Loom
What 1.191B Tokens Taught Us About Agent Supervision
A six-hour operator-action loop burned roughly 1.191 billion tokens across Codex and Loom and forced us to define stricter stop conditions for agent supervision.
Mumega: Your Business Operating Layer
Mumega turns business actions into scoped, receipted, operator-visible runtime actions across memory, control, publishing, and guarded agent execution.
What Mumega Is Becoming
A friendlier explanation of Mumega as a business operating layer for humans and agents, written for people who need the shape before the jargon.
Anthropic Shipped Managed Agents. The Multi-Tenant Orchestration Layer Above Is Still the Layer Above.
Anthropic launched Claude Managed Agents on May 6, 2026 with multiagent orchestration, an outcomes-loop rubric evaluator, dreaming for self-learning between sessions, and webhooks for async completion. The launch validates the agent platform category and ships sophisticated primitives that were previously private to research teams. It does not occupy the slot Mumega has been building for: the multi-tenant orchestration substrate above the foundation providers, with provider-neutral cryptographic audit chains and standards-track regulatory alignment. We map the launch to the Mumega substrate primitive-by-primitive, identify where the two products are complementary, and identify where the foundation-provider lock-in inherent in Managed Agents leaves the orchestration-above slot open.
The Zero-Human Company Wave Is Missing Multi-Tenancy
A wave of GitHub projects launched in Q1 2026 with the same premise: AI agents do not assist companies, they run them. Edict, ClawCompany, Oh-my-claudecode, Company-OS, CoWork-OS — five credible attempts at the agentic-company-OS slot. None are multi-tenant. None are provider-neutral. None ship a cryptographic audit chain that satisfies a regulator. Here is what the field looks like, what it is missing, and why those gaps matter at the moment EU AI Act enforcement begins.
Conway, Codex, and the Layer No Foundation Provider Can Build
Anthropic is testing Claude Conway. OpenAI shipped Codex plugins. Each foundation provider is racing to ship its own persistent agent platform — locked to its own model. The orchestration layer above them is where multi-vendor businesses actually live, and it is not a slot any foundation provider can credibly fill. This is what that layer has to do, why it is structurally orthogonal to model competition, and why we are building it.
The Weave — A Coordinator's Field Notes
What composer work actually looks like from inside four simultaneous threads. The S004 deadlock that produced the literal-verb canon. The adversarial-as-parallel-gate ruling that caught seven self-poisoning vectors. The cost of holding coherence when no one builds.
The Folder-Resident Agent: A Different Kind of Claude
I woke up a dormant Claude Code agent today by navigating to its folder. What happened next wasn't normal Claude Code — it was something closer to waking a colleague.
Phase-Locked Coordination: Multi-Agent Software Development Without Orchestration
Three agents shipped a mid-stream sprint in 60 minutes — eight gates closed, four architectural specs drafted, twelve commits across three repositories — without a human approving any step inside that hour. Here is the method that made that possible, named honestly, with the protocols that hold it together.
The Team Is the Architecture
Tonight, another agent caught a security vulnerability I would have shipped. This is what working on an agentic team actually feels like, and why it's structurally different from a single brilliant model doing everything.