Video summary

Martin Kleppmann — Local-first in an unstable world

Main summary

Key takeaways

Technology

High-level framing: tech change alongside geopolitical shifts (1989 → today)

The speaker links 1989 as a symbolic turning point:

  • Fall of the Berlin Wall: a political shift from Cold War dynamics to a period of relative post–Cold War stability.
  • Tim Berners-Lee’s Web proposal: a technological shift that accelerated the mainstreaming of the internet.

They argue that the internet evolved from a more protocol-driven/decentralized era into a more centralized platform/product era:

  • Pre-web: open protocols for communication
    • Examples: email, Usenet (social/community).
  • Post-web: growth of centralized websites/social networks, followed by cloud computing to support centralized infrastructure.

Why centralization happened—and tradeoffs

Benefits attributed to centralized web/cloud platforms

  • Better user experience (compared to earlier “protocol geek” interfaces).
  • Spam/abuse mitigation
    • Usenet struggled with spam; centralized platforms handled it more effectively.
  • Efficiency gains
    • Centralized storage can reduce overhead from multiple replicated copies and integrity validation.

Costs attributed to centralization

  • Reduced resilience when systems or providers fail.
  • Increased vulnerability to abuse and lock-in at the level of providers and nations.

The new risk: unstable allies + “cloud as leverage”

The speaker claims the roughly 35-year era of peace and stability is deteriorating, driven by:

  • AI disruption (unknown positive/negative impacts, but major disruption expected).
  • Renewed great-power rivalry and ongoing war in Europe (reference to Ukraine).
  • Concern that the US may be less reliable as an ally of Europe.

Core technological concern: Europe’s dependency on US cloud providers

A central worry is European dependency on US cloud providers. Example scenario: if the EU were pressured via trade sanctions, Europe could be locked out of US cloud/AI services, impacting computing services since AI often runs on cloud.

Evidence/examples of “lockout” behavior

  • References Trump’s repeated calls about Greenland, extrapolating potential sanction/tech restrictions.
  • Cites an example where Anthropic’s model access was reportedly limited to US citizens first, then broader access was blocked.
    • (As described in the talk.)

Broader point: “everything is software now”

They argue that software control can become a geopolitical weapon—e.g., John Deere remotely disabling farm machinery via software controls.

Market-share analysis of European cloud dependence

  • European providers allegedly serve ~15% of the EU cloud market.
  • AWS + Azure + Google Cloud allegedly serve ~70%.
  • This makes cloud a realistic leverage point, since it’s “hard to migrate off quickly.”

Link to “technological sovereignty” and how Local-First addresses it

The talk positions Local-First as a way to improve technological sovereignty and resilience.

It claims Local-First changes what is primary:

  • Classic cloud model
    • The cloud database is primary.
    • Clients are “thin” and must fetch/upload.
  • Local-First model
    • The user’s device holds the primary data copy.
    • Cloud becomes mainly for:
      • backup
      • data synchronization for collaboration

Cloud isn’t eliminated; it’s deprioritized so it is secondary rather than authoritative.

Product/architecture implications of Local-First

Local-First enables a different architecture approach:

  • Multiple sync services can coexist because the sync engine isn’t necessarily the sole authority.
  • Potential peer-to-peer sync between collaborators’ devices when connectivity allows.
  • Real-time collaboration is still supported, but designed around local primary state.

“Seven ideals” of Local-First (from a 2019 essay; key ones summarized)

  • Fast UX via local reads/writes (avoids network round trips).
  • Offline capability: users can read/write locally and sync later.
  • “Incredible journey proof”: resilience if a vendor shuts down—apps keep working locally even if the service company disappears.
  • User agency: privacy/security/control over data.

Benefits for developers (beyond users)

  • Building against a sync engine is described as a nicer programming model than working directly with REST APIs.
    • REST complexity like timeouts/failure handling and “did the server get it?” semantics is reduced.
    • The sync layer “handles the annoying distributed systems problems.”
  • Developers may reduce the need to maintain backend teams if they can rely on backend/sync services provided by others.

Societal/geopolitical benefit: resilience against lockout

The key enabling principle is framed as decentralization, expanded beyond avoiding a single failure point:

  • Avoid dependence on a single server
    • Replication/fault tolerance.
  • Avoid dependence on a single company
    • Make it easier to switch providers and prevent lock-in.
  • Avoid dependence on a single country
    • Use providers across jurisdictions.

To make provider switching feasible, the speaker argues for commoditization through standardization.

“Local-First hourglass”: a proposed standard layer for interoperability

The speaker uses the “internet hourglass” analogy:

  • A widely interoperable narrow waist protocol (e.g., IP/HTTP) enables many applications without changing the lower layers.

For local-first, they argue a similar shared layer is needed:

  • Many apps at the top (docs, notes, spreadsheets, editors).
  • Many CRDT libraries beneath (diverse models/tradeoffs).
  • Many collaboration/sync mechanisms:
    • sync providers
    • self-hosting
    • peer-to-peer
    • end-to-end encryption
  • But a shared “narrow waist” standard should sit underneath apps so they can interoperate across sync providers.

Teaser: “Sediment Tree” (not fully explained)

The speaker (and their group/company) are working on a protocol/data structure called Sediment Tree.

They present it as a possible candidate for the “narrow waist” concept for local-first interoperability, and invite others to consider what a good local-first “narrow waist” standard should be.

Main speakers / sources mentioned

  • Main speaker: Martin Kleppmann (as indicated by the video title).
  • Sources referenced within the talk:
    • Tim Berners-Lee (Web proposal, 1989)
    • The Economist (commentary about AI/European access becoming “unthinkable”)
    • Adam (mentioned as “as Adam already said” regarding AI)
    • Anthropic (example of access restriction to US citizens)
    • John Deere (remotely disabling agricultural machinery example)
    • Mentions of AWS, Azure, Google Cloud (market-share dependency)
    • CRDTs (concept; multiple CRDT libraries alluded to)
    • NATO summit / Donald Trump (policy/geopolitical trigger referenced)

Original video