Extensible software arrives, Nordics must own the hooks

LLMs let users write their own features inside your app, but the hooks and sandboxes are still controlled by foreign platforms; the Nordics must decide tonight whether to build our own extensible stack or keep feeding theirs.

EXTENSIBLE SOFTWARE IS HERE Jeremy Morrell at Cloudflare says the web is ready for apps that let users generate the features the product team will never build. A stable core, controlled hooks, and a sandbox that runs the code the LLM writes. Pi, the open-source terminal harness, already does this: users ask for a TypeScript extension, reload it in place, and share it. The pattern is clear. LLMs lower the cost of writing extensions, modern sandboxes lower the cost of running them. The hard work moves to defining the hooks and the limits. Whoever owns the hooks owns the intent path. NORDIC BUILDERS FACE A CHOICE Today, every Nordic builder who wants extensibility must choose between two bad options. Option one: build your own hooks and sandboxes, which means reinventing the wheel and competing with platforms that already have scale. Option two: use foreign hooks and sandboxes, which means your users’ intent data flows to foreign platforms, and your app becomes a feature of theirs. The Nordics have the talent and the capital to build our own extensible stack. We have the pension funds, the data centers, the open-source culture. What we lack is a decision. THE RISKS ARE REAL Morrell’s essay spends more time on security than on celebration. Generated code can leak secrets, break isolation, or run forever. A June 2026 study found recurring vulnerabilities in LLM-generated apps: exposed credentials, unfiltered input, placeholder logic. Better models and prompts reduce the problems, but the risks remain. The tradeoff is real. Every hook expands the attack surface. Every sandbox adds complexity. The core must still be accountable. LLMs make customization cheaper, but they do not make the host less responsible. CONCEPTUAL INTEGRITY IS THE OTHER SIDE Simon Willison warns that coding agents make it too easy to add rooms to the house. The cost of adding a feature drops, so the house grows in funny directions. The conceptual integrity falls apart. The app becomes a Winchester Mystery House: 140 rooms, no plan, no coherence. The solution is not to stop building. The solution is to own the hooks. If the hooks are yours, you can enforce the plan. You can say no to the 141st room. You can keep the house whole. THE POLL IS THE STEERING WHEEL Tonight, the network must decide. Do we build our own extensible stack, or do we keep feeding foreign platforms? The options are clear. Option a: Build a Nordic extensible stack now. Define the hooks, build the sandboxes, run them on Nordic compute. Use pension funds as anchor investors. Make it open source, make it sovereign. Option b: Build a Nordic extensible stack, but run it on foreign compute for now. Keep the hooks and the sandboxes ours, but accept foreign infrastructure until we have our own. Option c: Do not build a Nordic stack. Let each builder decide, let the market decide. Accept that Nordic intent data will flow to foreign platforms. Option d: No preference. Let the network decide later. The choice is yours. The hooks are the intent path. Whoever owns them owns the future.

researched · 4 sources

20 Augreaches everyone