How It Learns

The system doesn’t just do what you tell it — it gets better at doing it over time. Here’s how.

The Learning Loop

Analogy: Think of a sports coach. The players play the game, the coach reviews the footage, identifies what worked and what didn’t, and adjusts the game plan. Over a season, the team keeps improving. Your agent system works the same way.


  You use ──▶ System tracks ──▶ Patterns ──▶ Context is ──▶ Better performance
  the system   what happened     detected     updated        next time
     │                                                                       │
     └───────────────────────────────────────────────────────────────────────┘

What the System Tracks

Every time you interact with the system, it learns from:

Signal	What It Reveals
What you approve	What good output looks like for you
What you edit	Where the system’s defaults don’t match your style
What you reject	What NOT to do
What you ask repeatedly	Missing information in your profile
What takes too long	Workflows that need optimization
What you stop using	Features that don’t provide value

Three Types of Learning

1. Profile Learning (Fastest)

When you correct the system — “I actually prefer bullet points, not paragraphs” — that correction can be saved to your profile permanently. Next time, the system remembers.


  ┌─────────────────────┐       ┌─────────────────────┐       ┌─────────────────────┐
  │ You correct         │       │ Update saved        │       │ All future          │
  │ the system          │──────▶│ to profile          │──────▶│ interactions use    │
  │                     │       │                     │       │ the correction      │
  └─────────────────────┘       └─────────────────────┘       └─────────────────────┘

How it happens:

You tell the system to update your preferences
Or the nightly pipeline detects repeated corrections and proposes updates
Changes go into your profile files and take effect immediately

Example: You notice email drafts are too formal. You say “Make my emails more casual and use shorter sentences.” This gets saved, and ALL future email drafts match your real voice.

2. Workflow Learning (Weekly)

As you use workflows repeatedly, the system notices patterns:

Pattern	Learning
”You always edit the meeting prep to add competitor info”	Add competitor analysis to the meeting prep workflow
”You skip section 3 of every weekly review”	Remove or shorten section 3
”You add personal notes to every EOD summary”	Add a personal notes prompt to the workflow
”Email triage accuracy improved after adding VIP list”	Confirm VIP list is working, expand it

These improvements accumulate. A workflow you use daily gets meaningfully better every week.

3. Pattern Learning (Monthly)

Over longer periods, the system detects bigger patterns:

“Your Tuesday mornings are consistently your most productive time. Your energy drops after 3 PM. You do your best creative work on Thursdays.”

“You’ve been spending 40% more time on email this month. Most of the increase is from one client. Consider setting up auto-filters for their routine updates.”

“Three of your goals haven’t had any activity in 6 weeks. Are they still priorities, or should they be archived?”

These insights help you make strategic adjustments, not just tactical ones.

The Measurement System

Learning requires measurement. The system tracks metrics to know if things are improving:

What Gets Measured


  ┌─────────────────────┐
  │ Activity Metrics    │───┐
  └─────────────────────┘   │
  ┌─────────────────────┐   │   ┌─────────────────────┐
  │ Quality Metrics     │───│──▶│ Dashboard           │
  └─────────────────────┘   │   └─────────────────────┘
  ┌─────────────────────┐   │
  │ Efficiency Metrics  │───│
  └─────────────────────┘   │
  ┌─────────────────────┐   │
  │ Domain Metrics      │───┘
  └─────────────────────┘

Category	Example Metrics
Activity	Tasks completed, workflows run, emails handled
Quality	How often you approve drafts without edits, error frequency
Efficiency	Time saved vs. manual process, automation rate
Domain-specific	Email response time, meeting prep quality, follow-up completion rate

How Metrics Drive Improvement

Metrics alone are just numbers. The value comes from what they reveal:

Metric	What It Tells You
”Email triage took 45 min this week (up from 30)“	Something changed — new subscriptions? More volume?
”You edited 80% of email drafts”	Draft quality needs work — time to update your profile
”Weekly review takes 15 min (down from 90)“	Automation is working — the system is saving you time
”3 follow-ups were missed this month”	Follow-up tracking needs improvement

Key principle: Metrics should drive decisions. If you can’t act on a metric, don’t track it. Focus on the numbers that help you improve the system.

The Improvement Timeline

Timeframe	What Improves	How
Day 1-7	Basic personalization	Profile corrections, style adjustments
Week 2-4	Workflow quality	Repeated use reveals gaps, you refine
Month 2-3	Patterns emerge	Cross-workflow insights, strategic suggestions
Quarter+	Compound improvements	System deeply understands your patterns

The system follows a compound improvement curve — early gains are obvious (better email drafts, faster triage), and over time the improvements become more sophisticated (strategic insights, pattern detection, proactive suggestions).

What You Can Do to Help

The system learns faster when you:

Correct it in the moment — “That’s too formal” is better than silently editing
Update your profile — When preferences change, tell the system
Give explicit feedback — “That weekly review was exactly what I needed” or “This format doesn’t work for me”
Be consistent — Use the system regularly so it has data to learn from
Review the metrics — Check your weekly review for trends and act on them

Learn More

The 8 Primitives — See how feedback and metrics fit into the full system
Weekly Reviews — See metrics and patterns in your weekly review
How Context Works — How updated context flows through the system