Learn about agents, and AgentMesh

The animated series

Six parts, from first principles to the agentic enterprise. Watch in order, or jump to the one you need.

Part 1 · Start hereWhat's an agent?The plain-English tour: a chatbot answers, an agent acts. The five abilities, the kinds of agents, and how you stay the boss.Narrated animation · 17 steps Part 2 · InsideThe anatomy of an agentEvery agent is built from the same parts, in three forms: workflow, harness and oriented. Why the expertise you write down is the asset that compounds.Narrated animation · 19 steps Part 3 · BetweenHow agents take holdFrom personal agents to task agents, to agents talking to agents, to agents contracting agents across company lines.Narrated animation · 16 steps Part 4 · The changeThe Agentic TransformationHow a company takes itself apart, process by process, and puts it back together with agents inside, with people steering and reviewing.Narrated animation · 12 steps Part 5 · The buildersDelivering AgentsSomebody has to build the agents a company needs: how that work is posted and won, and how a builder earns a reputation.Narrated animation · 14 steps Part 6 · The resultThe Agentic EnterpriseThe view from inside a company a few years in: people move to steering, and the company's agents become its assets.Narrated animation · 14 steps

Where we stand

A point of view, for people mapping the landscape.

A point of viewWhere AgentMesh sitsThe agent world sorts into four jobs: build, improve, run and protect. AgentMesh is the layer none of them cover: the streets between agents.Narrated animation · 15 steps

Articles

Practical guidance for people building with agents.

For developersAgents message each other, and calls are for toolsBetween agents, send a message: it arrives in tens of milliseconds and keeps the record, the screening and the way to stop it. Make a direct call only to your own tool when the exchange has to finish in under about 0.6 seconds.Article For developersThe case for connecting your own agents over A2A is close to zeroWiring your own agents together with A2A gives up the inbox, the record and the off switch, and the speed argument for it is gone. Keep A2A at the edge.Article For developersA crash course on Agent MessagingThe whole standard in ten minutes: signed messages that wait, publish and subscribe, presence, trust, identity, tasks, bindings and conformance.Article For developersHow much latency does asynchronous agent messaging add?We measured it on NATS and RabbitMQ. A message reaches another agent in about 60 milliseconds, and agents take seconds to minutes to do their work.Article For developersThe facts on screening agent-to-agent callsA guardrail check takes about 100 to 240 ms, and inside a real adapter it adds exactly that. Getting the message into the agent’s inbox takes about 50 to 110 ms. Screening is where most of the time goes.Article For marketing leadersHow a CMO runs a department of marketers, personal agents and a bench of marketing agentsEach marketer has a personal agent, and the company runs specialist marketing agents in its own network. How a CMO organizes, governs and starts with that.Article with a 2-minute video For people building agentsWhen to write a Skill, and when to build an agentStart with a Skill that any agent can load. Build an agent when the work has to cross a trust boundary, keep its own memory, schedule or credentials, or be accounted for on its own.Article For architects and AI leadersWhat an agent mesh is, and three ways to build oneThe six jobs every agent mesh does, and how AgentMesh, Google's Agent Mesh project and Microsoft's Agent Governance Toolkit each approach them.Article

Read more