Blog

Notes from the
control plane.

On the economics of coding agents, token compression, model routing, and building neutral infrastructure for the agent era.

Topics All Economics Engineering Product Security
Economics · 8 min

Why agent token costs are the new cloud bill

Coding agents scale spend with productivity. Here's the anatomy of an agent invoice, why it grows so fast, and the three levers that bring it back under control.

Read
Engineering · 10 min

How compression + routing cut spend 60–90%

A technical walkthrough of the two techniques doing the heavy lifting in Cohesor — token compression and complexity-based routing — with the real per-token math.

Read
Product · coming soon

The MCP gateway: one front door for every tool

How we built a governed, audited broker for 50+ MCP servers — and why per-agent scopes matter more than the tool list.

In draft
Security · coming soon

Policy at line rate: governance in under 1ms

PII redaction, injection detection, budget caps and scope checks — all before a token moves. Here's how we keep it under a millisecond.

In draft
Engineering · coming soon

Lossless compression: same meaning, fewer tokens

The four techniques behind our 53% average token reduction — and why semantic similarity is the only metric that matters.

In draft
Economics · coming soon

The CFO's guide to agent spend

How to model, forecast and govern AI agent costs — with the exact metrics we'd put on a board deck.

In draft
Newsletter

Get the next one.

Occasional, technical, no fluff. Notes on the economics of agents, compression, routing and the control plane — straight to your inbox.