and designers. But it draws a distinction between a building and a city: a building is used and renovated after completion, while a city hosts many actors pursuing their own goals, so no single person can fix the final form. The essay suggests software may take on a similar character once AI moves beyond writing code into specs, design, testing, operations and proposing the next change.
The proposal then separates design authority into roles. A lead designer sets the scope delegated to AI, the change procedures and where responsibility sits. Developer citizen-designers extend and modify the shared mechanisms. User citizen-designers change their own environment, monitor AI judgments and file objections. AI is positioned as an agent that designs, generates and executes within delegated scope. The essay stresses that delegating design capability to AI should not narrow human participation.
From Ruby and Rails, the essay wants to carry over the freedom to modify existing mechanisms, citing the Rails Doctrine's "Provide sharp knives" principle of trusting developer judgment rather than removing powerful tools. It argues this freedom should extend to users who cannot write programs, while noting that giving human developers strong extension power differs from giving autonomous AI unlimited authority. It closes with open questions about who represents users, whether objection can keep pace with fast AI changes, and how to verify that a system has improved overall.
Pick your industry and the AI tools you use, and get news related to your work every day.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.