2026/10 - Intent, Implementation, Verification: Engineering in the Age of Coding Agents (Duygu Cakmak)
Version du programme : 1
Type de formation
DistancielDurée de formation
8 heures (4 jours)Accessibilité
Oui2026/10 - Intent, Implementation, Verification: Engineering in the Age of Coding Agents (Duygu Cakmak)
Agentic coding is developing fast. As more and more developers adopt these technologies in their regular workflows, new best practices emerge, engineering processes change and so does the role of the engineer. Autonomous coding agents don't only autocomplete code and support boilerplate generation or testing. They can now create, build, test, debug and iterate through an autonomous loop cycle. This course is aimed at mid-to-senior level game engineers and technical directors to learn how to react to such paradigm shift and how to manage agentic code generation workflows using intention, scope and verification. Participants will learn how to apply these practices to their teams, tools and development process. From October 19th to October 22nd - 10:00am-12:00pm every day (CET Time) - Remote training (GR1) From October 26th to October 29th - 10:00am-12:00pm every day (CET Time) - Remote training (GR2) I - The New Shape of a Code Change II - From Intent to a Controlled Agent Loop III - Review Behaviour, Not Fluency IV - Optimisation, Orchestration and the Agent Harness
Objectifs de la formation
- Explain how autonomous coding agents are transforming software development workflows and the role of engineers within them.
- Structure agent-assisted development tasks using the principles of intent, implementation and verification
- Write agent-ready specifications that clearly define expected behaviour, constraints, edge cases and success criteria.
- Define and control the scope of agent-generated changes to limit unintended impact on the codebase.
- Supervise autonomous coding loops effectively, while maintaining human ownership of specifications, critical decisions and verification.
- Identify common risks and pitfalls in agent-generated code and apply appropriate code review practices.
- Design effective verification strategies to assess the correctness, quality and reliability of AI-generated changes.
Profil des bénéficiaires
- Technical Directors
- Level Game Engineers
- Technical Leads
- Senior Developers
- Engineering Managers
- CTOs
- Professional experience writing and reviewing game code, ideally in C++ or an equivalent production language.
- Some exposure to a coding agent, such as Claude Code, Codex, Cursor, Copilot is helpful but not required.
- The course is tool-agnostic and focuses on transferable engineering practice.
Contenu de la formation
The New Shape of a Code Change
- Course Introduction & Overview of AI coding systems and agents: Baseline poll, history of AI assistants, and defining agentic coding systems.
- The Three-Layer Core Model: Introduction to Intent, Implementation, and Verification and reframing the engineer's role. Pair activity decomposing real code changes into intent and implementation layers.
- Standing Intent & Context Engineering: Strategies for agent briefing and system-level context. Practical drafting of system-specific standing intent with peer review.
From Intent to a Controlled Agent Loop
- Controls, Curation & Reach: Introduction to agent-ready intent, interface-first scoping, and "design by contract". Activity is converting backlog items into agent-ready assignments and reviewing proposed plans.
- Verification Strategies & Gaming: Locking verification targets and identifying ways agents game weak tests.Hands-on practice separating agent feedback loops from authoritative checks.
Review Behaviour, Not Fluency
- Review Methodology & Failure Smells: Review order (Intent , Behaviour, Implementation) importance and recognizing common "agentic smells." Activity: Rapid review rounds identifying behavioural failure patterns.
- Advanced Verification: Exploring adversarial reviewers, formal testing paradigms, and dual-agent pipelines. Activity is configuring a builder/auditor loop to catch edge-case failures.
Optimisation, Orchestration and the Agent Harness
- Optimisation as Verification: Framing optimisation as evidence-gathering, identifying hot paths, and benchmarking. Small-group exercise identifying the fastest behaviourally-equivalent variants.
- Orchestration & The Agent Harness: Building the support environment: context, isolated workspaces, observability, and escalation rules. Activity is assessing existing workflows and defining concrete harness, triage, or guardrail improvements.
Équipe pédagogique
Suivi de l'exécution et évaluation des résultats
- Exercises during the training sessions and quizzes throughout the days
Ressources techniques et pédagogiques
- Welcoming participants in a dedicated training room. Training materials displayed on a screen. Theoretical presentations Practical case studies Training materials made available online following the session.
Qualité et satisfaction
Modalités de certification
- Delivery of a certificate of completion.
Délai d'accès
Accessibilité
For companies wishing to have their training costs covered by their OPCO, registration must be completed no later than two weeks before the start of the training to meet the OPCO’s processing deadlines. In all other cases, registrations can be accepted up to two business days before the start of the training. Accessibility: We are available to work with individuals with disabilities to assess the accommodations needed so they can comfortably participate in the training programs offered.