On Building Malleable Software In the Age of AI | Notion After Hours

Malleable software is about letting people reshape their digital tools as they work—so the environment fits the way they think, not the other way around. Across the conversation, the central tension is clear: modern software often locks users into rigid interfaces designed by distant teams, while the physical world naturally supports adjustment. The payoff for getting this right is practical and cultural—people can express themselves, iterate faster, and share work without friction, instead of fighting the constraints of the platform.

Early memories of “hacking” personal computing—changing Mac menu items via Resedit’s code and resource forks, using Windows utilities like Resource Hacker, or building custom game experiences—set the tone for why malleability matters. Several participants point to small, everyday tools as proof that flexibility doesn’t have to be flashy: screenshotting apps like Clean Shot, drawing tools like TL Draw, and lightweight workflows that turn sketches into something actionable. Others connect malleability to mass adoption, comparing Instagram Stories’ editor to a kind of mainstream “malleable” creation surface—where sharing is built in and the loop from making to distributing is immediate.

The discussion also highlights what gets lost as software abstractions multiply. As standards rise and products chase higher production value, it becomes easier to “vibe code” a polished output without building a real mental model of how it was produced. That creates an exit problem: once the glossy first step is done, users can’t reliably understand or control what happened. A related frustration is interoperability: common protocols can reduce everything to a lowest-common-denominator baseline. One proposed way forward is layered compatibility—start with something broadly readable (like plain text), then optionally add richer annotations that only some tools can interpret, preserving both sharing and expressive power.

A recurring theme is the need for a “ladder of fidelity,” where users can move smoothly between simple representations (to-do lists, basic notes) and more structured, powerful data models without hitting a hard ceiling. AI is framed as a bridge across that gap: it can take messy, unstructured input and automatically structure it into forms that computers can work with—reducing the need for users to pre-plan their filing system. That aligns with a broader philosophy of tool design: successful platforms often begin with familiar primitives (spreadsheets, for example) rather than trying to solve everything at once.

The conversation ends by tying malleability to responsibility and learning. Restrictive safeguards can be justified when millions of users accidentally delete irreplaceable data, but over-restriction risks a self-fulfilling cycle where people never learn how to use tools well. Participants argue that users should be treated as capable learners—because when tools assume incompetence, the software ecosystem can drift toward safer but less empowering experiences. Notion’s block-based, hierarchical editing model is presented as a concrete example of a more expressive structure, one that can ramp from documents and creativity into deeper automation when users are ready. AI then becomes less about replacing authorship and more about meeting people where their information already lives—turning personal, cared-about data into something easier to query, transform, and build upon.