AI-Native Transformation Framework

Changelog

A record of significant updates to the framework.

v3.1 — May 18, 2026

A content release adding two new sections to the framework: a Role Catalog describing 31 individual roles as they operate in an AI-native organization, and an organization-level view that complements the role-level perspective with structural patterns (functions, leverage math, hierarchy flattening).

New: The Role Catalog

/roles — 31 role pages, each describing what the role looks like at T2-T3 maturity in an AI-native organization. Each page answers the same set of questions: what does the role actually do, what does success look like, what makes it interesting (and what may not appeal), who thrives in it, what skills to develop to get there, how this differs from the legacy version of the role, which role evolution patterns are in play, and which adjacent roles in the catalog connect.

Coverage at launch. 31 roles across 9 families:

  • Product — Product Manager, VP Product
  • Engineering — Software Engineer, Full-Stack Engineer, Tech Lead, DevOps Engineer, Data Engineer, Engineering Manager, Director of Engineering
  • Design — Product Designer
  • Marketing — Marketing Strategist, Demand Gen Marketer, Product Marketing Manager, VP Marketing
  • Sales — SDR, Account Executive, Solutions Engineer, VP Sales
  • Customer Success — Customer Support Specialist, Customer Success Manager, Director of Customer Success
  • Operations & People — Data Analyst, Head of People Operations
  • Executive — CEO, COO, CTO
  • Emerging — Workflow Architect, Agent Supervisor, Specification Owner, Governance Specialist, AI Transformation Lead

Audience. For practitioners deciding whether they still want this work, managers planning their teams, and HR designing tomorrow's workforce.

Hub design. A card grid at /roles with sticky family filter chips. Each card surfaces the role's dominant role-evolution pattern (Elevation, Specialization, Convergence, Emergence) visually. Read together with the org-level pages below.

New: The AI-Native Organization (org-level view)

/ai-native-organization — the structural counterpart to the role catalog. Describes the five functions every AI-native organization needs (Direction, Specification, Validation, Agent Operations, Trust & Human Relations), the shape of the org chart, what must exist, what disappears or absorbs, and the transition picture from T0 to T3. Includes seven diagnostic questions for leaders.

Two companion pages go deeper on the most-asked questions from executives:

  • /the-leverage-math — how AI-native leverage is distributed across compression, expansion, or some mix. Three scenarios with concrete headcount math, per-function leverage ratios, and the strategic factors that drive the choice. Audience: CFO, board, capital strategy.
  • /the-flatter-hierarchy — which management layers compress and which survive. The five functions middle management performs and which absorb into agents; the math (50-person legacy ~22-25 managers vs AI-native ~11-15); what changes for surviving managers; transition risks. Audience: CHRO, Head of People Operations.

Operational additions

  • The five role evolution patterns (Convergence, Specialization, Elevation, Absorption, Emergence) from v2.3 are now anchored with concrete role-level examples across the catalog. Each role page names its dominant patterns and explains how they apply.
  • Hierarchy compression as a structural consequence of AI-native operations. The original framework mentioned the "diamond of thinkers" via the Vision page; v3.1 makes the operational specifics explicit in The Flatter Hierarchy.
  • The org-level Absorption pattern — extending the role-level Absorption pattern from Role Evolution to organizational scope: transactional roles, coordination layers, routine quality functions, and specialized hand-off roles compress or disappear; strategic direction, specification quality, risk-graded validation, agent operations, governance, and live human relations stay or grow.

UX additions

  • Breadcrumb navigation on individual role pages: Home > Role Catalog > Role. Includes BreadcrumbList JSON-LD for SEO.
  • Visual role-page hero with role-evolution pattern badges and compact metadata (family, equivalent legacy role, reports to, works alongside).
  • Related roles card grid at the bottom of each role page, replacing the previous flat bullet list of adjacent roles.

Status of translations

This release ships English-only. Translations to FR, ES, DE, IT, and PT are planned for v3.1.1, following the same parallel-agent workflow used for v3.0 translations. Non-English locales currently fall back to English content for the new pages.

v3.0 — May 10, 2026

A major release reframing the framework's operational model at maturity. v2.x described the journey toward AI-native; v3.0 specifies what AI-native looks like operationally — and acknowledges where the framework is engineering-deep today and where parallel readiness models for other domains will follow on the v3.x roadmap.

Strategic position

  • The framework is engineering-deep today. That asymmetry is acknowledged structurally (in the Skill Progression Map and the Reference Framework) rather than papered over.
  • The patterns introduced in v3.0 are role-agnostic discrete-task patterns, not engineering-specific patterns. They apply to engineering, customer service, finance operations, legal review, and knowledge research. The v3 era will build out parallel readiness models for those domains.
  • Continuous / creative / interpersonal roles (sales, marketing creative, design, HR) need a different framework shape — AI as augmentation, not execution. Deferred to a future v4+ track.

Content

  • The five-stage operational unit as the canonical Tier 3 / Rung 5 working pattern: Context → Clarification → Execution → Validation → Recovery. Triangulates across spec-kit, Anthropic plan mode, AMPECO CODA, Cognition Devin, multi-agent debate research, Cemri et al. (2025), and frontier practitioner accounts. Lands in AI Lab, Reference Framework, Specification Guide, Skill Progression Map, and Standards.
  • Risk-graded validation gates (HITL / HOTL / HOOTL) split a previously monolithic Rung 5 into three operational stances by an action's blast radius, reversibility, and consequence. Drawn from SAE J3016, Anthropic ASL, OpenAI Preparedness, NIST Agentic Profile, OWASP LLM Top 10, and the HITL/HOTL/HOOTL practitioner literature. Lands in AI Lab § Risk-Graded Validation Gates, Reference Framework, and Standards § Failure Responsibility Model.
  • Process design as Layer 5 of the AI Execution Standards. The discipline of designing constrained, phased workflows for AI to operate consistently within — distinct from prompt engineering and from spec-writing per se. Topology vocabulary (chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer, autonomous agents) and decision rules. Lands in Standards § Layer 5.
  • AI economics at maturity — cost per unit of output replaces "time saved by AI" as the binding metric at Level 3; AI gross margin runs structurally below SaaS (40–55% vs 70–90%); per-token costs fall while per-task costs often rise (Jevons paradox applied to inference). Lands in Business Case § AI economics at maturity and Implementation Roadmap § Measurement vocabulary by maturity.
  • Failure modes and recovery patterns at high maturity — sycophancy as a structural concern (treat as engineering problem regardless of training trajectory); subjective edge cases (failures surfaced by users, not tests; recovery is conversation, not patching); recalibration vs debugging as operationally distinct responses; AI-bottleneck failure mode at Tier 2.5+. Lands in Engineering for unreliability, AI Lab § Stuck-State Protocol, and Leading the Transformation § AI bottleneck.
  • Codebase Readiness D6 enriched to treat CLAUDE.md / AGENTS.md as living artifacts validated by the agent against the live code, with skills and subagents created on demand. Reflects current vendor practice (AGENTS.md spec, Anthropic Claude Code memory). Lands in Codebase Readiness.
  • Permissions Owner as a fourth named organizational role at production-grade AI systems. Accountable for what each agent can and cannot do, and for the validation gating tier per action class. Lands in Standards § Organizational Roles.
  • T3 / Rung 5 operational depth in the Skill Progression Map's Engineering column — unit of work, cycle time, day-shape vignette, AI-bottleneck failure mode, recalibration vs debugging, risk-graded validation, and metrics that replace "time saved." Lands in Skill Progression Map.
  • Tier 3 day-shape addendum in Transforming Your Role — what the day looks like once the transition arrives; cognitive demand at boundaries, not in execution.
  • Glossary — 16 new entries across four sections (Operational Reality at T3/R5, Risk-Graded Validation, Failure Modes and Recovery, AI Economics at Maturity).

Framework updates

  • Reference Framework: T3 row tightened to "AI is the execution layer; humans give direction and validate"; new prose paragraph on T3 day shape and process design as the distinguishing discipline. R5 row updated from "Nobody (scenarios verify)" to "Agent reviewer + scenarios; risk-graded human gates."
  • Skill Progression Map: New depth-asymmetry note up front naming the engineering tilt and the v3 era roadmap; bimodal productivity callout (L1–L2 spurious gains vs L3 4× gains for restructured teams).
  • AI Lab: Working Mode renamed and restructured around the five-stage unit; new sections for Risk-Graded Validation Gates and Stuck-State Protocol; token economics named as a Lab metric; AI-bottleneck pitfall added.
  • Engineering for Unreliability: New section "Failure Modes and Recovery at High Maturity" covering sycophancy, subjective edge cases, and recalibration vs debugging; agent-as-reviewer named as production default.
  • Specification Guide: New "The operational loop" section with clarification dialogue as a discrete stage; spec-writing described as a spectrum (human-written → AI-assisted → AI-drafted-human-ratified) with proper hedging.
  • Standards: Pre-execution clarification allowed in Core Principle; Layer 5 (Process Design) added; failure responsibility model extended with Damaging output / Unnoticed output / Wrong-direction-defended-confidently rows; Permissions Owner role added.
  • Business Case: New section on AI economics at maturity (cost per unit, gross margin floor, Jevons paradox).
  • Implementation Roadmap: New section on measurement vocabulary by maturity — productivity-tier metrics at L1–L2 transition to unit-economics-tier metrics at L3.
  • Leading the Transformation: New diagnostic — AI bottleneck as the high-maturity failure mode; updated Workload Inflation to connect with bottleneck dynamics.

Rationale

The framework's v2.x releases described the journey to AI-native maturity from a Tier 1–2 vantage point. Practitioners operating at Tier 3 / Rung 5 — where AI is the execution layer rather than an assistance tool — found the descriptions of their actual operational reality too abstract. v3.0 specifies that reality: the operational unit, the validation gating structure, the discipline of process design, the economics of unit-cost, and the failure modes that emerge once the agent does the execution.

The reframing is honest about where the framework is engineering-deep today and explicit about the discrete-task domains where parallel readiness models will follow. The v3 era roadmap is internal at this release; it becomes a public commitment when v3.1 ships.

Acknowledgments

This release was prompted by detailed feedback from Vincent Lamanna (Crewdle) on the framework's self-assessment questionnaire. Vincent's critique surfaced the gap that v3.0 closes — the framework's descriptions of mature AI-native work didn't match the operational reality of his organization. The five-stage operational unit, risk-graded validation gates, AI-bottleneck failure mode, recalibration vs debugging vocabulary, and AI infrastructure economics all triangulate against multi-source literature, but Vincent's account is what made the gap visible. Thank you.

v2.3 — April 19, 2026

Content

  • Codebase Readiness for AI-Native Development: New framework page — a five-level readiness model (Opaque → Instrumented → Validated → Legible → Specified → Scenario-governed), each level defined by the feedback mechanism it adds. The Codebase Readiness Grid — a nine-dimension diagnostic (test coverage and feedback latency, type strictness, file size and context legibility, module boundary clarity, API directness, documented intent, observability, dev and deploy simplicity, dependency and runtime currency) with a 1–5 scoring rubric per dimension. Four scoring rules: ceiling = lowest score; three dimensions are blocking (D1, D2, D5) and six are constraining; intentional documented deferrals earn one-level credit; never summarize with an average — the Grid is a vector, not a scalar. Reference profile describing what a Level 5 codebase looks like. The hard rule: a codebase's readiness level is the ceiling on the engineering Rung that can operate reliably on it (read it)
  • Brownfield Engineering Strategy: New framework page — how to transition an existing codebase to AI-native development. Codebase-state triage (greenfield / brownfield / hybrid) before picking a mode. Four modes framed as paths to Level 5: remediate in place, strangler-fig migration, full rebuild, or isolate and bypass — each answering "where does Level 5 happen" rather than "what does the team do." "Do not invest" folded into Mode 4 as a legitimate outcome when remediation cost exceeds remaining value. The Research-Review-Rebuild methodology (Fowler/EPAM) with five Black Box to Blueprint reverse-engineering techniques. Spec-from-code as the brownfield inversion of spec-first. Economics anchor: Bahmni case study (~$2/component in under an hour vs. 3–6 days manually, with human review as the throughput constraint) (read it)
  • Companion tool — codebase-readiness skill: Open-source Claude Code skill that runs the nine-dimension assessment on any repo, applies scoring rules (ceiling, blocking/constraining, deferral credit), classifies codebase state, and recommends a path to Level 5. MIT licensed. Hosted at github.com/Kenogami-AI/codebase-readiness. Referenced from both framework pages via a new <ToolCallout> MDX component.

Framework updates

  • AI Lab: Brownfield admission criteria now frames the upstream decisions (readiness assessment, mode selection) as outside the Lab's scope, pointing to the two new pages.
  • Engineering for Unreliability: Mirror-framing paragraph linking codebase readiness as the inverse problem — same uncertainty, reflected.
  • Reference Framework: Added a Rung ceiling note — the achievable engineering Rung on a given codebase is capped by codebase readiness.
  • Assessing Your Organization: New common pitfall — skipping codebase readiness when assessing engineering teams. A team at Level 2 practice on a Level 0 codebase is effectively at Level 0.
  • AI Execution Standards and Specification Guide: Note added that brownfield codebases invert the spec-first model; specs must be reverse-engineered (spec-from-code) before new spec-first work resumes.
  • Glossary: Added Codebase Readiness section (7 entries) and Brownfield Strategy section (8 entries). Terms: codebase readiness levels, harness, ambient affordances, feedback loop topology, dependency and runtime currency, blocking dimensions, constraining dimensions, four brownfield modes, isolate and bypass, Research-Review-Rebuild, spec-from-code, strangler-fig migration, Technical Debt Quadrant, seam identification, Black Box to Blueprint techniques.
  • Navigation: New "Tools" dropdown (header + footer) listing the Transformation Platform and the Codebase Readiness skill. New <ToolCallout> MDX component for pointing to open-source tools from within framework pages.

Rationale

The framework covered the human side of AI transformation (cognitive cost, roles, leadership, contributor transition) and the engineering practice side (AI Lab Rungs 0–5, execution standards, specification engineering). What was missing: the codebase as an object of transformation. AI coding agents amplify whatever structure already exists — which means the single most consequential investment for brownfield teams is building the harness (tests, typing, module boundaries, documented intent) before deploying agents at scale. This release adds a diagnostic (codebase readiness), a strategic playbook (brownfield engineering strategy), and an open-source tool that implements both.

v2.2 — April 17, 2026

Content

  • The Cognitive Cost of AI Transformation: New framework page — the mental-energy cost of AI transformation is a structural constraint, as real as budget or headcount. Covers eight distinct challenges (cognitive overload, decision fatigue, vigilance fatigue, work intensification, workload inflation, AI anxiety, identity disruption, learned helplessness, transformation fatigue), the cognitive J-curve at Tier 1.5, and what leaders and individuals can do about it (read it)

Framework updates

  • Leading the Transformation: J-curve section expanded to describe the cognitive J-curve alongside productivity dip. Managing adoption speeds now flags cognitive overload as a common root cause of the "stuck" group. New subsections on workload inflation and the cognitive cost of transition. Pitfalls list extended with workload inflation, concurrent-tools cap, and T1.5 burnout pattern.
  • Transforming Your Role: Guiding principle now acknowledges that sustained judgment is cognitively demanding. Layer 4 (Implementation) expanded with a cognitive J-curve subsection — what to watch for, what helps. Boundaries extended to cover workload inflation absorbed as personal quota.
  • Vision: "Let's Be Honest" now includes the cognitive cost of transition — not just reskilling and reinvention.
  • Implementation Roadmap: Infrastructure prerequisites now include cognitive capacity protection (workload headroom, concurrent-tool cap, output-quota discipline, manager awareness). Pilot checkpoint at Day 60 asks whether people are burning out, not just whether workflows improved.
  • Glossary: Added Cognitive Cost section — 9 new entries.

Rationale

The framework previously treated the shift from execution to judgment as an unambiguous elevation. The research is clear that it is — and that sustained judgment is cognitively demanding in a way production work isn't. Pretending otherwise is how transformations stall: the most engaged people burn out first. This release integrates that constraint across the framework rather than hiding it in a corner.

v2.1 — April 8, 2026

Content

  • Expanded individual tier scale. The individual maturity scale goes from 3 tiers to 7: T0 (Unexposed), T0.5 (AI-Curious), T1 (AI-Aware), T1.5 (AI-Building), T2 (AI-Augmented), T2.5 (AI-Advanced), T3 (AI-Native). The .5 tiers capture the transitional states where most people actually are – proof of movement, not just a label.
  • Skill Progression Map expanded. Self-assessment tables for all 5 role families now include Level 1.5 and Level 2.5 columns, giving people a closer "next step."
  • Tier references across the framework. Leading the Transformation maps adoption speed groups (Accelerators, Progressors, Stuck) to tiers. Transforming Your Role references tiers in Layers 1 and 3. Assessing Your Organization uses the individual tier scale for finer-grained team assessment.

v2.0 — April 5, 2026

Content

  • The Business Case: New page — why invest in becoming AI-native. Economic signals, competitive risk, investment profile, timeline expectations, and how to pitch internally (read it)
  • Assessing Your Organization: New page — a team-by-team diagnostic for AI maturity. Assessment methodology, observable behaviors by level, common pitfalls, and how to build your maturity map (read it)
  • Skill Progression Map: New page — what Level 1, 2, and 3 look like concretely for engineering, marketing, customer service, sales, and design. Self-assessment questions for each role family (read it)
  • Implementation Roadmap: New page — where to start, how fast, and what to watch for. Sequencing criteria, 30/60/90 plan for leaders, decision checkpoints, and scaling from pilot to organization (read it)

Reorganization

  • Content tightened across the framework: removed redundant statistics and duplicated sections between pages. Each page now has a clearer scope with less overlap.
  • The Reference Framework page now focuses on the conceptual model. Acceptance criteria and transformation path moved to the Implementation Roadmap.
  • The Vision page defers to The Business Case for the full economic argument instead of duplicating it.

Languages

  • Four new languages: Spanish, German, Italian, and Portuguese (Brazil). Full translations of all 18 pages and 3 blog posts, with localized URL slugs.

v1.5 — April 4, 2026

Content

  • Role Evolution: New page — five patterns of role transformation (convergence, specialization, elevation, absorption, emergence), a Role Decision Matrix, and mapping to organizational maturity levels. Based on Patel (IJSR 2026) research, adapted to the framework's vocabulary (read it)

Blog

  • Your Role Is Not Your Tasks: New blog post — the biggest barrier to AI transformation isn't technology, it's that people confuse what they do with who they are (read it)

Cross-references

  • Added role evolution links to: framework, vision, employee guide, manager guide, legacy work patterns
  • Added glossary terms for all five evolution patterns and the Role Decision Matrix

v1.4 — March 24, 2026

Content

  • Specification Guide: New page — practical companion to the Execution Standards with examples, templates, context failure modes, and worked specs for engineering, marketing, and support roles (read it)

v1.3 — March 12, 2026

Blog

  • The Real 50x: New blog post — everyone talks about AI making individuals more productive, but the bigger gain is eliminating coordination entirely (read it)

v1.2 — March 12, 2026

Content

  • Engineering for Unreliability: New page — how to build reliable systems from unreliable AI components, and why this isn't as new as it feels (read it)

Blog

  • Blog section: Added blog with index page, tag filtering, and bilingual support
  • Managing AI Is Managing: First blog post — how working with AI agents teaches management skills, not coding (read it)

v1.1 — March 7, 2026

Content

  • Vision: Added labor market evidence from Anthropic's research on AI labor market impacts — young worker hiring shifts (~14% drop in AI-exposed occupations), knowledge worker exposure demographics (47% higher earnings, 4× more likely to hold graduate degrees)
  • Framework: Added measurable capability-usage gap to Level 1 description — 94% theoretical AI task coverage vs. 33% actual usage in technical roles (source)
  • Vision: Added section on accountability evolution — how accountability shifts from inspection to system design as AI takes over execution

Terminology

  • Engineering scale rungs renamed from role-based to process-based names: Assisted coding, Scoped delegation, Supervised generation, Directed development, Spec-driven development, Autonomous production — replacing previous names that implied a hierarchy of human roles rather than a progression of working modes

v1.0 — March 3, 2026

Initial public release. Nine pages covering vision, reference framework, AI Lab, execution standards, employee guide, manager guide, glossary, legacy work patterns, and about.

← Back to home · About this framework