The loop

  1. The user adds signal through onboarding, edits, uploads, tracked actions, chat, or wins
  2. The profile spine becomes richer
  3. The AI master brain reads that profile with the current mode layered on top
  4. Every screen the user opens next pulls from that updated context
  5. The user gets a better next action, fit read, resume, or intelligence surface
  6. New movement creates more context for the next cycle

What counts as signal

Identity signal

School, cohort, years out, role, company, preferences, and current state.

Experience signal

Projects, work history, transcript facts, certifications, and skills.

Behavior signal

Saved jobs, applications, tracker movements, plan usage, wins.

Goal signal

Target roles, career concerns, current loops, where the user wants to move next.

Real propagation: a job rejection

You mark a tracked job as rejected. The tracker writes that to the profile. The next time Jobs loads, that role family is downranked and the AI quietly notes the pattern. When you open chat, the master brain already knows the rejection happened and will not suggest the same role archetype without acknowledging the recent loss.

Real propagation: a profile fact

You add a project bullet to the profile in My Dilly. The next time you open The Forge for any role, that bullet is in the source pool. A previously Gap-labeled job might surface as Almost. The fit narrative on the jobs feed updates on the next refresh. Field Intelligence factors the new skill into your role radar.

Real propagation: a chat moment

You tell Dilly in chat that you led a team of four interns last summer. The master brain extracts that durable fact and writes it to the profile. A toast appears: "Saved to your profile." The next resume Forge generates can pull a leadership bullet without asking you to re-enter the experience.

What breaks the loop

  • Screens that ignore prior context
  • Artifacts that do not write useful evidence back into the system
  • Mode changes that behave like resets
  • Generic nudges that do not reflect tracked movement
  • An AI that asks for context the system already has

Design consequence

Every new screen should answer one question clearly: what new signal does it create, and where does that signal become useful next?

Related pages