The question
Online-store work is fragmented across catalog, storefront, order, support, and growth tools. A merchant repeatedly transfers context between those tools and remains responsible for every operational decision.
FaStart is researching a narrower question: which repeated workflow can an agent safely own while the merchant keeps authority?
Current product evidence
The current FaStart recording demonstrates product-draft setup and AI image generation in a working prototype interface. It does not demonstrate production availability, active merchants, storefront publishing, payment processing, orders, or fulfillment.
That boundary matters. A convincing interface is not the same as a reliable operating system.
Reliability before autonomy
An agent workflow should graduate only when it has:
- a defined permission boundary;
- visible inputs and proposed actions;
- merchant review for consequential changes;
- an audit trail that explains what changed;
- a fallback or rollback path;
- evaluations that can detect regressions.
The research focus is therefore not a catalogue of named agents. It is the control system around one bounded action.
Current status
Research programme with an in-development product prototype. FaStart does not claim unattended production operation.
The next proof milestone is a connected workflow that records the proposal, merchant approval, outcome, and recovery path from end to end.
Compare notes
If you work on agent reliability, evaluation, permissions, or LLM observability in real operational systems, contact research@fastart.tech.