Architecture & Philosophy · May 2026

Two Teams,
One Architecture

When a bottom-up ecosystem analysis and a top-down platform design converge on the same architecture, it suggests the team is circling something real.

The Cortex Convergence
Sam Schillace · Brian Krabach · MADE:Explorations
The Evidence

Nobody drew
a master plan

And yet, when you step back and look at what the team built over the past year, what emerges is an operating system.

442+
Capabilities tracked
15
Team members
500+
Repositories
20
Capability clusters

Three people independently built personal memory vaults. Four people independently built M365 connectors. The same convergence loop — propose, judge, refine — appears in 8+ projects that never referenced each other. This isn't a coordination failure. It's a demand signal.

— The Semantic OS, Section 2: The Landscape at a Glance
The Semantic OS Vision

Intelligence as
ambient medium

Not a tool you invoke, but the medium everything runs in. Agents communicate through semantic contracts: natural-language descriptions of what they can do and what they need.

Traditional OS

  • Manages bits and bytes
  • Processes are code
  • APIs are syntactic contracts
  • Memory is RAM / disk
  • Users invoke tools

Semantic OS

  • Manages intentions and context
  • Processes are agents (code + model + context)
  • APIs are understood, not just parsed
  • Memory is context windows + knowledge
  • Intelligence is ambient substrate

The system's purpose is to elevate. Every cycle of operation raises the level of abstraction at which the human works. The objective function is leverage: maximize the long-term value of outcomes per unit of human attention.

Framework

Three layers, genuinely different in kind

Not separate subsystems managed by a central kernel. Emergent properties of a system where every component is itself an agent that can reason, negotiate, and adapt.

3

Elevation

The system's purpose. Continuously raise the level of abstraction at which the human operates. The floor rises — what was hard last month is infrastructure this month.

2

Self-Model

The system understands itself. A live, continuously-updated model of what exists, what's overlapping, what's missing, what's evolving. The highest-leverage gap in the current ecosystem.

1

Substrate

The plumbing. Semantic contracts, the memory hierarchy (L1–L5), progressive formalization, the communication bus, and the process scheduler.

The abstraction ratchet: the substrate formalizes because agents notice patterns (Layer 2 observing Layer 1). The self-model curates because it understands what matters (Layer 2 serving Layer 3). Every formalization removes something from the human's attention queue.

Core Innovation

Progressive formalization

Every interaction in the system exists on a spectrum. The semantic OS operates at all points simultaneously and moves in both directions.

Fully Semantic
burn tokens every time
Semi-Formal
LLM resolves structure
Fully Formal
rigid API, zero tokens

Leftward (toward semantic)

A formal protocol breaks. Instead of crashing, the system falls back up the spectrum. The LLM reads the error, understands the intent, finds another way. Degradation is graceful.

Rightward (toward formal)

A pattern repeats enough times. A convention agent identifies recurring structures and proposes semi-formal conventions. High adoption hardens the convention.

The key property

Formalization is reversible. Every formal contract has a semantic interpretation underneath. If the formal version breaks, the semantic version takes over. The LLM is the universal fallback.

Landscape Analysis

The team is voting
with their code

Eight overlap clusters where multiple people independently solved the same problem — demand signals for what the platform should provide natively.

6 Memory Systems

dev-memory, Lume, LifeOS, agent-memory, team-knowledge, GitNexus — all built store/retrieve/reflect independently.

4 M365 Connectors

connector-m365, teams-intelligence, openm365, lattice — same API surface, four implementations.

5 Schedulers

Loom, routines, worker-machine, dev-machine, self-driving — three independent scheduling systems that don't know about each other.

6 Orchestrators

Recipes, Attractor, project-orchestrator, longbuilder, workgraph, orchamp — sequential and graph-based models coexisting.

5 Code Quality Gates

nexus-dev, deliberate-dev, superpowers, metacog, foundation validators — same role, different philosophies.

8+ Convergence Loops

UI Studio, BookFoundry, Reality Check, recipe gates, nexus-dev, SOAR — all discovered propose/judge/refine independently.

10+ Browser Agents

operator, researcher, tester, debugger, documenter, audit, form-automation — differentiated by purpose.

7+ Content Pipelines

BookFoundry, Stories, nanoppt, paper-to-podcast, comics, deck-engine — all share the same multi-stage pipeline pattern.

The Other Path

Meanwhile,
independently…

In March 2025, Brian Krabach produced a comprehensive vision for the Cortex Platform: three documents totaling roughly 100KB of architectural specification. In January 2026, a concrete core design followed. This work was done without reference to the Semantic OS framework.

Bottom-Up

Sam Schillace · April 2026

  • Analyzed 442+ capabilities across 500+ repos
  • Identified 8 overlap clusters empirically
  • Named 10 unnamed concepts the team had already built
  • Derived architecture from demand signals
  • Emphasis on philosophy and systems theory

Top-Down

Brian Krabach · March 2025 – Jan 2026

  • Vision & Values, Technical Architecture, Day-in-the-Life narrative
  • Specified component interfaces and contracts
  • Designed JAKE memory system and Cognition System
  • Mapped Cortex Core onto Amplifier infrastructure
  • Emphasis on buildable component specs
The Convergence

Architecturally compatible conclusions

Two groups, working without knowledge of each other's framing, arrived at the same central organizing ideas.

Concept Cortex Semantic OS Match
Central orchestrator Cortex Core (6 subcomponents) Kernel (amplifier-core + amplifierd) High
Specialized agents Domain Expert Entities Expert Machines / Delegates High
Unified memory JAKE memory system Memory hierarchy (L1–L5) High
Communication Structured / NL / hybrid spectrum Progressive formalization High
Human escalation Guided Conversation Protocol Human Attention Scheduler High
Adaptive reasoning Cognition System (3 modules) Self-Model + Evolutionary Runtime Medium

The convergence is remarkable and should be taken seriously. When independent design efforts — one top-down from architectural vision, one bottom-up from ecosystem analysis — arrive at compatible conclusions, the team is circling something real.

— The Semantic OS, Section 10: Cortex Convergence
Complementary Strengths

Each provides what
the other lacks

Cortex adds specificity

  • Guided Conversation Protocol — six-step ASK_USER flow with priority, modality, timing
  • Workspace Model — projects/topics as organizational unit for memory
  • Cognition System — three named, swappable subcomponents
  • Domain Expert Contracts — four formal interfaces for agent integration
  • Day-in-the-Life narrative — morning voice to in-game AI avatar
  • Core as Amplifier Session — validates the kernel already exists

Semantic OS adds philosophy

  • Progressive Formalization — how interfaces emerge and how to grow without API gatekeeping
  • Self-Model Layer — live system awareness, convergence detection across projects
  • Attention as Objective Function — measurable: outcome value per unit of attention consumed
  • Elevation Ratchet — dynamic trajectory from operator to strategic director
  • Convention Agent — watches for patterns, proposes formalization
  • Demand-Signal Methodology — empirical evidence from ecosystem redundancy
The Unified Vision

Both buildable
and principled

Cortex provided the component architecture. The Semantic OS provided the systems philosophy. The merge produces something that is both.

Kernel

amplifier-core + amplifierd manages agent processes the way Unix manages computational processes.

Memory

JAKE interfaces at each tier. Workspace model for human-facing organization. Context managed like RAM.

Cognition

Reasoning, planning, self-reflection, and system model. Reasons about tasks and about its own architecture.

Experts

Domain Expert Entities with four-interface contracts. Progressive formalization for onboarding new agents.

Attention

Guided Conversation Protocol as Human Attention Scheduler. Track, batch, optimize cognitive load.

Evolution

Convention Agent watches for patterns. Adversarial Verification provides immune system. Elevation Ratchet trends upward.

The path forward is a single unified document that reads from the Semantic OS philosophy through to Cortex-level component specifications, grounded in the empirical evidence of what the team has already built. That document should be the foundation for everything we build next.

Sources & Methodology

How this story was built

Primary Documents

Cortex Platform Source Material

Repository

Contributors

Methodology

More Amplifier Stories