Background
In September 2019, Fibery founder Michael Dubakov published "No-code Revolution. Why Now?", betting that a fourth wave of productivity software (Airtable, Coda, Notion, Retool, Fibery) would wash away third-wave collaborative vendors within a decade. In August 2026, his follow-up essay "Malleable software = solid bases + custom code" grades that bet as "aged so-so": no-code tools did grow mainstream, but third-wave vendors like JIRA were not displaced. The real surprise came at the end of 2025, when LLMs made every no-code and low-code tool increasingly dependent on code — a return he did not predict.
Key points
- Five routes to building internal tools, each missing something:
- *Generate from scratch* (Claude Code, Codex, 2024): smooth for the first 80%, painful for the last 20% (hosting, auth, permissions, databases).
- *Vibe-coding* (Lovable, v0, 2023): faster out of the box, but stalls once needs exceed the generator's scope.
- *Low-code* (Retool, Softr, 2017): sells "base + code" for a decade, but provides an application base only — data lives elsewhere, with no collaboration, comments, or change history.
- *Malleable tools* (Notion, Fibery, 2013): rich out of the box, but extension points fall short of your specific workflows.
- *Vertical software* (e.g., Kinoko, MycoSense for mushroom farms): fits immediately, but locks you in; if it becomes flexible, it stops being vertical.
- Three types of bases (the essay's hardest section):
- *Technical base*: hosting, authentication, raw storage (what vibe-code platforms provide).
- *Application base*: prebuilt components, access control, automation (Retool, Softr) — but your data lives elsewhere.
- *Work base*: domain model, collaboration, history, and permissions grow around the data itself (Notion, Fibery).
- Custom code covers the remaining ~20%, but "it is your company — no vendor will model it exactly." An anecdote from the Filestash author: splitting his product into 80% core + 20% plugins resulted in plugin code 10x the core size, because everyone's 20% differs.
- Two conditions for viable custom code: (1) it inherits the base — permissions, history, and data integrity apply automatically; (2) it is bounded — custom code can break itself but must not corrupt the base, and rollback must be easy.
- The market inversion: for two decades vendors sold interfaces while bases (storage, permissions, history) were invisible plumbing. Now interfaces can be generated in minutes while bases still take years to build. Dubakov's new bet: solid bases + custom code wins, with malleable tools arriving first — extension points take quarters to add, bases take years.
- Practical principle: choose bases, not interfaces. Data, history, and permissions accumulate year over year and are hard to migrate; UI is becoming the cheapest, most replaceable layer. For individuals, vibe-code for fun; if vertical tools cover 90% of workflows, buy them; for teams with evolving processes, start from a malleable tool and vibe-code the missing 20% — Dubakov built the mushroom-farm management space in Fibery in roughly an hour.
Caveats
The author runs a malleable-tool company and his second bet is on himself — not a refutation, but worth keeping in mind. On Hacker News, skeptics asked why to trust a badly wrong forecaster; the top answer: nobody predicted AI. He promises a check-in by 2030 to see whether the mushroom farm has finally escaped spreadsheets.
Source: Dubakov, M. (2026-08-18). Malleable software = solid bases + custom code. https://www.mdubakov.me/malleable-software-solid-bases-custom-code/ · Prior essay: https://fibery.com/blog/gems/no-code-revolution-why-now/ · HN discussion: https://news.ycombinator.com/item?id=49508059