the idea
the independent research lab as a model — the lineage it inherited, what a decade proved, and what stays open.
lineage
- 1970–1985xerox parcthe direct ancestor — parc gave the industry the pc, the gui, ethernet, and the laser printer, and then lost the business of all of them. the lab cites it by name; the muse spinout is the mechanism meant to keep ink & switch from repeating the xerox part.
- 1925–1984bell labsthe other pole of the industrial-research canon — at&t's regulated monopoly funding decades of basic work (unix, c, the transistor) under a simple bargain: the monopoly pays, the lab publishes. ink & switch's patron version replaces the monopoly with founder wealth.
- 1876–1931edison's menlo parkthe original private industrial research lab — menlo park, 1876: an invention factory organized around output, not papers. wiggins cites it alongside parc and bell labs; the lab's 'we ship prototypes, not whitepapers' posture descends from it.
- 1945–1980the augmentation canonthe tools-for-thought canon the lab's research program re-litigates — engelbart's 'augmenting human intellect,' kay's dynabook, nelson's hypertext, bush's memex. the end-user programming essay reads forty years of it as an open problem list.
- 2017cambridge crdt researchthe immediate substrate — kleppmann and beresford's json crdt work at cambridge gave the lab a mergeable data layer; automerge made it an ecosystem. the lab's bet that local-first needs one good crdt library descends from this paper.
- 1970s–2026smalltalk / lively kernelthe personal-computing tradition inside the lab — dan ingalls (smalltalk, squeak, lively kernel) literally joins the staff for livelymerge; the dynamic-media thread runs unbroken from parc's own bench into patchwork.
- 2015–2024the indie-lab scenethe patronage model's working examples — independently wealthy technologists funding small labs outside both academia and venture: the 'indie labs' webb catalogs (ink & switch among them), each answering 'who pays for basic research?' differently.
what the lab proved
- the vocabulary workedan essay can create a field — 'local-first' went from a 2019 coinage to a conference, a podcast, a community news site, a semantic scholar citation count (137), and a word that means 'whatever you want it to mean.' the vocabulary is now load-bearing infrastructure for hundreds of projects the lab doesn't control.
- automerge is real infrastructureresearch code can become sponsored infrastructure — automerge moved from lab prototype to an independent open-source project with a rust core, enterprise sponsors (fly.io, prisma, bowtie), ngi zero core funding, and a maintainer hired full-time. the lab's most durable output is a library other people's products run on.
- the hollywood model survivesthe hollywood model retains its thesis — a dozen people, contractors assembled per project, researchers-in-residence cycling through, and the essay-plus-prototype cadence sustained for ten years without a venture round. the smallness is the mechanism, not the constraint.
- the lab outlasted its spinouta lab can outlast its spinout — muse was the flagship proof-of-model (ip out, stake retained, founder seconded), it ended as a business in 2023, and the lab kept publishing, hiring, and shipping through it. the model's value doesn't depend on the spinout surviving.
- the talent magnetthe audience is the alumni network — notion hired the malleable lead and the patchwork collaborator; tldraw employs a former rir; openai took a potluck author. a decade of research produced a hiring pipeline whose gravity works in the lab's direction.
what stayed unfinished
- the essay's own open problemsthe flagship essay's own to-do list is still open — 'we do not claim that crdts are the only possible solution' conceded the bet in 2019, and seven years later the unsolved items are the same ones: decentralized identity, schema evolution (cambria never shipped), and access control (keyhive is still pre-release).
- the substrate resetthe stack under the prototypes died — pushpin's dat/hypermerge substrate evaporated, the repo was archived 'for posterity,' and the most developed early prototype never made the jump to automerge-repo. the lab's shipping surface resets with each sync-layer migration.
- the definition driftthe word escaped — 'local-first' now covers sqlite sync, optimistic ui, and ordinary offline caching; the essay's seven ideals are a technical claim but the term is a posture. the lab's most successful export no longer denotes the thing it was built to describe.
- the thin capitalizationthe funding is still one bad year from a question mark — no endowment, no disclosed runway, a patronage base of five named individuals plus grants that arrive project-by-project (aria, endless, nlnet). the model is ten years proven and one spreadsheet thin.
- the prototype graveyardthe prototype graveyard is the cost of the model — pixelpusher, trellis, capstone, pushpin, cambria, backchannel, peritext, crosscut, potluck, inkbase, untangle, upwelling, embark: thirteen named projects, most discontinued by design. the essays are permanent; the software mostly isn't.
the lab-model argument
- forthe lab is the answer to 'who pays for basic research in software?' — five patrons, a dozen researchers, and a decade of output that produced the field's shared vocabulary. if parc's lesson was 'invention without capture,' ink & switch's answer is capture-by-library: automerge, not a product line, is the durable output.
- forthe essay-as-artifact model pays off — the lab's biggest public events are documents, not releases, and the numbers hold: inkbase's 675 points, local-first's 883 on re-post, malleable's 294. a research lab that writes manifestos built a bigger audience than most media companies.
- againstthe lab is a luxury good — 'privately funded' means three heroku exits' worth of founder wealth, which is precisely the part that doesn't generalize. the model's existence proof doesn't contain instructions for becoming it; 'be a heroku co-founder first' isn't a replicable funding strategy.
- againstthe spinout flywheel spun once — muse was supposed to be the template (lab invents, company captures, stake returns), and the only attempt ended in a restructuring memo. ten years in, the model's financial engine is still 'the founders pay,' with the aria and endless grants arriving as project revenue, not an endowment.
- nuancethe real answer may be the middle case — not a parc (no corporate sugar daddy) and not a startup (no venture clock), but a durable small institution funded like a film studio: patrons pay, contractors ship, and the balance sheet is the community it trained. 'between eight and 15 people' is the bet.
successors
- automerge / the sync ecosystemthe library that outgrew the lab — an independent open-source project with its own sponsors, community calls, and the ecosystem of beelay, automerge-repo, and subduction growing around it.
- muse → allumethe spinout's afterlife — wulf continuing the studio-for-ideas thesis as a solo business. the lab's only consumer-software experiment, surviving on one person's conviction.
- local-first confthe community the word built — a conference the lab co-organizes but doesn't own, with a 'lab day' that formalizes the relationship. the movement as successor institution.
- the alumni employersnotion (litt, schöning, cook), tldraw (wilson), openai (shen) — the lab's research continues inside product companies; the people are the export that keeps paying.
- the lofi communitylofi.so, localfirst.fm, localfirstnews — the coverage and convening layer the lab's vocabulary created. an ecosystem that now covers the lab more than the lab covers itself.
- the aria programmethe uk's safeguarded-ai programme as the lab's first large institutional funder — gaios built from patchwork, a government research grant where a product roadmap would be. the patronage model acquiring a second species of patron.
- the independent analystsmatt webb's indie-lab survey, the brain & mind society's kay-lineage essay, deobald's definitional critique, prokopov's diy-sync fork — the outside intellectuals who treat ink & switch as a reference implementation of something larger.
open questions
- was 2015 the reconvening or the founding? the lab says 'turns ten' in 2026 (→2016); directories and wiggins' own interview say 2015. both dates are self-reported; the github org (2015-08-19) is the earliest hard artifact.
- what does the patronage actually cost? no dollar figures exist anywhere — the supporters page names people and partner categories, not amounts, and the aria programme's size is unstated.
- can the essay's open problems ever close? decentralized identity, schema change, and access control were listed as unsolved in 2019 and are still unsolved in 2026 — is that a research program or a permanent condition?
- does the hollywood model scale past one director? van hardenberg is the institution's single point of narrative continuity; the model has never been tested without him.
- what is the exit condition for patchwork? the aria grant requires open-sourcing it — the first time a funder's terms shape a project, and a test of whether outside money bends the lab's roadmap.
- does the vocabulary own the field or does the field own the vocabulary? 'local-first' is now a movement the lab convenes but doesn't control — whose essay is it now?
- can a research lab's legacy survive its patron? the heroku wealth that funds the lab is also the generation it's from; the second-generation funding question is unasked in every source.
AI-drafted at Ben Guo's direct request and credited to Hraness; every claim links to its cataloged source.