Among interesting things from the current AI moment: it looks like many patterns from the world of coding are quietly diffusing into adjacent disciplines—sometimes deliberately, sometimes by osmosis.
OMCR is a good illustration—thank you Young-Eun Lee. Built on the shoulders of the OMX ecosystem ("oh-my-codex," "oh-my-claudecode," and others), it applies harness engineering—orchestrating specialized agents with structured handoffs, review loops, and traceable iteration—to scientific research. This may open fascinating perspectives for the world of R&D.
Scientific discovery has always been deeply iterative and collaborative by nature: hypotheses, experiments, peer review, refinement, interpretation. What AI-native harness engineering introduces is the possibility of making these loops more explicit, orchestrated, traceable and composable.
More broadly, what OMCR illustrates is a deeper pattern: the grammar of software development is being ported, almost intact, into the production of knowledge itself. Minimalistic UI, agile rituals, iteration, traceability, modularity, structured handoffs between specialized actors.
Over time, coding tools are quietly reshaping the broader product narrative—and beginning to define the gravitational center of entirely new platform strategies. A few examples: ChatGPT's forthcoming super app is said to be organized around Codex. Claude Code is driving the fastest-growing share of ARR at Anthropic, while absorbing increasingly diverse use cases outside code. Microsoft is rapidly reframing decades of productivity software around "Cowork" and agents—a striking repositioning effort that feels like a large-scale narrative re-skinning. Meanwhile, a growing number of SaaS platforms suddenly rediscover the virtues of being headless, scriptable, CLI-friendly.
That looks like a narrative inversion: not a chatbot that can also write code, but a coding agent becoming the core runtime of a broader productivity platform. In a way, the patterns of software development seem to be catching up with—and progressively reorganizing—many layers of office work itself.
There is something almost paradoxical in this: forty years of GUI logic were built on abstracting the command line away from end users, on the assumption that its minimalistic UI was too hostile for non-developers. What we may be witnessing now is the terminal intuition returning—not as regression, but as the architectural foundation of the next generation of products, this time available to everyone.
Marc Andreessen famously argued in 2011 that software was eating the world—and the prediction largely held. The more interesting observation, fifteen years later: the tools built to write software are now restructuring the digital product stack itself.
Perhaps software's deepest legacy is not just its products—but its discipline.
