10 articles in this category.
An agent without a computer is a chat window. A team of agents without a team computer is a set of chat windows that cannot see each other. Both have a place, and settling for one costs more than it saves.
Most products say they are multiplayer the moment two people can see the same chat. The shape of the collaboration decides what breaks first, and what control the product needs.
Adding an assistant to a chat window does not make a workspace where humans and AI agents work together. Six design decisions do, and most of them are about control rather than capability.
Three very different products are sold under the same name. Only one of them is somewhere a team can actually work, and six questions tell you which is which.
AI can generate a hundred screens in a minute. Most of them are slop. Here is the working method I use to keep AI in the loop without letting it flatten the product.
Google Docs beat Word by going multiplayer. Figma beat Photoshop. AI has not had that moment yet, and the missing piece is not shared context. It is shared control.
How to design agent identity, delegation, authorization, and credentials without turning automation into invisible impersonation.
The engineering behind approval gates: suspension, evidence, expiry, revalidation, revocation, and safe continuation.
Human-in-the-loop design can be precise rather than performative. The checkpoint should appear at the moment risk changes, with enough context for a fast decision.
A principal architect’s guide to state, failures, retries, and consistency when humans and AI agents work together.
