6. Communities of Practice

Why this page closes the collection

The previous five pages give a position. Page 1 sets the political character of the question. Pages 2 and 3 give formal definitions of functional and intrinsic accessibility. Page 4 names the cost-aware analysis that holds the formal definitions honest. Page 5 gives the methodological substrate that lets the formal definitions be built rather than just stated.

Page 5 ends with the structure: domains, bridges, compilers, the recursive-design frame. This page asks what is happening in that structure. The actors inside the structure are not passive compositional units; they play roles, they collaborate, they compete, they learn. Treating them as a community of practice rather than as a compositional graph names the dynamics that the recursive-design frame leaves implicit. Those dynamics are also the bridge to the multi-agent framing that the 2029 framework resumes.

The user interface as a community of practice

A community of practice, in Wenger’s sense, is a group of actors playing roles in a shared enterprise, with collaboration, learning, identity, and shared repertoire as the binding forces. The pattern was developed for analysing professional groups — midwives, claims processors, Naval quartermasters — but it generalises in interesting ways to other actor-collections, including the components of a running interface.

At least nine entities can be named as actors in the ordinary user interface: the User Capability Context (what the user can do in the body and the moment), User Preference (what the user has chosen to ask for), the Content Provider (the source of the underlying material), the Content Manager (which material is currently in scope), the Content Presenter (how the material is rendered), Custom & Practice (the convention that says “forms look like this”), Cultural Context (the convention that says “red means danger”, or “red means luck”, depending on where you are), the Device Capability Context (what the device can do), and the Operational Context (where the user is, what time it is, who else is around).

Each of these has what Wenger would call an identityin the community — a stable role and a characteristic pattern of contribution. Custom & Practice is the damping factor, pulling every interface toward rectangular grids and visual metaphors regardless of the user-platform. Cultural Context is the guardian of meanings, intervening when the Content Presenter would otherwise commit a cultural error. User Preference is the king the whole enterprise exists to serve, and the actor whose objections, when they appear, are the most expensive to override. The roles are not symmetrical, and the asymmetries themselves carry meaning.

The shared enterprise is straightforward: provide content and collate feedback between device and user. Every actor contributes to that enterprise, by negotiation and by routine. The enterprise is the shared object that holds the community together.

User populations as a community of practice

The same analysis applies one level up. The users of an interface, considered collectively, also form communities of practice — particularly when the users share a disability-relevant capacity profile and are connected through peer-group infrastructure (advocacy organisations, online forums, mutual-aid networks, family circles, professional bodies for accessibility practitioners). The RNIB, the RNID, the MS Society, and their international counterparts are communities of practice in this sense.

The interesting consequence at this level is that portable user profiles act as memes. A configuration that one user finds works for them is shared with peers in the same community; the peers copy it with variation; the variations that work propagate further; the variations that don’t get dropped. Peer-group infrastructure provides the selection pressure that makes profile evolution non-random. Disability communities, in this frame, act as evolutionary substrates for accessibility settings.

That observation has a falsifiable consequence:user interfaces that optimise user profiling to improve the evolutionary process will be more accessible than those that do not. Profiles that propagate easily, that vary at the right granularity, and that recombine cleanly when users move between communities should produce accessibility outcomes that converge faster on good fits than profiles which stay siloed, unsharable, or coarse. The claim is testable; the experimental apparatus does not yet exist; the prediction is on the table.

The reframe also shifts what “accessible” asks for. The narrow reading is does this interface adapt to a given user profile? The wider reading is does this interface support the social processes by which good profiles emerge and spread? An interface that nails the first while ignoring the second is helping each user in isolation; an interface that does both is helping the community learn.

Inaccessibility as community dysfunction

The reframe yields a useful definition by negation. If a working user interface is a functional community of practice, an inaccessible user interface is a dysfunctional one. Three modes of dysfunction recur, and each maps onto a class of accessibility failure that the formal definitions of pages 2 and 3 can detect but cannot diagnose.

1. Competition between actors for limited resources. The audio channel is finite; the screen real estate is finite; the user’s attention is finite. When multiple actors want the same resource without a mechanism for arbitration, the resource is claimed by whichever actor gets there first, which is rarely the one with the strongest claim. The accessibility prompt loses to the marketing pop-up because the marketing pop-up was added last and runs first. The screen-reader announcement loses to the live-region update because the live region changes too often. Each individual collision is a bug; the pattern is a failure of the community to arbitrate. (This is also the failure mode that the Shlaer-Mellor assigner state model exists to handle.)
2. Inappropriate power structures. The Content Presenter overriding User Preference is the canonical case — a developer fixes a font size into the stylesheet despite the user having set a different one in their browser; the Presenter has decided it knows better than the Preference. The reverse is rarer but exists: User Preference being permitted to override Cultural Context in ways that produce accidentally offensive output, or User Preference being permitted to override Custom & Practice in ways that leave the user navigating an interface no other user would recognise. Both directions are governance failures: the wrong actor is making the call.
3. Inappropriate decision-making. The community needs models of its actors and feedback channels between them. When the models are missing, the loop is open. When the granularity is wrong — all blind users can’t see, all deaf users can’t hear, all motor-impaired users use the same switch interface— the decisions get made against a category that does not match the user. The user is the residual; their experience is the accident.

All three failure modes are visible in current accessibility practice; all three are addressable in the recursive-design frame; all three become more tractable when you have a vocabulary for the community-of-practice dynamics. The formal definitions on pages 2 and 3 detect the failures; the community-of-practice analysis explains them.

Why competition-only agentic AI is the wrong frame

The current agentic-AI literature is competition-shaped. Auctions, negotiations, marketplaces, multi-agent reinforcement learning — all of them model the relationships between agents primarily as competition for scarce resources, with cooperation as an emergent equilibrium when the game permits it. That framing is plausible; it is also incomplete.

The community-of-practice framing is collaboration-and-learning-shaped. Agents do not just bid against each other for the audio channel; they tune their relationships, they learn collectively, they hold shared identity, they migrate between roles as the shared enterprise demands. Same agents; richer model. The contemporary literature has the first half of the story. The second half — communities of practice with multi-agent dynamics, applied to a domain where the shared enterprise actually means something — is materially open.

Whether the game is cooperative or non-cooperative is the wrong question. The actual question is how do agents in a community of practice negotiate competition for resources while sustaining the shared enterprise that makes the cooperation valuable in the first place? That question is what the 2029 framework takes up.

Where this points

Six pages, one position. Accessibility is political; functional and intrinsic accessibility are the formal measures of two distinct properties; equivalent experience is the cost-aware analysis that holds them honest; recursive design is the methodological substrate that makes the definitions buildable; and communities of practice name what is actually happening inside the structure once it is built.

The point of the collection is not to settle the field. The point is to articulate a coherent position that is formally precise, methodologically grounded, and politically honest — one that a serious practitioner or a serious researcher can argue with, extend, or refute. The forward-looking work that flows from this position is the 2029 framework; the practical work that flows from it lives in Paradise; the substrate that makes both possible runs in code at Action Language.

Continue

← Previous: 5. The Shlaer-Mellor lens · See also: The 2029 framework →