Outline updated November 2026
AI for Software Developers
Class Duration
35 hours of live training delivered over 5 days.
Student Prerequisites
- At least one year of professional software development experience in any language
- Familiarity with Git
- No prior AI tool experience required
Target Audience
Professional developers, tech leads, and engineering managers on mixed-stack teams who want one shared curriculum, independent of language or vendor. Teams on a single stack can take a language edition instead: AI-Assisted Development with Python, Rust, JavaScript and TypeScript, or C#.
Description
This five-day intensive teaches developers to work with AI coding assistants and agents in any language. Participants start with how frontier models generate code and how prompting and context engineering steer them. They then compare current tools, delegate multi-step tasks to agents, and review what the agents produce. Later days cover AI for design, refactoring, testing, debugging, and documentation, then project context, MCP, security, and governance. Shorter courses go deeper on prompting, coding agents, spec-driven development, and MCP.
Learning Outcomes
- Explain how LLMs generate code and choose a model for a task.
- Apply prompting and context engineering to get correct, idiomatic code.
- Compare IDE assistants, agentic IDEs, and terminal agents for your team.
- Delegate multi-step tasks to coding agents and supervise them.
- Review agent output with a consistent checklist.
- Use AI for design, refactoring, testing, debugging, and documentation.
- Ground assistants with instruction files, Agent Skills, and hooks.
- Connect assistants to tools and data through MCP.
- Apply security, licensing, and governance guardrails.
- Measure productivity impact and plan team adoption.
Training Materials
Comprehensive courseware is distributed online at the start of class.
Software Requirements
Students need a laptop with Git and a modern IDE or editor, plus access to an agentic AI coding assistant such as GitHub Copilot, Claude Code, Cursor, OpenAI Codex, Kiro, Google Antigravity, or Devin Desktop (a free tier is sufficient). Labs provide production-style codebases in several mainstream languages, so no language-specific toolchain setup is required in advance.
Training Topics
How Coding Assistants and Frontier Models Work
- How LLMs generate code
- Tokens and context windows
- Claude, GPT, and Gemini model families
- Speed, cost, and reasoning tradeoffs
- Where AI helps in engineering work
- Where AI reliably fails
- Assistant, pair, or junior engineer
Prompting and Context Engineering for Code
- Prompts for correct, idiomatic code
- Choosing context from a large codebase
- Few-shot examples
- Iterative refinement
- Structured outputs
- When to start a fresh conversation
Current Coding Tools
- GitHub Copilot and Cursor
- Kiro and Google Antigravity
- Devin Desktop
- Claude Code and OpenAI Codex
- Copilot CLI and the cloud agent
- Autonomy levels
- Selection criteria and cost
- Multi-tool strategies
Delegating Work to Coding Agents
- The agentic loop
- Writing task briefs
- Scope, constraints, and acceptance criteria
- Plan-then-execute sessions
- The supervision spectrum
- Issue-to-PR with cloud agents
- Tests in the loop
Reviewing and Steering Agent Output
- A review checklist for AI code
- Hallucinated APIs
- Subtle logic errors
- Over-generation
- Interrupting and redirecting an agent
- Small, reviewable diffs
- Understanding code you sign
AI Across the Lifecycle: Design, Implementation, and Refactoring
- Exploring architecture options
- Tradeoff analysis and spikes
- From specification to reviewed code
- Spec-driven development and AI-DLC
- Human approval gates
- Multi-file refactors
- Legacy code behind characterization tests
- Dependency and framework upgrades
AI-Driven Testing and Debugging
- Generating test suites
- Coverage gaps and edge cases
- Mutation-style review of AI tests
- Stack traces and logs to hypotheses
- Bisection with agents
- Validating proposed fixes
Documentation and Code Understanding
- Onboarding to an unfamiliar codebase
- Explanations and call graphs
- Architecture summaries
- READMEs and how-to guides
- Architecture decision records
- Summarizing pull requests
- Keeping generated docs current
Project Context and Custom Instructions
CLAUDE.md,AGENTS.md, and rules files- Agent Skills and
SKILL.md - Hooks as guardrails
- Encoding repo conventions
- Team prompt libraries
- Keeping context files current
Extending Assistants with MCP and Tool Integrations
- What MCP standardizes
- Connecting internal tools and data
- Installing existing MCP servers
- Security boundaries for tools
- Building versus adopting an integration
Security, Licensing, and Governance
- Prompt injection in agent workflows
- Keeping secrets out of prompts
- Vulnerabilities in generated code
- Licensing and IP of AI code
- Approved tools and audit trails
- Review requirements
Measuring Impact and Sustainable Adoption
- Cycle time and review iterations
- Defect rate
- Vanity metrics
- Adoption anti-patterns
- Tokens, model tiers, and budgets
- Champions, guilds, and paired onboarding
- Re-evaluating tools over time