Coding


Active development

The process of designing, coding, and implementing new software features or systems. It involves translating requirements into functional code, applying best practices, and ensuring that new components integrate smoothly with existing systems.

Maturity levels

Level

Name

Description

Technology

Example tools

0

Non-existent

No AI capabilities are available for development tasks. All code creation, syntax validation, and design decisions are performed manually. There is no AI functionality in tools or processes.

  • Manual
  • IDE without AI features
  • Text editor

1

One-off assist.

AI can generate small code snippets and perform syntax correction. Suggestions can be copy/pasted from standalone tools when needed. There is no integration with IDEs or the development pipeline.

  • Prompt-based code generation (chat)

2

Integrated assist.

AI works inside the IDE with project context: completions, next-edit suggestions, multi-file edits and refactorings proposed against existing coding standards. It can generate whole functions or classes and iterate with the developer, but the developer stays at the keyboard and accepts or rejects each change.

  • Context-aware assistants
  • In-IDE agents

3

Supervised delegation

A scoped task (an issue, a ticket, a spec) is handed to an agent that plans and executes it unattended in an isolated workspace, runs the build and the tests, and returns a pull request together with its session log. Several agents can run in parallel on separate worktrees. Nobody watches the work; the human owns review and merge.

  • Asynchronous coding agents
  • Agent orchestration surfaces
  • Github Copilot Cloud Agents
  • Claude Code Managed Agents

4

Full autonomy

AI autonomously manages the entire development cycle, including design, coding, and documentation. It interprets requirements, produces optimized solutions, runs validations, and integrates code with minimal human supervision.

  • Agentic autonomous systems
    None proven
  AI Maturity Level: Indicates the level the technology vendors claim to have reached in deploying AI solutions that actually work in real-world applications

Legacy code refactoring

The practice of improving the structure and quality of existing code without changing its external behavior. The goal is to increase readability, maintainability, and performance while reducing complexity and technical debt.

Maturity levels

Level

Name

Description

Technologies

Example tools

0

Non-existent 

AI has no capability to assist in legacy code improvement. All restructuring, modernization, and dependency management are performed manually. Code complexity and technical debt are addressed reactively, if at all.

  • Manual
  • Traditional IDEs

1

One-off assist.

AI can suggest small refactorings such as renaming variables, removing unused code, or simplifying isolated logic. Actions are triggered manually and applied without system-wide context.

  • Prompt-based refactoring suggestions

2

Integrated assist.

AI detects code smells and proposes cross-module modernisations inside the IDE and the CI pipeline, on top of deterministic recipes that perform the mechanical part of the change. The developer validates the impact and resolves conflicts change by change.

  • Pattern-aware refactoring tools

3

Supervised delegation

AI analyzes full codebases, identifies systemic issues, and proposes major refactoring strategies. Developers collaborate by validating impacts, resolving conflicts, and adjusting AI-generated solutions to align with business goals.

  • Large-context AI Agents
  • Github Copilot Cloud Agents
  • Claude Code Managed Agents
  • Devin

4

Full autonomy

AI autonomously identifies legacy code, creates a refactoring plan, executes changes across the system, and validates functionality through automated testing. It maintains compatibility and performance without human intervention.

  • Autonomous codebase evolution systems
  • None proven
  AI Maturity Level: Indicates the level the technology vendors claim to have reached in deploying AI solutions that actually work in real-world applications

Security assessment

The evaluation of software systems to identify vulnerabilities, weaknesses, and compliance issues. It includes analyzing code, configurations, and dependencies to ensure the application meets security standards and protects sensitive data.

Maturity levels

Level

Name

Description

Technologies

Example tools

0

Non-existent 

AI provides no security analysis. All vulnerability checks, threat modeling, and mitigations are manual or handled by non-AI tools.

   

1

One-off assist.

AI can scan code for common vulnerabilities and report results on demand. These assessments are reactive and limited to specific requests or periodic checks, with results manually interpreted.

  • Prompt-based vulnerability checks

2

Integrated assist.

AI continuously scans for known security risks and dependency vulnerabilities as part of CI/CD pipelines. Findings are enriched with runtime context (internet exposure, data sensitivity) so they can be prioritised, and remediations are proposed for the developer to implement. The developer applies and validates each fix.

  • Static Application Security Testing (SAST)
  • SCA with AI triage

3

Supervised delegation

An agent monitors commits continuously, reasons about exploitability, and opens fix pull requests for review. Findings can be batched into remediation campaigns covering many alerts at once.

  • AI threat modeling and remediation agents
  • GitHub agentic autofix (Copilot cloud agent)
  • Google Big Sleep

4

Full autonomy

AI independently monitors systems, detects vulnerabilities, applies patches, and validates security measures. It dynamically adapts to new threats and maintains compliance without human involvement.

  • Autonomous security auditing and patching agents
  • None proven

* A lot of vendors advertise that their AI agents are fully autonomous at discovering security breaches and fixing them. But the actual results are often “showcases” that do not really translate into real-life situations.

  AI Maturity Level: Indicates the level the technology vendors claim to have reached in deploying AI solutions that actually work in real-world applications

Unit tests generation

The creation of tests that verify the correctness of individual units of code. These tests help ensure that small parts of the software behave as expected and continue to do so over time, preventing regressions during future development or system changes.

Maturity levels

Level

Name

Description

Technologies

Example tools

0

Non-existent 

AI offers no capability for generating or maintaining unit tests. All test creation and updates are manual.

   

1

One-off assist.

AI can generate unit tests for individual functions or classes when requested. It creates simple assertions and edge cases without deep code context.

  • Prompt-based test generation

2

Integrated assist.

AI produces and updates test suites as code is written, identifies coverage gaps, and proposes edge cases and integration points. Deterministic generators cover the bulk regression cases. The developer validates each test against business logic and critical paths.

  • Embedded and context-aware test generation

3

Supervised delegation

An agent is given a coverage or regression objective, writes and maintains the suite unattended as the code evolves, and returns pull requests. Test quality is verified by mutation testing rather than by coverage percentage.

  • Autonomous test maintenance under review
  • Github Copilot Cloud Agents
  • Claude Code Managed Agents

4

Full autonomy

AI autonomously creates, updates, and maintains test suites throughout the software lifecycle. It ensures complete coverage, adapts tests as code changes, and validates results without human intervention.

  • Autonomous test maintenance systems
  • None proven
  AI Maturity Level: Indicates the level the technology vendors claim to have reached in deploying AI solutions that actually work in real-world applications

Code Documentation

Code documentation relies on three distinct but complementary aspects:
  • Inline Documentation (like Javadoc) explains the immediate purpose, usage, parameters, and return values of specific code elements directly within the source files for developers working with the code.
  • Architecture, Design & Implementation documentation details the high-level structure, technical rationale, data flows, and component interactions of the entire system. This is crucial for maintaining the system’s design integrity, supporting complex troubleshooting, and guiding the long-term evolution of the software.
  • Framework/Library Tutorials & User Guides provide goal-oriented, step-by-step instructions and practical examples to help a developer successfully utilize the technology.

Maturity levels

Level

Name

Description

Technologies

Example tools

0

Non-existent 

No assist. Documentation is manual, and often external (e.g., separate wiki pages) with no AI contribution.

  • Manual
  • Markdown
  • MS Word
  • Confluence
  • …

1

One-off assist.

Reactive Documentation. AI generates comprehensive documentation (inline comments, high-level summaries) for code blocks or files, utilizing a single chat session. The assist is reactive and requires explicit commands, with context lost once the session is closed.

  • Chatbot

2

Integrated assist.

AI proactively drafts and updates inline documentation and API summaries inside the IDE, accepted change by change. Repo-resident context files start to make documentation an input to the tooling and not only an output.

  • IDE Assistant

3

Supervised delegation

On merge, or on a schedule, AI regenerates and reconciles architecture, design and cross-module documentation against the code, and opens a documentation pull request. The human retains review, approval and strategic direction over architectural narrative and compliance statements.

  • Scheduled documentation agents
  • Github Copilot Cloud Agents
  • Claude Code Managed Agents

4

Full autonomy

Self-Maintaining Documentation. The AI maintains a continuous, self-correcting external knowledge base (living documentation) that mirrors the code, generating user guides, tutorials, and training materials without human intervention.

  • Autonomous Agent
  • None proven
  AI Maturity Level: Indicates the level the technology vendors claim to have reached in deploying AI solutions that actually work in real-world applications

Major upgrade

A Major Upgrade is a technology update that cannot be automated by the normal build pipeline (e.g., Renovate) because it introduces a fundamental paradigm shift in the underlying framework. Such upgrades typically require manual code adjustments, changes in application architecture, or adoption of new APIs. For example, upgrading from Spring Boot 2 to Spring Boot 3 involves the platform-wide transition from Java EE to Jakarta namespaces (javax.* → jakarta.*), which affects imports, libraries, and compatibility assumptions across the entire codebase.

Maturity levels

Level

Name

Description

Technologies

Example tools

0

Non-existent 

No AI is used in the development workflow; all tasks (coding, reviews, testing, documentation) are performed manually.

  • Manual
IDE (without AI features)

1

One-off assist.

Developers occasionally query standalone AI tools outside the IDE (e.g., to draft snippets, explanations, or regex), with no integration into the toolchain or codebase context.

  • Chatbot

2

Integrated assist.

Mechanical transformations are executed by deterministic recipes (namespace rewrites, API replacements) while an in-IDE agent handles the residue: broken configuration, changed defaults, transitive dependency conflicts, tests that no longer compile. The developer drives the sequence and validates module by module.

  • Recipe engines plus context-aware agents

3

Supervised delegation

The upgrade is handed to an agent that performs the migration across the repository, fixes the build, runs the suite and returns a reviewable pull request per module.

  • Asynchronous agents on repository-wide change
  • Github Copilot Cloud Agents
  • Claude Code Managed Agents
  • Cognition’s Devin

4

Full autonomy

AI agents plan and execute end‑to‑end tasks (issue analysis, coding, tests, PRs, validation) with minimal human intervention, operating across tools, repos, and environments under policies/guardrails.

  • Autonomous Agent
  • None proven
  AI Maturity Level: Indicates the level the technology vendors claim to have reached in deploying AI solutions that actually work in real-world applications

Migration

Migration is the transformation of an application from one technology ecosystem to another when the target platform requires a new architecture, runtime, or programming model.

This can involve in‑ecosystem modernization (e.g., JEE → Spring Boot, where architectural decomposition and framework replacement are needed) or cross‑ecosystem re‑platforming (e.g., COBOL → Java, where business logic must often be re‑extracted and re‑implemented).

Effective migration requires two essential preparation phases:

Functional Analysis
  • Understanding current business rules, workflows, edge cases, domain objects, and user interactions.
  • Often requires domain expert interviews, reading COBOL business logic, or mapping JEE services to domain processes.
  • Ensures that what is migrated is functionally correct and complete.

Technical Analysis
  • Inventory of components, modules, interfaces, data structures, copybooks, integrations, deployment models, and operational dependencies.
  • Identifies the architectural gaps between source and target platforms (e.g., EJB container → Spring container, or COBOL file I/O → Java persistence).

Through these analyses, teams establish the migration blueprint, after which the transformation involves:
  • architectural redesign,
  • library and API replacement,
  • code rewrite or translation,
  • test regeneration,
  • integration and deployment modernization.

Migration is therefore the most complex form of code evolution—beyond refactoring and version upgrades—because it alters how the system is built, run, and structured.

Maturity levels

Level

Name

Description

Technologies

Example tools

0

Non-existent 

All analyses and migration activities are manual. Engineers read COBOL logic or JEE modules, map components, redesign architecture, and rewrite code without AI assistance.

  • Manual tooling
Traditional IDEs, mainframe tools

1

One-off assist.

Developers use external chatbots for isolated help (e.g., asking “How to convert JAX‑RS to Spring Web?”), but AI has no system context and cannot reason about the project architecture.

  • Chatbots

2

Integrated assist.

AI operates on the repository and its artefacts. In the analysis phases it extracts business rules, maps copybooks, dependencies and interfaces, and helps produce the migration blueprint. In the transformation phase it generates module templates, proposes target layering and translates programs unit by unit. Engineers validate semantics at every step.

  • Repository-scale analysis, discovery tooling, design co-pilots

3

Supervised delegation

Agent workflows chain dependency mapping, impact analysis, code generation and compilation verification, producing reviewable pull requests across layers. Functional equivalence is demonstrated by parallel run and output comparison against the legacy system, not by review alone.

  • Agentic migration workflows, parallel-run verification
  • IBM watsonx Code Assistant for Z
  • Google Cloud Dual-Run

4

Full autonomy

AI agents perform end-to-end migration tasks: analyzing the legacy system (including data flows, business logic, dependencies), producing architectural proposals, rewriting code across layers, generating integration adapters, creating tests, and submitting PRs. Engineers remain supervisory reviewers.

  • Autonomous Agent
  • None proven
  AI Maturity Level: Indicates the level the technology vendors claim to have reached in deploying AI solutions that actually work in real-world applications