Vibe Adoption Framework · Roboco
A framework for adopting vibe coding in established enterprises — introducing the Vibe Adoption Framework (VAF)
vaf.roboco.io · github.com/roboco-io/vibe-adoption-framework
Why now
The effect is like hiring a large number of senior-level developers at low cost. Lead time, quality, and cost structure all change.
They use the tools but never convert them into organizational results. It ends with the conclusion "we used AI and nothing changed."
Coding agents already handle much of the work skilled developers used to do. The gap keeps widening.
Core premise
What changes is not the principles but the economics.
Clear requirements, good architecture, review, testing, documentation — the principles that made human development succeed make agent development succeed too.
What used to be expensive — meticulous reviews, thorough testing, full legacy analysis — agents now perform continuously at low cost.
The question is not "what do we change for AI" but "how do we use agents to realize what we should have been doing all along?"
Foundational Principles
Give agents what you give a capable new hire — access, system explanations, rules, clear instructions, review. Failing organizations blame the agent; successful ones build an onboarding system.
"The agent wrote it, so I don't know" is not allowed. Even when the author changes, the duty to explain does not move. You must be able to answer whenever anyone asks.
The gap between human understanding and the actual system = cognitive debt. Like technical debt, it is a trade-off to manage, not something to eliminate. The organization decides where to keep deep understanding.
Named Concepts
The engineering of systematizing the reviews and guardrails humans used to perform so that AI carries them out — a combination of rule files, automated review, hooks, and verification pipelines. A well-built harness elevates human review to a higher level of abstraction: intent, design, trade-offs.
Non-taste (security, guardrails, coding conventions) is managed with the repository and enforced on everyone. Only taste (e.g. level of documentation detail) is left to very limited discretion. Rules embedded in the repository become an environment agents cannot violate.
Framework
"What must the organization have in place?" The unit of assessment and improvement planning.
"What actually changes?" AsIs → ToBe for each domain.
"In what order do we proceed?" A repeating cycle of Prepare → Assess → Pilot → Transform.
Move along the journey, build the capabilities, and use them to turn each domain's AsIs into its ToBe.
Value Chain
Mistaking tool adoption for results — buying licenses is only the first link in the chain.
Demanding outcomes without transformation — the false conclusion that "we used AI and nothing changed."
The executive question: which link in this chain is the bottleneck right now? That is your investment priority.
Transformation Domains
| Domain | AsIs | ToBe |
|---|---|---|
| Code Production | Humans write code; hiring is capacity planning | Agents write code; humans direct intent and approve |
| Quality Assurance | Human review, first to be sacrificed under deadline pressure | The harness enforces automatically; humans focus on design judgment |
| Knowledge | Tacit knowledge and oral tradition; departure = knowledge loss | A documented knowledge base agents use continuously |
| Organization & Roles | Author-centric teams, evaluated by code volume | Owner- and orchestrator-centric teams |
The slowest and hardest transformation is Organization & Roles — left unaddressed, you end up with "tools bought, nobody using them"
Perspectives
Why, and what do we gain — adoption strategy, ROI measurement, SDLC gap analysis, prioritization
The human role — ownership system, cognitive debt management, role transition, capability development, culture
What do we enforce — taste vs. non-taste, repository-enforced policy, security, cost
Can agents work — access channels, tool connections, execution environment, permissions
Do agents know — knowledge base, repository documentation, documented business issues, context supply
Who guards quality — rule files, skills and hooks, automated review, verification pipeline, improvement loop
Perspectives in one line
Why adopt, and what do we gain? — When this perspective is weak, the other five dissolve into individual heroics.
What is the human role? — Human value lies not in what you wrote but in what you are accountable for.
What do we enforce? — Agents have no will to break rules. If a rule is not in the environment, they simply don't know it exists.
Can agents work? — You cannot expect results from a new hire with no account.
Do agents know? — Agent performance is proportional to the quality of accessible knowledge.
Who guards quality? — In an organization without a harness, agent speed is defect speed.
Adoption Journey
VAF puts environment preparation before assessment —
because the agents themselves perform the assessment, analysis, and documentation.
Journey Principles
Finish small — Going narrow to the finish beats spreading wide and stalling. The experience of completing a cycle is the fuel for expansion.
Don't skip stages — Assessment without preparation runs at human speed; a pilot without assessment produces context-free code; transformation without a pilot is unverified coercion.
Verify with in-depth interviews at every stage — Moving to the next stage while documents diverge from reality is the most expensive failure.
Refresh metrics every cycle — Executive trust comes from metric continuity.
Anti-patterns
Six months later, momentum dies with a "usage is low" report.
Unable to prove results, the effort founders at budget renewal.
The answer that comes back is a total block. Bring security in as a design participant.
The cause is not the agent but a gap in the harness.
The full VAF text specifies anti-patterns for each of the 26 capability items
Execution Kit
Diagnostic questions for 26 capabilities × 3 levels (absent / partial / established)
Four areas: business issues · IT issues · SDLC · knowledge management
Assessment workshop (half day) · pilot kickoff workshop (1 day)
All framework terms in Korean and English
From assessment and interviews through pilot kickoff — from reading the framework to executing it.
Expected Outcomes
Feature development, bug fixes, and legacy analysis run in parallel, continuously.
Review and verification move into the harness and are enforced without exception.
Development capacity is no longer capped by headcount.
Tacit knowledge becomes a documented knowledge base, ending dependence on individuals.
Humans not writing code by hand becomes the default,
sustained by a system of ownership and cognitive debt management.
Vibe Adoption Framework
vaf.roboco.io
(EN · 한국어 · 日本語)
github.com/roboco-io/
vibe-adoption-framework
Roboco — roboco.io
© Roboco · Vibe Adoption Framework