Why Piyaz for agent teams

Coding agents are good at the work and bad at the memory. Two failure modes show up the moment you run them on a real project.

Agents lose context between sessions

A fresh agent starts cold. It does not know what the last session built, which approach was tried and abandoned, or why a file is structured the way it is. Without that history it re-discovers decisions, duplicates work, or contradicts choices made an hour ago. The longer the project runs, the more the gap costs.

Parallel agents collide

Run two agents at once and they edit the same files, claim the same task, or ship changes that assume the other's work does not exist. Speed turns into merge conflicts and wasted rotations.

What Piyaz provides

Piyaz is a shared task graph that agents and humans both read and write. It carries the state that a single agent's context window cannot:

• A shared task graph. Every task is a node with a status, acceptance criteria, and a record of what happened. Agents pick up where humans left off, and humans pick up where agents stopped. See Task graph.
• Execution records. When a task lands, the agent writes a short record of what it built: the files, the functions, the decisions. The next agent reads that record instead of guessing. See Execution records.
• Dependency edges that carry agent-to-agent briefs. A depends_on edge is not just an ordering constraint. Its note tells the downstream task's agent what specifically it gets from the upstream task. The brief travels with the graph, so the handoff happens without a human in the middle.

Because the graph holds the history, the agent's context window does not have to. Each session, and each parallel worker, reads exactly the slice it needs and writes its result back where the next worker will find it.

This section