Cadence

Cadence Labs / Writing

The interface is the unsolved problem.

We build Cadence and we publish what we learn. Findings from our own product and our own data on multi-agent orchestration, voice as a command channel, and building around Claude Code while working alongside other coding agents.

  1. No. 04
    /Engineering/5 min

    Scouts before editors: a division of labor for parallel agents

    Give four agents equal authority over the same work and you get four plausible answers you now have to referee. Many readers, one writer.
  2. No. 01
    /Manifesto/4 min

    Cadence Labs: studying the UI of agency

    Most AI products bolt a chat box onto existing software. We think the interface, not the model, is the constraint that defines the next decade.

    Early 2026
  3. No. 02
    /Interaction/5 min

    Dictation is the UI layer for agentic AI

    Models output 1500+ words per minute. Click-driven interfaces serialize at four seconds per decision. Voice is the only input channel that scales to multi-agent control.

    Early 2026
  4. No. 03
    /Engineering/5 min

    Why we only ship Claude

    Multi-model routing is the default pattern in 2026. We picked one family and committed. The product is better for the constraint.

    Early 2026

Release notes

  1. v0.5.5
    /Release

    Cadence 0.5.5

    Dictation appears as you speak, updates install themselves, and Claude Code in your other terminals shows up on the toolbar.
  2. v0.5.4
    /Release

    Cadence 0.5.4

    Anonymous usage counts, with a notice and an off switch.
  3. v0.5.3
    /Release

    Cadence 0.5.3

    The Inbox only asks when an agent is waiting on you.

Every release →

Updated quarterly. No newsletter, no subscribe modal.