Run the Automated Design
The automated design loop is the heart of the product: it generates an architecture from your Design Inputs and the approved technology catalog, analyzes it against your controls, remediates what it can, and pauses only when a decision genuinely needs you. This page walks the whole loop, including what to do when it asks for decisions.
Before You Start
- The package has its inputs filled in on the Design Inputs tab: at minimum a summary and a few functional requirements. Data classes and regulations make the result meaningfully better.
- You have a role that can run designs (architect or above).
- The org's technology catalog (Settings → Technology catalog) has approved entries. An empty catalog is fine for a first run; the loop will surface every capability as a decision instead.
Start a Run
- Open the package and go to the Architecture tab.
- In the Automated Design panel, choose Run Automated Design (first run) or Re-run Design (later runs).
- Watch the status move through Queued → Designing → Analyzing (and Remediating when the loop fixes failing controls itself). Runs take seconds on the mock provider and a minute or two with a real model.
The run ends in one of two states:
- Design complete: a full design exists. The Design segment shows components, data flows, and trust zones; the Diagrams tab renders the C4 views; Assurance fills with compliance results.
- Needs your input: the loop paused because one or more capabilities have no approved technology. This is normal on early runs; it is the product asking you to govern, not failing.
When It Needs Your Input
The Decisions needed panel lists each open capability with ranked options: capability fit, policy alignment, integration fit, trade-offs, and a chip when an option leverages an existing vendor contract.
- Press Choose on the option you want for each capability. One technology can fill several gaps; the button says Choose for all when it would.
- The header counts your selections ("3 of 6 selected, not applied yet"), and your picks survive navigating away and coming back.
- Press Apply N Decisions (or the main button, which becomes Apply & Re-run). Each chosen technology is approved into the catalog for that capability, and the design re-runs.
Type the product name into …or enter a different technology on the capability's card. Applying approves it into the catalog exactly like a recommended option.
If the decision is not yours to make today, create a deferral exception on the Assurance tab instead (Exceptions and Deferrals). An active exception covers the gap for approval purposes until its review date.
Decisions Stick
An applied decision is recorded state, not a suggestion to the model. On every later run, a capability covered by the catalog is selected deterministically, so the loop converges: one decision, one re-run, done.
Re-running Without Losing Anything
Re-runs are governed:
- Your edits survive. Components you edited or locked (Edit and Lock the Design) override the AI on every run.
- The boundary is stable. A re-run cannot silently add components: additions must cite the control, finding, or requirement that forces them, and anything uncited is quarantined for your accept-or-discard decision (Keep the Scope Stable). Check Preserve scope to tighten this further.
- People and externals never ask. Actors, external systems, and the legacy system a design is migrating off are context: they never appear as technology decisions.
After a Completed Run
| Look At | You Should See |
|---|---|
| Design segment | Component cards with technology, purpose, status, and leverage chips; flows with protocols and encryption posture |
| Diagrams tab | Context, Container, Component, and Deployment views regenerated from this design |
| Assurance tab | Compliance results per control, and the design review when you run it |
| Decisions & Value | The run's ADRs, indicative cost, estimated value, evidence, and exports |
Common follow-ups: run the design review on Assurance for the multi-agent critique, use Remediate Findings to feed current findings back into another pass, or head to Review & Approval when the readiness board is clear.
If the Run Fails
A failed run keeps the previous completed design; nothing is lost. The panel shows the error and Re-run Design retries. Runs interrupted by a worker restart fail with a re-run hint rather than hanging.