Simon Willison quotes Jeremy Morrell on extensible software

Key facts

• On 19 August 2026, Simon Willison published a “Quoting” post on simonwillison.net excerpting a passage from Jeremy Morrell’s essay “Extensible Software in the age of LLMs.”
• Morrell’s hypothesis: large language models radically lower the cost of authoring software extensions, while modern sandbox primitives lower the cost of deploying them and provide what the passage calls “good security boundaries.”
• Morrell’s proposed pattern: build the app as a “solid, accountable core,” then let users safely extend it in many directions with an LLM filling in the code for each extension.
• Morrell frames the outcome of this pattern as giving users “super powers.”

Why it matters

Morrell frames this as a hypothesis rather than a settled result: that lower authoring costs from LLMs, combined with lower deployment risk from what the passage calls modern sandbox primitives, open a new opportunity for extensible software on the web. The architecture proposed is to keep a single “solid, accountable core” that a team builds and controls, while letting an LLM generate the extension code needed for each user to customize the app in their own direction, with the sandbox supplying what the passage calls “good security boundaries” around that generated code. If the cost reductions described hold up, the practical effect is to shift extensibility from a feature a team designs and maintains by hand toward something individual users can effectively generate for themselves.

Who it affects

As described, the idea is aimed at teams building web software who want users to be able to customize it without the team hand building every extension itself, and at the users of that software, whom Morrell says the pattern would give “super powers” to extend the app in their own directions.

How to use it

Morrell’s own passage doubles as the recipe proposed: build the application as a “solid, accountable core” that the team fully controls, wrap the user extensible parts in a sandbox that supplies strong security boundaries, then have an LLM generate the specific code for each user’s extension rather than hand building every possible extension in advance.

How solid is it

Morrell introduces the idea explicitly as “my hypothesis.” The passage Willison quotes states the claim in general terms, including that sandboxes provide “good security boundaries,” without walking through a specific extension, product or test that would demonstrate it holding up in practice. It reads as a proposed direction for software architecture, not a report of something already built.

Risks and caveats

The passage compresses two separate claims, that LLMs make extensions cheap to write and that sandboxes make them safe to run, into one sentence each. “Good security boundaries” is Morrell’s own phrase for the safety claim; how strong a boundary needs to be depends heavily on what the extension can access, and that judgment is left to whoever adopts the pattern rather than settled by the passage itself. This is one paragraph excerpted from a longer essay, so it reads as the headline claim of Morrell’s argument rather than the whole of it.

“My hypothesis is that there is a new opportunity for Extensible Software on the web. LLMs radically lower the cost of authoring extensions, and modern sandbox primitives lower the deployment cost and provide good security boundaries. We can build our app as a solid, accountable core, and allow users to safely extend it in many directions by having LLMs fill in the missing pieces. We can give our users super powers.”

— Jeremy Morrell, “Extensible Software in the age of LLMs”