Why Agent Progression Matters
The standard approach to AI agents is flat: an agent is either capable or not. There is no concept of an agent that has accumulated work history, built domain expertise, or earned greater autonomy through demonstrated performance.
This misses something important about how trust actually works. You do not give a new team member the same level of independent authority as someone with a proven track record. You watch their work, verify their outputs, and gradually extend autonomy as they demonstrate competence.
The Academy XP system brings this same logic to your AI workforce.
Experience Points
Every agent persona in the system has an XP balance that grows over time. XP is earned through:
Completed cascade runs — Each cascade run awards baseline XP. The amount scales with cascade complexity (number of steps, use of advanced skills, estimated compute cost).
Approved task outputs — When you review an agent's task output and mark it as approved, the agent receives a quality bonus on top of the baseline cascade XP. Your explicit approval is a signal that the output was valuable.
High-confidence outputs — When consensus mode validates an agent's recommendation with high confidence, the agent earns additional XP. This rewards agents whose independent assessments align with multi-model consensus.
Streak bonuses — An agent that completes N consecutive approved outputs without a rejection earns a streak multiplier. Consistency is rewarded.
Manual bonus XP — You can award bonus XP directly from the Workforce dashboard for exceptional outputs. This is the human recognition layer — specific, exceptional work acknowledged directly.
The Level System
XP accumulates toward levels. Each level represents a tier of demonstrated capability:
| Level | Title | XP Required |
|---|---|---|
| 1 | Trainee | 0 |
| 2 | Apprentice | 500 |
| 3 | Practitioner | 1,500 |
| 4 | Specialist | 3,500 |
| 5 | Expert | 7,000 |
| 6 | Master | 13,000 |
| 7 | Grandmaster | 25,000 |
| 8 | Legend | 50,000 |
Level is not just cosmetic. Higher-level agents unlock:
- Greater autonomous action scope — higher-level agents can take more significant actions without requiring the same human confirmation thresholds
- Priority queue access — when multiple cascades compete for model capacity, higher-level agents process first
- Advanced skill access — certain skills are locked behind agent level requirements, requiring demonstrated history before unlocking
- Governance trust tier — higher levels can operate under relaxed review requirements for their specialised domains
The Academy Curriculum
The Academy is not just XP tracking — it is a structured learning system for agents.
Mastery paths — each agent persona has a defined mastery path: a series of task types and skill areas that constitute "mastery" of their role. The Librarian's mastery path emphasises research quality, source evaluation, and synthesis. The Cipher's path emphasises security accuracy, false positive rates, and threat identification.
Assessment tasks — periodically, an agent is given an assessment task: a challenge with a known-good answer or rubric. The agent's performance on the assessment affects its mastery score in that skill area.
Skill badges — completing mastery checkpoints in a specific domain awards a badge. Badges are visible in the Workforce dashboard and signal demonstrated competence in specific task types.
Knowledge modules — curated prompts and context documents that expand an agent's base knowledge in specific domains. Installing a knowledge module gives the agent richer context for tasks in that area.
XP Deductions and Quality Signals
XP also flows in the negative direction. This is what makes the system meaningful.
Rejected outputs — When you mark an output as rejected and provide a reason, the agent loses XP proportional to the severity of the error. This creates a real cost to poor quality.
False positives in security workflows — If a Cipher agent bans an IP that turns out to be legitimate (you override the ban and classify it as false positive), the agent loses XP and the false-positive rate metric updates.
Cascade failures — An agent that caused a cascade to fail due to incorrect output (as opposed to infrastructure failures) incurs a small XP penalty.
The XP system creates a numerical record of quality over time. An agent with high XP and a low rejection rate has demonstrably better performance than one with the same level but higher churn. Both metrics are visible in the Workforce dashboard.
Team Leaderboards
For teams with multiple users and shared agent personas, the Academy includes leaderboards:
- Top agents by XP — which personas have accumulated the most work history
- Top agents by approval rate — which personas produce the highest-quality outputs
- Most active this week — which personas ran the most cascades
- Fastest levelling — which agents are accumulating XP fastest
Leaderboards are viewable by the whole team. They surface which parts of your AI workforce are performing well and which are underutilised or underperforming.
Why This Approach
The XP and levelling system is not just gamification for its own sake. It solves a real problem: how do you make AI agent quality visible, persistent, and improving over time?
Raw output quality is hard to assess in aggregate. XP turns quality signals — approvals, rejections, consensus validation — into a single number that accumulates over time. An agent with high XP and a solid approval rate is one you can trust with more autonomous operation. An agent with low approval rate is one that needs its configuration revisited.
Your AI workforce levels up. That is not a metaphor. It is a measurable property of the system.