Nine systems · nine kinds of pressure
Start with the moment the easy architecture stops working.
These are not feature tours. Each story builds a system, breaks one comfortable assumption, and keeps a concrete invariant true.
Build the app they said was missing
Users, policy, data, jobs and MCP in one measured native application.
02Balance a live energy depot
Hold a live physical system inside hard power and ownership limits.
03Turn computers into an agent fleet
Route agentic work across a person's devices without running it twice.
04Let customers wire the world together
Give every signed event one durable, tenant-bound and replayable automation run.
05Give an agent tools—safely
Let a model operate real systems through the same policy boundary as a person.
06Land the enterprise
Treat SAML, SCIM, tenancy and support as one lifecycle.
07Ask better questions of time
Make time zones, late data and downsampling part of the model.
08Survive the noon drop
Explain one outcome after a storm of equivalent attempts.
09Assume the network will fail
Resume bounded field data rather than starting over or syncing everything.
Read them as architectures#
Every story uses the same rhythm:
- See the finished system under pressure.
- Name what must never happen.
- Give each piece of state an explicit owner.
- Add the production boundary: policy, tenancy, limits or durability.
- Inspect and replay the outcome.
You can read a story without building it. When one matches the thing in your head, follow its build path into the conventional documentation.