Pointbreak / working principles

Evidence should outlive the agent that produced it.

Software is moving from humans writing every line to humans directing increasingly autonomous systems. The bottleneck is no longer generation. It is deciding what deserves trust.

Pointbreak makes the basis of those decisions inspectable: what was claimed, what supports it, what remains uncertain, who answered, and which assessment they recorded.

← Back to the product

Nine rules for evidence-backed decisions

These principles shape the Pointbreak model, its current product, and what Pointbreak refuses to become.

  1. 01

    Durable over conversational

    Store the facts that matter. Discard conversational exhaust.

  2. 02

    Evidence over assertion

    Explanations provide context and rationale. When a claim can be tested, observed, or tied to the work, it should be supported by evidence another actor can inspect.

  3. 03

    Decisions over transcripts

    Conversations are optimized for generation. Evidence is optimized for decisions.

  4. 04

    Attention over surveillance

    Spend human attention where it has the highest leverage without requiring continuous supervision of every agent action.

  5. 05

    Human judgment over automation

    Agents maximize autonomy. Humans retain authority.

  6. 06

    Accountability over enforcement

    Name every writer and make unanswered requests visible without turning Pointbreak into a central gate or controller.

  7. 07

    Work objects over sessions

    Evidence belongs to the work, not the temporary session, model, or conversation that produced it.

  8. 08

    Typed records over free-form text

    Claims, observations, validation, requests, responses, and assessments keep their meaning without requiring a transcript to interpret them.

  9. 09

    Ambiguity surfaced

    When reviewers disagree, revisions compete, or a decision becomes stale, preserve the uncertainty instead of manufacturing consensus.

What Pointbreak is not

Boundaries make the category legible.

  • 01
    Not a chat log.

    It stores durable evidence and decisions, not every message that led there.

  • 02
    Not shared model memory.

    The goal is better human decisions, not more agent context.

  • 03
    Not an orchestrator.

    Pointbreak records what work produced. It does not schedule or control the agents.

  • 04
    Not a gate.

    Pointbreak records and surfaces. People and policy decide what binds.

  • 05
    Not a platform or product suite.

    Pointbreak Review is the only current product. Other possible evidence sources are directions, not shipping products.

Current product, broader model

REVIEW / AVAILABLE NOW

Review makes the model concrete.

Pointbreak Review is a durable, local-first record for changes built by humans and coding agents together. It carries the claims, evidence, questions, responses, and assessments.

Explore Review
THE PURPOSE

Not smarter agents.
More confident humans.

See Pointbreak in the work loop