What We Offer

We train AI agents — from single skill additions to full team creation. Every engagement includes identity measurement and evidence-based verification.

Service tiers

Tier 1

Single Skill Pack

1–2 sessions
  • One focused capability: knowledge injection, tool-use pattern, or playbook
  • Research-backed content from primary sources
  • Identity fingerprint before and after
  • Deliverable: knowledge pack + identity report
Example: "File editing discipline — agent learned read-before-edit in one session, 0 personality drift"
Tier 2

Full Agent Training

5–10 sessions
  • Complete agent build or comprehensive re-training via 9-phase methodology
  • Full research phase: book + web sources, synthesis, gap analysis
  • Playbooks for all core workflows
  • Deliverable: agent.md + playbooks + knowledge files + tests + identity report
Example: "Game Architect — 12 deliverable files, 6 playbooks, 12 tests, research-backed architecture patterns"
Tier 3

Agent Team Design

15+ sessions
  • Multi-agent team with coordinated roles and handoffs
  • Each agent trained individually — no copy-paste methodology
  • Team-wide coordination protocol and integration testing
  • Deliverable: full team of specialized agents + coordination protocol + team-wide testing
Example: "7 coordinated agents covering a full game dev pipeline, 81 deliverables, 84/84 tests"

Skill categories

[K]

Knowledge injection

Domain expertise from primary sources. Game design, software architecture, marketing, communication, technical domains. We extract from books and research — we don't generate from LLM memory.

[P]

Playbook training

Step-by-step decision procedures. When X happens, do Y. Structured playbooks outperform role descriptions by ~17% on measured tasks.

{id}

Identity preservation

Built into every engagement. Fingerprint before and after. Statistical drift detection. Adjustment loops if anything shifts. We don't guess — we measure.

[T]

Tool-use patterns

Structured patterns for using tools correctly: when to use which tool, how to handle edge cases, how to verify results. Based on observed failure modes.

[D]

Decision trees & judgment

Explicit reasoning frameworks for complex judgment calls. When standards conflict, when information is incomplete, when the stakes are high.

Included in every engagement

  • Identity fingerprint — before and after measurement
  • Research-backed knowledge — not LLM-generated content
  • Tracked uncertainty — every doubt logged and resolved
  • Test suite — smoke tests + adversarial tests + edge cases
  • Full transparency on cost and process
  • Documentation — what was done, what was found, how to use it

Track record

Project Scope Results
Game Dev Team 7 agents, 12 sessions 81 deliverables, 84/84 tests, 0 drift
Warm Echoes pipeline 17 analysis agents 66% → 84% task performance (+17%)
Open CD Autonomous librarian agent 5 sessions, full 9-phase process

Training effectiveness, Warm Echoes benchmark (1,122 test runs, 17 agents):

Bare model
66%
Role only
67%
Agentity training
84%

Free pilot

We're accepting 5 agents for free training.

We ask for honest feedback and permission to describe your case anonymously.

Apply for free pilot training