the idea
the fast-editor-as-collaboration-layer idea — the lineage from engelbart through atom, what zed proved, what stalled, and what delta is for.
lineage
- 1968engelbart's nls and xanaduthe mother of all demos and the mother of all unbuilt software — sobo explicitly cites engelbart's nls/augment vision and doug's lost collaboration paradigm in the first zed post, and 'xanadu was waiting for agents' puts delta in the docuverse lineage.
- 2011–2022atom and electronthe first attempt — hackability plus usability on a webview shell, which delivered electron to the world and the performance ceiling to its own authors. 'we created electron to create atom, but i can't imagine a worse application for electron than a code editor.'
- 2013–tree-sitterthe parser that outlived its editor — brunsfeld's incremental glr engine went to github.com, neovim, helix, and emacs, and became zed's syntax layer and extension-grammar format.
- 2017teletype and crdtsthe collaboration primitive — teletype's custom string-wise crdt proved real-time multi-cursor editing in production; 'conversations should happen close to code' became zed's founding assumption.
- 2017–2019xray and memothe failed rehearsal — xray's rust core and client/server crdt architecture died at github, but its memo side-project ('an operation-based version control system that tracks changes at the level of individual keystrokes') is the direct ancestor of deltadb eight years later.
- 2016 → 2025lsp → acpthe protocol precedent — lsp unbundled language intelligence from monolithic ides; acp applies the same move to agents. zed's stated strategy is to be the neutral surface rather than the deepest bundled model.
- 2018–2021the rust-native stackthe native-stack conviction — raph levien's rust ui writing and xi-editor experience as the cited nudge; gpu-rendered immediate-mode rust (gpui) as the only way to 'pixels on the next refresh of your display — every time.'
what zed proved
- the native editor worksa native, gpu-rendered editor can be built and loved — 'hundreds of thousands of developers' daily at 1.0 (self-reported), ~90k github stars, a two-year-long 'windows when?' demand curve, and 2,153 hn points for the 1.0 post. the speed claim is the moat.
- the open-source flywheelopen source is the distribution and the hiring engine — 1,576 points at relicensing, the registry grew to ~1,500 extensions in two years, the debugger shipped as a community-led effort, and 'hired through github' converted contributors into employees.
- agents come to the neutral surfacethe neutral-surface bet works — google initiated contact for acp, jetbrains adopted it, claude code and codex and opencode and cursor all run inside zed. when the model vendors fight, the editor can host them all.
- the craft is crediblethe team keeps its own credential — the people who built atom's collaboration stack wrote the crdt papers' production sequel, and the engineering blog ('zed decoded') became a respected rust publication independent of the product.
- the post-commit world existsthe second half of the plan exists in code — deltadb and delta shipped to public beta with pull requests disabled on the company's own repo: '33 of us have landed 570 changes to main' (self-reported). the memo idea from 2018 is a running product.
what stayed unfinished
- collaboration in the editorcollaboration inside the editor didn't stick — 'very few people leverage collab today' (company admission), the chat panel was removed wholesale, and channels survive as the office the team itself runs, not the product users adopted.
- the extension gapthe extension gap persists — a deliberately conservative wasm api produced ~1,119 listed extensions mid-2026 against vs code's tens of thousands; 'my one plugin isn't there' remains the standard switching objection.
- free aifree ai didn't survive contact with the bill — free beta → at-cost billing → prompt bundles → token pass-through in eighteen months, each step read by some users as a smaller version of the same bait-and-switch.
- gpui for everyonegpui as a standalone framework stays aspirational — apache-2 and documented, but unpublished as a supported crate; the community (gpui-component, loungy, zedis, termy) builds on it anyway, which is evidence for and against the 'ui framework for everyone' claim.
- the ai trust ledgerthe ai-direction argument never fully closed — disable_ai shipped as a genuine concession, yet zedless exists and 'not much of an actual refutation' was the hn verdict on the economics defense.
the agent-era turn
- forthe whole history points one direction — teletype's crdt collaboration, xray's operation-based memo, channels, acp, and deltadb are one fifteen-year project. the company isn't pivoting toward the agent era; the agent era finally caught up to what they were building.
- againstthe collaboration record cuts the other way — the team that argued 'conversations should happen close to code' watched 'very few people leverage collab today,' removed chat, and moved the bet to a separate app. delta is a relaunch, not a continuation.
- forthe neutral-surface strategy is correctly calibrated for a model-vendor knife fight — acp means zed doesn't have to out-model openai or anthropic, just out-host everyone. jetbrains adopting the protocol is the strongest external validation yet.
- againstthe economics are still unproven — the editor is free, ai tokens were sold at a loss until token billing, business seats are six months old, and delta's paid plans don't exist yet. sequoia's $32m is a bet on the second half of the plan, not a verdict on it.
- nuancethe moat may be the stack rather than the editor — if deltadb wins, delta matters more than zed; if it doesn't, zed is an excellent editor in a market where the giants give editors away. the company's own framing already separates the two bets.
- againstthe trust ledger is the real ceiling — telemetry, cla, arbitration, token pricing, and feature removals each passed, but they compound into the question every 'can i switch yet' thread asks. zedless is the ledger made executable.
- forthe strongest evidence for delta is that the company runs on it — 'zed is our office' made the dogfooding literal, and disabling pull requests on delta's own repo is a risk no marketing team would fake.
successors
- deltathe idea's own successor — the collaboration thesis grown into a separate product where threads replace pull requests. if it works, zed becomes the on-ramp, not the destination.
- cursorthe opposite bet running at full speed — fork vs code, bundle the model, own the agent. the reference incumbent zed's neutral-surface strategy is defined against; cursor now also speaks acp.
- vs code + githubthe incumbent stack zed has to displace twice — the editor monopoly atom's own electron enabled, plus the commit/pr workflow delta is built to replace. atom alumni now staff both sides.
- warpthe road not taken — sobo's three-month employer built the agent workbench inside the terminal instead of the editor. same 'rebuild the tool for agents' impulse, different surface.
- zedlessthe community fork as successor-with-an-asterisk — the editor thesis stripped of the company's services layer: no telemetry, no cla, no rugpulls. what gpl was always for.
- jetbrains + acpthe protocol succeeding without zed — if acp becomes lsp-for-agents across the industry, the neutral-surface win outgrows the editor that minted it.
open questions
- the series b valuation is undisclosed — sequoia led $32m at an unknown price; crunchbase news reports 'over $42m' total but no post-money figure.
- adoption is self-reported — '150k monthly active developers' comes from sequoia's blog, 'hundreds of thousands daily' from the 1.0 post; no independent measurement exists.
- does delta's 'threads replace prs' survive contact with real teams — 33 employees landing 570 changes on their own repo is evidence of dogfooding, not of the market?
- is the revenue model pro + business + delta plans enough to clear a $42m+ raise — or does the open editor stay subsidized by services that haven't yet proven their take rate?
- the warp ip separation is undocumented — how did the codebase sobo carried into warp for three months come back out clean?
- collab adoption inside zed stayed low — does the delta re-platforming fix a product problem or a demand problem that will follow it?
- does acp commoditize zed — if every editor speaks the protocol and the agents belong to the model vendors, what is the editor's durable position?
- the extension gap is structural — a sandboxed wasm api may never host the plugin surface vs code users expect; is 'conservative api' a strategy or a ceiling?
AI-drafted at Ben Guo's direct request and credited to Hraness; every claim links to its cataloged source.