The Reference Framework
This document consolidates the maturity model, the operating principle, and the two scales that structure the AI transformation.
The Universal Translation Rule
The operating principle of the entire transformation fits in one sentence:
Replace "the human produces the artifact" with "the human defines the spec → the system produces the artifact."
What This Means by Department
The Litmus Test
If this person disappeared, could a system execute 80% of their tasks?
- If no → the role is still execution-based
- If yes → the role is AI-native
This isn't "AI adoption." It's the shift from a labor-based company to a systems-based company.
Organizational Scale — Levels 1 to 3
This scale applies across the entire group — engineering, marketing, sales, finance, customer service.
AI-Assisted
What it looks like:
- AI is a tool that individuals choose to use
- Same structures, same processes, same roles
- If AI disappeared tomorrow, nothing structural would change
Typical behaviors:
- Using ChatGPT/Claude like Google or a spell checker
- Isolated prompts, no iteration
- AI outputs manually pasted into work
- No shared prompts, no documentation
- Adoption is uneven and optional
The gap is measurable: in technical roles, AI has 94% theoretical task coverage but only 33% actual usage. Level 1 organizations leave most of AI's capability untouched.
AI-Integrated
What it looks like:
- AI is integrated into workflows and systems
- Some processes redesigned around AI capabilities
- Roles start shifting from "doing" to "directing" (see role evolution patterns)
- If AI disappeared tomorrow, some workflows would break
Typical behaviors:
- Saved prompts, templates, prompt libraries
- AI used across multiple steps of a task, not just one
- Tools like Copilot, Notion AI, Zapier, n8n in active use
- Prompts and workflows shared among colleagues
- AI usage is expected, not optional
AI-Native
What it looks like:
- Organizational design assumes AI as a first-class resource
- Roles are defined by judgment and direction, not execution
- Headcount is a fraction of a traditional company at the same output
- If AI disappeared tomorrow, the company couldn't function
Typical behaviors:
- The starting question is: "What part should be automated?"
- Agents, pipelines, and decision systems built (code or no-code)
- Processes designed so humans handle judgment, AI handles execution
- AI impact is measured (time saved, costs reduced, quality improved)
- AI literacy is a condition of employment
Engineering Scale — Rungs 0 to 5
Engineering needs finer granularity. Based on Dan Shapiro's framework, this scale describes the progression of software development. The AI Lab details it and how it operates.
| Rung | Human's role | Who writes the code | Who reviews the code |
|---|---|---|---|
| 0 — Assisted coding | Human codes, AI suggests | Human | Human |
| 1 — Scoped delegation | Human assigns scoped tasks | AI | Human (everything) |
| 2 — Supervised generation | Human supervises multi-file changes | AI | Human (everything) |
| 3 — Directed development | Human directs, reviews at feature/PR level | AI | Human (PR) |
| 4 — Spec-driven development | Human writes the spec, verifies results | AI | Nobody (tests verify) |
| 5 — Autonomous production | Spec goes in, software comes out | AI | Agent reviewer + scenarios; risk-graded human gates |
The achievable Rung on a given codebase is capped by codebase readiness — a team at Rung 5 on a codebase at Readiness Level 2 will produce fast hallucinated output with no correction signal.
What Rung 5 actually looks like
Rung 5 is not monolithic. Validation is risk-graded by an action's blast radius, reversibility, and consequence:
Human approves before execution.
Default for irreversible high-impact actions: financial transactions, production deploys, customer-facing changes.
Agent acts autonomously; human monitors with intervention authority.
Default for reversible production work with eval coverage.
Agent acts within pre-defined boundaries; no real-time human involvement.
Reserved for sandboxed, reversible work with strong tests and an agent reviewer on every PR.
A Rung 5 team operates all three concurrently, picking the gate per action class. Each gate runs the same operational unit — Context → Clarification → Execution → Validation → Recovery — described in detail in the AI Lab.
Beyond engineering
The engineering scale is the most documented instance of a broader pattern. The Rung 5 operational reality — recurring loop, risk-graded validation, agent-as-reviewer — generalizes to other discrete-task domains (engineering, customer service, finance operations, legal review, knowledge research) where AI similarly operates as the execution layer. Parallel readiness models for those domains are on the framework's roadmap. See also the depth note in the Skill Progression Map for how this asymmetry shows up role-by-role.
Mapping
| Organizational scale | Engineering scale |
|---|---|
| Level 1 — AI-Assisted | Rungs 0-1 |
| Level 2 — AI-Integrated | Rungs 2-3 |
| Level 3 — AI-Native | Rungs 4-5 |
Diagnostic Questions
Three quick tests to gauge maturity. For a full team-by-team assessment methodology, see Assessing Your Organization.
"If AI disappeared tomorrow, what would change?"
- Nothing structural → Level 1
- Some workflows break → Level 2
- The company can't function → Level 3
For acceptance criteria per level and the transformation path — timelines, prerequisites, and what to watch for — see the Implementation Roadmap.
Leadership Tiers
The company can't exceed the tier of its leadership. Leadership is the ceiling.
Publicly endorses AI. Uses it personally. Doesn't push adoption.
Sets expectations by role. Asks "how did AI help?". Funds automation before hiring.
Redesigns the organizational structure. Rewrites roles and KPIs. Makes AI literacy a condition of leadership.
Individual Tiers
The individual tier scale measures where a person is in their AI transition. The three main tiers (T1, T2, T3) describe established states. The intermediate tiers (T0.5, T1.5, T2.5) describe the transitions between them – proof that someone is moving, not just labeled.
AI is not part of work in any form.
Has tried AI but it hasn't changed how work gets done.
"AI helps me do my job faster."
Actively designing and testing AI workflows. The construction phase.
"AI helps us do this task better and more systematically."
Building systems where AI handles most execution. The role is transforming from the inside.
"This role should exist differently because AI exists."
The three main tiers are operational states – they describe how someone works. The .5 tiers are transitional – they describe that someone is actively moving between states.
Tier 0 / 0.5 – The gap between T0 and T1 isn't knowledge, it's the habit of reaching for AI when work starts. A person at T0.5 has experimented but hasn't integrated. The risk: people stay here indefinitely without structured nudges.
Tier 1.5 – This is where most people stall. They're past ad-hoc usage and actively building workflows – designing prompts, testing systems, iterating. Some experiments fail, some stick. The transition from T1 to T2 requires that these experiments become established workflows, not just personal experiments. Context files and shared prompt libraries become critical at this stage.
Tier 2.5 – Multiple processes redesigned around AI. The person spends most of their time on direction, judgment, and review rather than execution. Starting to eliminate coordination overhead – doing alone what used to require team handoffs. The distinction from T3: the role still has the same title and boundaries, but the work inside has fundamentally changed. These people demonstrate what Level 3 looks like before organizational structure catches up.
Tier 3 – AI is the execution layer; humans give direction and validate. Work is structured around a recurring operational unit (Context → Clarification → Execution → Validation → Recovery), with cycle times collapsed to days at the feature level rather than weeks at the sprint level. The person's value concentrates at the boundaries: front-boundary work (specification, alignment, clarification dialogues with the AI) and back-boundary work (validation, edge-case sessions, first-user UX testing). Inside the loop, the agent runs. The discipline distinguishing T3 from T2.5 isn't more tool use — it's process design for AI: defining the constraints, gates, and validation tiers within which AI operates consistently. The role and the work inside it look fundamentally different from T2.5; titles may not have changed but the day-to-day has.
Mapping to organizational levels
| Individual tier | Organizational level |
|---|---|
| T0 / T0.5 | Below Level 1 (awareness without integration) |
| T1 | Level 1 (AI-Assisted) |
| T1.5 | Transition from Level 1 to Level 2 |
| T2 | Level 2 (AI-Integrated) |
| T2.5 | Transition from Level 2 to Level 3 |
| T3 | Level 3 (AI-Native) |
← Back to home · Implementation Roadmap · AI Execution Standards · The AI Lab · Glossary