Skip to content

Building OS-Intelligence a reasoning layer for Claude Code

View live GitHub
Case study content
Open-source Apache 2.0 Claude Code 15 skills 7 sub-agents Launched May 2026

The reasoning layer for your operating system.

Built solo in Claude Code over three months. Capture, synthesis, retrieval - the reasoning layer the substrate didn't have, shipped open-source. Now in production with the first pilot users.

A current-state.md file from a live project, showing project position, stakeholder dynamics, in-flight work and standing decisions - all auto-synthesised from meetings, documents and notes.
A live current-state.md from one of my projects - auto-synthesised from meetings, documents and notes, loaded by /os-start at the top of every session.

The Challenge

AI without memory is a chatbot. AI with memory is a reasoning partner. PMs run hundreds of conversations across stakeholders, projects and decisions - and lose the thread the moment the session closes. OS-Intelligence holds the thread, so reasoning compounds.

The
Process

  1. 1

    PM-OS as the proving ground

    • – Built 40+ Claude Code skills covering the full PM workflow - discovery, PRDs, decisions, status updates, stakeholder prep.
    • – Layered memory: project context, stakeholder intelligence, session continuity.
    • – Sub-agents for review - engineer, designer, executive, customer voice.
    • – Three months running my entire PM career on it. Every PRD, status update, brief.
  2. 2

    Identified what generalises

    • – Skills hierarchy and routing logic.
    • – Memory design and context loading patterns.
    • – Eval loops and sub-agent review chains.
    • – The line between "my workflow" and "a system other PMs can install."
  3. 3

    Productized as OS-Intelligence

    • – Open-source, Apache 2.0 licensed, one-command install in any Claude Code repo.
    • – Sub-agent fan-out for synthesis - a two-hour transcript doesn't block a thirty-second answer.
    • – Model routing - Sonnet for routine reads, Opus for structural reasoning. Pay for cognition where it matters.
    • – Local-first - every entry is a markdown file in your repo. Nothing leaves your machine.
  4. 4

    Shipping in public

    • – Soft launch at os-intelligence.ai in April. Public launch this week.
    • – Pilot conversations across builder-PMs, scale-up Heads of Product, and mentees.
    • – Weekly releases. User feedback compounding directly into the roadmap.
    • – Companion stakeholder-intelligence app and n8n job-search pipeline open-source on GitHub - proof points for the broader AI-native practice.

The
Outcome

A reasoning layer live in production. Open-source, installable, shipping weekly.

Live and open-source

Shipped at os-intelligence.ai. Apache 2.0. Clone, install in any Claude Code repo.

Memory across the whole job

People, projects, meetings, decisions, open questions - extracted, linked, resurfaced. The dividend: 15 hours a week back on my own work.

First pilot users live

Onboarding repos seeded with the first collaborators. Weekly cadence, feedback flowing into the roadmap.

Henry Latham
"Simon is a talented AI product manager. A stand out feature for me is his ability to work effectively at pace in a highly ambiguous environment."
Henry Latham - Founder of Product MBA

Key takeaways

What does AI-native product management look like as a system, not a chatbot? A reasoning layer: capture, synthesis, retrieval. Scattered meetings, documents, notes and chats turned into accurate persistent memory. PM-OS proved the pattern in private. OS-Intelligence ships the reasoning layer in the open. Built in Claude Code over three months. First users live.

Built independently. Available for senior AI PM and AI transformation roles at AI-native companies.

Get in touch

More work

See all case studies →