BOOT CONSTELLATION

A.005 · 04.12

DELEGATE OK

DECOMPOSE OK

PERSIST OK

ORCHESTRATE OK

SPECIALIZE OK

TRAIN OK

Specialized Subagent Architecture

Generic delegation → context-aware multi-agent coordination. How to decompose, delegate, persist, and train specialized agents.

READ →

A.001 · 05.04

STACK-LB

THE SHAPE
OF A WORKING
MAS

Eight parts on context, write authority, orchestration, protocols, control plane — and a reference architecture you can build today. Plus six staked bets for May 2026 → May 2027.

SECTION INDEX · 12

CONTEXT

AUTHORITY

ORCH

PROTOCOL

CONTROL

REFERENCE

EVAL

GUARD

MEMORY

OBSERVE

RECOVER

6 BETS

SIX BETS · MAY 26 → MAY 27

B1

B2

B3

B4

B5

B6

·

·

REFERENCE ARCH · TOPO BUS / CUBE LINK

READ →

RESEARCH SURFACE / 7 STREAMS → 3 SURFACES

A.003 · 05.12

CLAIM BEAD CHECKPOINT ANCHOR ×SURFACE

Routing the Routers

7 research streams of LLM model routing mapped onto one stack — with an off-policy evaluation playbook and an EU AI Act audit mapping.

READ →

AGENT INSTRUMENTS

A.002 · 04.15

FORK

MERGE

LATCH

PROBE

PRISM

FILTER

SPOOL

SCOUT

BEACON

PATCH

SEAL

QUARANTINE

Self-Evolving Skills

Skills became the npm of agentic AI in 8 weeks. A survey of the explosion — and what an optimal architecture looks like.

READ →

CONTEXT COMPACTION GRID / 128K WINDOW

A.004 · 04.13

LIVE INFLOW 240 CHOKE ×4.6 OUTFLOW 186 COMPACT 0.78 DROPS 412

Context Compaction for Long-Running Agents

Agent sessions degrade as context grows — not at the limit, but continuously. A research-grounded design synthesizing 40+ papers.

READ →

POSTS · 2026

05

SHIPPED

INDEX / 01 — 05

RTR

MAS

SKL

CTX

AGT

5 RESEARCH STREAMS · 1 ARC