Panel Discussion: The Big Questions of Local-First

Overview

Three of the four original local-first-software essay authors (Kleppmann, Wiggins, van Hardenberg) sit on a panel with Aaron Boodman (Rocicorp/Zero) — and the result is the most honest airing of internal tensions in the local-first community. The central question isn't about CRDTs or sync protocols. It's about values: is local-first a technology choice or a moral commitment to users?

Aaron's Concentric Circles

The most useful framework from the discussion. Aaron proposes three rings to classify apps that use sync engines:

flowchart TD
    subgraph outer["Synced Applications"]
        direction TB
        O1["Instant reactivity, optimistic UI"]
        O2["Server remains source of truth"]
        O3["Linear, Figma"]
        subgraph middle["Offline-First"]
            direction TB
            M1["Designed and tested for offline"]
            M2["Genuine offline resilience"]
            subgraph inner["Local-First"]
                direction TB
                I1["All seven ideals"]
                I2["User owns data forever"]
                I3["Survives developer shutdown"]
            end
        end
    end

Most developers using Zero today sit in the outer ring — they want instant UI, not offline resilience. That's fine, Aaron argues. The inner ring is the aspiration, not the entry requirement.

The Offline Writes Debate

This is where the panel gets heated. Aaron temporarily disabled offline writes in Zero, and the reactions split perfectly along ideological lines:

Boodman's position: CRDTs give developers a false sense of safety. Two concurrent edits to the same paragraph produce "consistent but nonsensical" results — text that neither user typed. The longer someone is offline, the more work they invest and the higher the chance of destructive merges. A cruel double whammy.

Kleppmann's position: Blocking a user from editing their document on a train is patronizing. If you're the sole editor, there's zero conflict risk. The solution isn't to remove the capability — it's to build version-control UX (branching, diffing, review) for structured data, the way developers already do with Git.

Van Hardenberg (mediating): Both are right at different layers. Ink & Switch built Patchwork — a prototype that gives rich documents pull-request-style review. The branching approach makes offline tractable without pretending auto-merge solves everything.

AI and Local-First: Complementary, Not Conflicting

Van Hardenberg had the sharpest framing here, citing Geoffrey Litt: "AI collaboration is a version control problem." When an agent edits a 17,000-word document, you need the same branching and diff primitives you use for code. Local-first's change-tracking capabilities fit naturally.

On the centralization fear — van Hardenberg argued it's already fading. DeepSeek proved you can train cheaply by drafting off existing work. Unified memory architectures are putting 64-256 GB of usable GPU RAM on consumer devices. Capable models will ship with your OS within a few years.

Notable Quotes

"For me, sync engines are a particular technology. Local first for me is a value, and that value is empowering users." — Martin Kleppmann

"I sometimes tell people that working on CRDTs is my penance for creating a million single points of failure by running all those Heroku Postgres databases." — Peter van Hardenberg

"I've experienced a product that works really well offline like five or less times in my life." — Aaron Boodman

"Stopping the user from editing their document on the train — it's almost patronizing." — Martin Kleppmann

The Deeper Tension

What makes this panel fascinating is that the disagreement maps to a fundamental question: who is your primary stakeholder?

Kleppmann optimizes for the end user — their agency, their data, their freedom. Boodman optimizes for the developer — ship tools that work reliably, don't moralize about what gets built. Van Hardenberg mediates — protective of the essay's ideals but welcoming the broad community.

This mirrors exactly the idealist vs. pragmatist framing from local-first-software-pragmatism-vs-idealism, except here it's not abstract — it's three people who deeply respect each other disagreeing in real time about whether ideals should constrain shipping decisions today.

Connections

local-first-software — The foundational essay co-authored by three of the four panelists. Every argument in this panel traces back to whether the seven ideals are a north star or a shipping requirement
sync-panel-discussion — The companion panel at the same conference where Aaron Boodman debates sync architecture with different panelists (Cowling, Schickling, Mathews). That panel is about how to build sync; this one is about why
local-first-software-pragmatism-vs-idealism — Adam Wiggins' solo talk articulating the idealist-pragmatist spectrum. This panel is the live version of that tension, with Kleppmann and Boodman embodying opposite poles
the-past-present-and-future-of-local-first — Kleppmann's keynote at the same conference where he proposes the definition that software must survive its creator's shutdown. His position in this panel is the debate-ready version of that definition