Founders' note
Coming first
Why we're building Seam: the operational layer agents are missing
Models keep getting better. Deployments keep getting harder. The problem isn't reasoning — it's coordination, context, identity, and audit. Here's what we've learned from talking to fifty enterprises trying to put agents into production, and why it convinced us to build Seam.
Coming soon →
EngineeringQ3
Designing for replay: how every Seam session becomes auditable by default
Replayability is a property of architecture, not logging. We walk through the data model that makes every agent decision reconstructable months later.
In draft
ProtocolsQ3
MACP, CTXP, AITP — three protocols that make multi-agent systems behave
An overview of the open protocols Seam speaks: what each one solves, why we made them open, and how they fit together.
In draft
Field reportQ4
A loan-origination workflow, six months in: what changed, what broke, what got better
An anonymised field report from one of our first finance design partners. The boring parts of running agents in regulated environments — and how the boring parts are the whole game.
Drafting with partner
Short noteQ4
The bandit at the edge: how Seam learns from session-level signals without retraining a model
A short engineering note on the contextual-bandit layer behind Seam's policy adaptation, and what it lets you tune without touching the underlying agent.
In draft