Agentic Software Engineering with AI Coding Agents

Move beyond isolated prompts and use coding agents as part of a disciplined software engineering workflow. Your team learns how repository context, tools, hooks, tests, permissions, and human review form a harness that helps Codex, Claude Code, GitHub Copilot, and similar agents produce useful, reviewable changes in Java projects.

Book a 30-min call Request by email

Two-day in-house course for up to 12 participants. Delivered in Zürich, at your office across Switzerland or Europe, or fully remote; exercises use a prepared Java repository or your own codebase where suitable.

When This Training Is Worth Your Team's Time

For engineering teams that already have access to coding agents and want repeatable delivery improvements rather than occasional individual experiments.

  • Adoption Is Uneven

    Some developers get useful results while others lose time correcting broad, unverified changes. The team needs a shared working method.

  • The Repository Fights the Agent

    Knowledge is implicit, setup is fragile, and quality checks are slow or hard to discover. Agents cannot reliably observe whether their changes work.

  • Governance Must Become Practical

    The organisation needs clear boundaries for tools, data, permissions, validation, and human accountability without blocking useful engineering work.

What Your Team Will Build and Practise

Participants prepare a repository for agent-assisted work, give an agent a bounded engineering task, and take the resulting change through automated checks and human review. Tool-specific examples are adapted to the coding agents your team uses.

  • Agent Capabilities & Limits

    Choose tasks that fit the agent's available context, tools, and feedback. Recognise where uncertainty, missing system knowledge, or risk requires a smaller step or human decision.

  • Agent-Ready Repositories

    Make architecture, conventions, commands, and boundaries discoverable. Write concise repository instructions that guide work without duplicating documentation or going stale.

  • Task Design & Context

    Turn a ticket into a bounded task with relevant evidence, explicit constraints, and acceptance criteria that an agent and reviewer can verify.

  • Harness Engineering

    Connect the agent to the files, search tools, build commands, tests, and local services it needs. Design fast feedback loops so it can inspect the effect of each change.

  • Tools, MCP & Integrations

    Decide which external capabilities the agent should access and expose them with narrow, understandable interfaces. Use MCP when it improves the workflow and control the data shared.

  • Reusable Workflows & Delegation

    Package recurring work as project instructions, commands, skills, or playbooks. Delegate focused work to specialised or parallel agents when ownership, boundaries, and integration checks are clear.

  • Hooks & Automated Checks

    Run formatting, static analysis, focused tests, and policy checks at useful points in the workflow. Keep checks fast enough to guide the agent while reserving broader validation for the right stage.

  • Permissions, Secrets & Boundaries

    Scope file, command, network, and credential access to the task. Establish approval points for consequential actions and handle untrusted repository content deliberately.

  • Reviewing Agent Work

    Inspect the diff, assumptions, test evidence, and unintended effects. Treat generated code as a proposed change whose owner remains responsible for correctness and maintainability.

  • Team Adoption & Measurement

    Agree where agents fit the delivery process, which practices the team shares, and how to evaluate lead time, review effort, defect risk, and developer experience.

Who Is This Training For?

The course is for people who build, review, or guide professional software and want AI coding agents to work inside their existing engineering standards.

  • Java & Spring Developers

    Developers who want to use coding agents for implementation, debugging, refactoring, testing, documentation, and code review in established repositories.

  • Tech Leads & Architects

    Leads responsible for repository structure, technical guardrails, review quality, and deciding which agent-assisted workflows the team should standardise.

  • Engineering Enablement Teams

    Platform and developer-experience teams designing shared tools, templates, policies, and feedback loops for responsible adoption.

Training Details

Short explanations lead into repository work, agent runs, and review. We align the exercises with your selected tools, development environment, and engineering policies.

Format

In Zürich, on-site at your office, or fully remote. Participants work in small groups on connected exercises.

Duration

2 days. For in-house delivery, we can adapt the agenda or split it into focused modules.

Language

Delivered in English or German. Italian available on request. Course materials are in English.

Prerequisites

Professional software development experience, working knowledge of Git, and familiarity with the build and test workflow of a Java project. No AI or ML experience required.

Frequently Asked Questions

Which AI coding agents does the training cover?

The engineering practices are tool-independent. Exercises can use Codex, Claude Code, GitHub Copilot, or the combination your team has selected, subject to the capabilities and policies available in your environment.

Is this a prompt engineering course?

Prompting is one part of the course. The main focus is the complete engineering system around an agent: context, tools, permissions, hooks, build and test feedback, acceptance criteria, and human review.

How is this different from Spring AI & Embabel?

This course teaches developers to engineer software with AI coding agents. The Spring AI & Embabel training teaches teams to build model-backed features and AI agents into Java applications.

Can we use our own repository?

Yes, when access, confidentiality, and setup allow it. Otherwise we use a prepared Java repository that exposes the same decisions about instructions, tooling, tests, hooks, permissions, and review.

How much does the training cost?

We deliver in-house, for up to 12 participants. The final price depends on group size, duration, and how much we tailor the content. Book a 30-minute call or email hello@42talents.com. We reply personally to emails within one business day.

Related Training

Combine agent-assisted development practices with the technical foundation your Java and Spring systems require.

Build a reliable coding-agent workflow

Tell us which agents your team uses and where the current workflow breaks down. We'll shape the course around your repositories, toolchain, delivery goals, and policies.

In-house, up to 12 participants. Content tailored to your codebase and engineering environment.

Book a 30-min call Request by email