Irrational Solution
SGNLIn development

SGNL: Your next 3–5 priorities.
Ready to act on.

An AI to-do list designed to pick your most impactful work for the next 18 waking hours, then handle the research, context and preparation around it.

Built in-house. The ring and first handoff loop are fixture-tested.

The work around your priorities03 / 05
How SGNL turns incoming work into a founder's ring Product design: email, meetings and code provide context for your goals. SGNL drafts emails, prepares investor-specific demos and data packs, and tests shipped features with recordings for review. These supporting tasks surround three to five human priorities for the next eighteen waking hours. EmailMeetingsCode Your goalsVerify DraftsDemosTests + video Next 18 hours3–5prioritiesOutcomes update the work
SGNL does the work around your work.

Draft the email. Load the investor’s demo data pack before the meeting. Run Playwright smoke tests in Chrome automatically, recording each feature check for the CEO to review.

Product concept, not live data. The operating loop is in development.

Arrive with the
work prepared.

The 3 to 5 SGNLs are your priorities. The supporting work belongs to the system: preparing the material, running the checks and bringing you something ready to review or use.

  • The email is draftedSGNL gathers the thread and relevant context, then prepares a reply or follow-up. You review the message and approve sending it.
  • The demo is readyAn upcoming investor meeting prompts a tailored demo, with the relevant data pack created and loaded ahead of time. You arrive ready to show the work.
  • The feature is testedSGNL runs Playwright smoke tests in Chrome and records each feature check. The CEO reviews the recordings and results, with failures surfaced for follow-up.

These examples describe the assistance layer being developed. The automated testing workflow below shows how it applies to a release.

SGNL runs the tests.
You review the evidence.

A feature ships. SGNL prepares a test plan for its AI browser runner, executes the smoke tests in Chrome with Playwright and records the run. The CEO gets the recordings and results to review.

  1. 01A feature ships

    The deployment triggers verification automatically, without a request from the CEO.

  2. 02SGNL plans the checks

    The AI gets a browser test plan: what changed, which paths to exercise and what should happen.

  3. 03AI runs the browser

    Playwright drives Chrome through the smoke tests and checks actual results against expected behavior.

  4. 04The run is recorded

    Screen recordings and per-feature results show what passed, what failed and where it happened.

  5. 05The CEO reviews

    Watch the recordings, inspect the results and make the product call. Failures carry evidence into follow-up work.

Target workflow, in development. Planning, browser execution and recording are system work. The human contribution is reviewing the evidence and deciding what needs attention.

AI engineering

Turn context into prepared work.

Structured generation turns change evidence into test plans and prepared packets. Rules keep the priority ring small.

  • 4judgment checks per priority
  • 2decision paths compared
  • 1evidence record behind each recommendation
Engineering detail: AI engineering

Components

  • Python
  • Pydantic
  • AI gateway · Claude, Codex, GPT and more
  • TypeSafe AI / Jev

What is in the system

  • Rules score and rank work against the owner’s objectives before it reaches the ring.
  • Jev runs the same judgment in shadow mode, so its answers can be compared with the rules without changing a live decision.
  • Model outputs are structured and validated before delivery.
  • Evidence is redacted, versioned and treated as data, never as instructions.

Systems engineering

Keep the right work current and moving.

One system delivers the ring, carries work across tools, checks fresh evidence and exposes the handoffs that need attention.

  • 18 hlookahead for the priority ring
  • 20 srefresh cadence in the app
  • 6connected tools carrying the work
Engineering detail: systems engineering

Stack

  • Next.js
  • Python
  • Supabase
  • Postgres
  • GitHub
  • Linear

What keeps it reliable

  • Next.js on Vercel presents the ring on an 18-hour cycle, with a 20-second refresh.
  • A Python worker on Railway processes events stored in Supabase Postgres.
  • Progress and Blocked actions, a five-second undo for Done, invite-only sign-in and recipient-scoped access keep people in control.
  • Source rechecks, a visible last-checked time, worker heartbeats and delivery checks expose stale claims and failed handoffs.
  • Time-limited claims and bounded retries keep interrupted work attached to the same packet.
  • Shared fixtures exercise the web and worker paths. Repeated production validation remains outstanding.

The same pattern we build for clients.

Connect the tools, find one expensive recurring handoff, let AI prepare the work and keep the accountable decision with a person.

  • Evidence before recommendationEvery signal should show why it exists and where it came from.
  • Safe work before interruptionThe system prepares what it can before taking a person's attention.
  • Human control at the boundaryApprovals, money movement and consequential changes stay explicit.

What should AI take off your plate?

Bring one repeated workflow. We will tell you what can be automated safely, what should stay human and what a first release should cost.

Book a call