Agent Readiness

Most engineering teams I talk to have the same story: a handful of developers use Copilot or Claude daily, a few more experiment occasionally, and the rest are skeptical. With a few exceptions, even the enthusiastic adopters are usually doing something more modest than they realize: just small steps. This is not a matter of skills or vision. The real question most of the time turns out to be: is this codebase even ready for agents to do more?

My focus is primarily about closing that gap between "we use AI tools" and "we get systematic value from agents". Realizing that agent readiness is the first step I developed a framework to measure and improve both the structural conditions in the codebase that enable agentic coding. For completeness I developed also a framework to measure and improve the level of adoption of agentic engineering. The result is Agentize, an open-source framework with two complementary maturity models:

Agent Readiness
Agent Adoption

Agent Readines is a measure of how well the codebase supports agentic coding. Agent Adoption is a measure of how deeply agentic coding is embedded in the team's workflow. Both are important, but you need a certain level of readiness to get adoption. Of course agentic adoption can make it easier to reach readiness, but in my experience most teams are held back by readiness gaps that adoption alone can't fix.

Getting StartedRisk ZoneMissed PotentialFull Valueideal pathmost teams0123Readiness01234Adoption
Both dimensions need to advance together. The amber dot marks where most teams currently sit.

The Readiness Model

The Agent Readiness Maturity Model measures structural, environmental, and documentary properties of a codebase. It has four levels.

LEVEL 0
Uninstrumented
very little or no benefit from applying agents, agents have no context or feedback mechanism
LEVEL 1
Disconnected
agents would be able to perform small steps, but humans needs to micromanage; agents have some feedback (e.g. compilation, unit tests)
LEVEL 2
Connected
humans still guide but agents now have context to perform larger steps (plan a story, write tests, etc.)
LEVEL 3
Reflexive
agents have full context and strong feedback loops (e2e tests, screenshots) and can autonomously implement most stories

The model is scored across 11 specific criteria in 8 groups. The overall level is the highest at which all required criteria are met — so you can identify exactly which criterion is dragging your level down.

Codebase AccessibilitySetup AutomationArchitecture DepthRequirements AccessRunnabilityUnit Test CoverageIntegration & E2E CoverageStatic AnalysisTest IsolationCI/CD AutomationObservability

The key insight:

Agents amplify what's already there

A codebase with good tests and documented conventions lets an agent work autonomously and catch its own mistakes. A codebase without them produces output that breaks things in ways that are hard to detect.

The Adoption Model

The Agent Adoption Maturity Model measures behavioral dimensions: not whether agents are installed, but how deeply they're embedded in how the team actually works. It has five levels.

LEVEL 0
Unassisted
No systematic agent use, no agent-authored code in git history.
LEVEL 1
Vibe Coding
Agents as smart assistants: autocomplete, code generation, explaining errors. The human is still the author. A valid starting point, but not the productivity transformation the hype promises.
LEVEL 2
Agentic Engineering
Agents implement full stories. The developer reviews the resulting diff, not each step. Agents run their own feedback loops: write tests, fix failures, present only passing results. The key difference from Level 1: self-verification.
LEVEL 3
Software Factory
A continuous pipeline pulls stories from the backlog on a schedule. The team defines requirements, reviews PRs, handles exceptions. No human initiates each story.
LEVEL 4
Sustainable Autonomy
Agents proactively open PRs for tech debt, dependency updates, and security fixes. Incidents trigger automated diagnosis. Most teams won't need this level.

The model is scored across 8 criteria. The overall level is the highest at which all required criteria are met.

Agent Context AvailabilityAgent-Authored ContributionsFeedback Loop ClosureTask ScopeWorkflow IntegrationAutonomous OperationProactive Quality ManagementPlanning Integration

Evaluate Your Project

Answer 19 questions to get your Readiness and Adoption scores, plus a gap analysis showing exactly what to improve next.

Part 1: Agent Readiness
How ready is your project's structure for agentic coding?
1 / 19
C1.1 Codebase Accessibility

Automated Assessment with Claude Skills

If you want to skip the quiz and get an immediate, automated assessment of your project, I've created a GitHub repository that you can use. The Agentize repository ships a suite of Claude Code skills that do the assessment for you. The main advantage is that it evaluates the criteria objectively, from the point of view of the agent, avoiding any human bias or blind spots. Your projects do not need to be public: the skills run locally on your machine and can access private repos without any data leaving your environment.

Besides the skills for assessing the Agent Readiness level and the Agent Adoption level, the repository also includes a set of skills to improve your project's readiness and adoption by automatically generating pull requests that implement the missing criteria. For example, if your project has low unit test coverage, the skill will generate tests for untested code. If your project lacks architecture documentation, a skill can generate a system overview based on code analysis. The possibilities are endless, and you can even customize the skills or create your own to target specific gaps in your project.

Check out the Agentize repository to get started.

© 2025 Jaksa Vuckovic. Built with SvelteKit