The runtime decides
It runs continuously, choosing the next action from what was observed and the criteria written in advance. It does not write to a channel or a supplier, and it records what it chose along with why.
What is a commerce agent runtime
A commerce agent runtime is an execution environment that reads sales and supply data continuously and chooses the next action.It runs all the time, and it never writes outside itself.
Deciding and writing.
The two layers do not overlap. What the two layers actually do is in sourcing automation.
The system writes
Decisions become listings, collected orders and purchase orders. Permissions and limits are checked first, and afterwards the result is read back from the official state at the channel and the supplier.
People set the boundary
Changing the criteria themselves, and the line around actions that are hard to undo, stay with people. So do exceptions that have not been reduced to rules.
Five stages and who owns them.
ObserveSystem
Demand, competition, price and stock are collected continuously from channels and suppliers.
DecideRuntime
The next action follows from what was observed and the criteria set in advance.
ExecuteSystem
Listings, orders and purchase orders go out only after permissions and limits are checked.
VerifySystem
Every result is read back from the official state at the channel and the supplier.
LearnRuntime
Reasoning and outcome are kept together, so recurring exceptions become rules.
Separated, it can grow without loosening.
One place holds the write permission
When every external write funnels through the system, limits and kill switches sit in one place. More decision-making does not mean more points of control to maintain.
Changing judgement leaves execution alone
Because deciding and writing are separate, how decisions are made can change while the verified execution path stays exactly as it is.
What was done, and why, survives
Decision and execution are recorded separately, so a bad outcome can be attributed to one or the other. The fix starts at the step that failed instead of rerunning everything.
On the agent runtime
What is a commerce agent runtime?
A commerce agent runtime is an execution environment that reads sales and supply data continuously and chooses the next action. Unlike a tool that answers once when prompted, it runs all the time and hands what it decided to an execution layer.
Does the runtime list products or place orders itself?
No. Writing outside is done by the commerce system, which holds the permissions and limits. The runtime goes as far as deciding what should happen and recording the basis for it. Separating the deciding side from the writing side is the point of the arrangement.
Why not combine them?
Because external writes need to funnel through one place for limits and kill switches to mean anything. How decisions get made keeps changing; the execution path is better left verified and still.
How is this different from other AI commerce tools?
Tools that draft copy or gather product listings hand a person the output to carry to the next step. Here the decision continues into execution, and the result is confirmed against the official state before it returns to the criteria.
What do people do?
They set the criteria, draw the line around actions that are hard to undo, and judge exceptions that have not been reduced to rules. Those judgements are stored with their reasoning and outcome, so repeats are handled as rules.
Contact
contact@deepeet.com