Agents
In Agents24, an agent is not just a prompt. It is a runnable graph with explicit nodes, edges, inputs, and output behavior.
What matters first
- drafts can be incomplete and still saved
- runnability is decided at compile and execution time, not by forcing every draft to be perfect
- runtime values come from workflow input, state, or node outputs
Practical reading of the model
For the first implementation pass, think in this order:
- What enters the run?
- What decisions does the graph make?
- What outside capabilities can it call?
- What leaves the run?
What to avoid in the first build
- branching-heavy graphs
- broad state contracts
- multiple tools with overlapping responsibility
- large knowledge scopes you cannot validate manually