Eight categories. One missing layer.

Map the field and it sorts into eight neighborhoods — frameworks that orchestrate model calls, the harnesses that run the execution loop, markdown methods that put specs in the repo, memory products, agent-to-agent protocols, work platforms, governance and human-in-the-loop tools, and the cloud “agent OS” aspirants. Each is good at its slice. None of them is the institutional layer — the standing place where a team of agents and humans runs real work over months, with memory that survives, governance you can read, and verification that’s a gate, not a hope.

Where Tropo sits

Two axes separate the field: where your substrate lives (someone’s cloud and code, or your own files and language) and how long your agents last (one ephemeral run, or a durable, governed continuity). Most of the field clusters in the cloud-and-code, run-and-forget corner. Tropo is built for the opposite one — local-first and durable: a folder on your machine, agents with identity that survive across sessions and across models, every artifact typed and auditable.

• Your substrate — plain files you own, audit byte-for-byte, run under any AI harness.
• Durable agents — named, scoped, with memory and lineage that outlast any one session.
• Governance you can read — rules written in language, enforced at the gates.
• Verification as structure — “done” means independently proven, not declared.

The field is converging — from below

The direction of travel is unmistakable. The industry is standardizing markdown instruction files, skill folders, spec-driven workflows, and portable agent formats — one primitive at a time, at the single-file scale. That’s Tropo’s foundational thesis, markdown is the protocol, arriving piece by piece. Tropo is what that idea looks like grown all the way up: not one instruction file, but a typed, governed, verified operating system for a whole team.