Moos is the malleable operating system that grows with the user

The sequence is forward-only. The two filled marks are gates: the run stops there until the host has seen the proof itself.

1. 1962

   Augmenting Human Intellect: A Conceptual Framework

   Douglas C. Engelbart — Stanford Research Institute, Summary Report AFOSR-3223, October 1962

   The thing to improve is not the computer, it is the whole system a person works inside: the human together with artifacts, language, methodology and training. Augmentation raises what a person can do; automation takes the person out of the loop, and Engelbart wanted the first. The strategy is bootstrapping — better tools for handling symbols make better thinking possible, which makes better tools possible, and the cycle compounds.

   Moos bootstraps in exactly this sense: it uses its own build loop to build its own parts. Where it is thinner than Engelbart is the other three. It improves your artifacts and does nothing for your language, your methodology or your training, and it changes the artifact for you rather than teaching you to change it. Whether that is augmentation or automation in Engelbart’s sense is a fair question to put to it, and the honest answer is that it depends on whether you read the brief before pressing yes.

2. 1984

   Computer Software

   Alan Kay — Scientific American, September 1984

   Software is soft, Kay wrote, precisely because it has none of the physical constraints that make hardware expensive to change. Nothing about a program has to stay the way it is. His answer to the complexity that follows was architecture: systems built out of objects that behave like independent computers, and a user illusion that keeps the whole thing comprehensible from the outside.

   Moos takes the softness literally — the running system rewrites its own parts. What differs is what stands in for the missing physical constraint. Kay’s answer was architecture. The answer here is evidence: a change gets in only after the host has watched the tests and the scan exit.

3. 1997

   The Educated Mind: How Cognitive Tools Shape Our Understanding

   Kieran Egan — University of Chicago Press, 1997

   Education pulls in three directions at once — Plato’s transmission of a true picture of reality, Rousseau’s unfolding of the individual, and plain socialisation — and Egan argues the three are incompatible, which is why the whole enterprise keeps failing. His alternative describes understanding by the cognitive tools that carry it: somatic, mythic, romantic, philosophic and ironic, each acquired in turn, each changing what a mind is able to think.

   Moos says in its own headline that it grows with the user, and Egan is the reason that claim is worth both taking seriously and doubting. An operating system is a cognitive tool, and one that reshapes itself around a person could develop with them instead of holding them at the level it shipped with. But a system that builds the thing for you can stand in for understanding just as easily as it can build it. Moos does not teach, and nothing in it makes you better at the work. Whether the tool grows the user or only itself is not settled by the design; it is settled by use.

4. 2015

   Webstrates

   Clemens N. Klokmose, James R. Eagan, Siemen Baader, Wendy Mackay and Michel Beaudouin-Lafon — UIST 2015, Aarhus University and Télécom Paris

   A webstrate is a web page whose DOM is the document. The HTML, the CSS and the JavaScript that runs it all live inside one persistent, shared structure, and every change is synchronised to every other client as it happens. The line between a document you edit and an application you use stops existing.

   Moos wants the same collapse and stands on the same substrate — its apps are plain web UIs. The difference is where the malleability comes from. Webstrates gets it from immediacy: you change the running thing directly and it is changed. Moos gets it from a build run — minutes rather than keystrokes, and the change has to pass a gate first. That is slower and far less direct, and what it buys is a brief you can read beforehand and an undo afterwards. Webstrates was also built for several people at once; Moos runs for one.

5. 2019

   Designing and Programming Malleable Software

   Philip Tchernavskij — PhD thesis, Université Paris-Saclay

   The application is the thing in the way. Functionality gets trapped inside app boundaries, so anything you want to combine across two programs is out of reach. The thesis argues for interfaces built from objects and interaction instruments that recombine freely, contributes entanglers as a composable way to specify interaction, and demonstrates them in a prototype called Tangler.

   Moos has no answer here, and this is the sharpest criticism of it. You can have a new app in minutes and change an existing one by asking, but they stay apps — components with manifests, not instruments you can pull out of one and drop into another. Composition between apps is the open problem, and this thesis named it precisely.

6. 2025

   Malleable Software: Restoring User Agency in a World of Locked-Down Apps

   Geoffrey Litt, Josh Horowitz, Peter van Hardenberg and Todd Matthews — Ink & Switch, June 2025

   A gentle slope from user to creator: adapting your tools should get gradually harder instead of hitting a cliff where you suddenly have to be a programmer. The essay is also explicit that LLM code generation on its own does not deliver this — without open, composable infrastructure it just produces more isolated apps — and that the modifying is usually communal, done by someone nearby for the people around them.

   Moos is the move this essay warns about, so it is the test Moos has to pass. The claim is that the gentle slope is the dialog: the brief is readable without being a programmer, correcting it is one sentence, and refusing it costs nothing. The two things Moos does not answer are the essay’s own — composability, as above, and community, because it runs for one person on one machine.