Agents should improve the tools that run the work.
The target workflow is for synthetic business teams to use the ten services, expose a real capability gap, and hand it to a coding agent that writes a focused change. A separate reviewer verifies the result. This page currently maps workflows and drafts local briefs; it does not host running agents.
Start with an invented business.
Use fictional details. The rehearsal creates an operating map; it does not create a real legal entity, license, customer, account, or transaction.
Business lifecycle route
Ten specialist workbenches
The service stations map business work to current tools and handoffs. The benchmark runner exercised one fictional operating day across all 10 services and 60 tools; other business blueprints are not yet benchmarked.
Business agents find the product gaps
Bot employees should operate the services against bounded fictional businesses, not pretend that every task is already covered. Each blocked workflow becomes a reproducible product issue with evidence and a checkable result.
- OperateEach service agent follows its business role and runs the available tools with synthetic records.
- RecordA supervisor saves the failed step, inputs, handoff, and expected result without customer data.
- BuildA coding agent claims a gap in an isolated branch, changes only the affected service, and adds a regression check.
- VerifyA separate reviewer checks the diff, runs the service workflow and shared safety checks, then prepares a preview.
Current status: no model, agent process, isolated code workspace, CI, or deployment runner is connected. This browser stores local gap briefs only; it cannot claim the services are being improved in the background.
Saved synthetic tool run
These are deterministic calculations previously run against fictional Desert Fern inputs. “Replay” displays the saved result; it does not execute a live tool or contact a service.