Search

Home > Machine Learning Guide > MLA 024 Agentic Software Engineering
Podcast: Machine Learning Guide
Episode:

MLA 024 Agentic Software Engineering
Category: Technology
Duration: 00:15:14
Publish Date: 2025-04-13 05:55:00
Description:

Development is shifting from manual coding to agentic engineering, where users orchestrate AI agents to produce production code. This approach uses git worktrees for parallel execution and automated hooks to ensure technical quality and context efficiency.

Links Agentic Engineering

Agentic engineering treats AI agents as the primary interface for production code. Developers manage "jagged intelligence" by deciding when to trust agents and when to run parallel instances.

MCP and Context Management

Minimize active MCP servers to preserve the 200K token context window. Each server uses 4-10K tokens. Use project-scoped claudecode.json to load tools only where needed.

  • GitHub MCP: Handles PRs, issue management, and code search.
  • Context7: Fetches version-specific library documentation to prevent hallucinations.
  • Sequential Thinking: Structures architectural reasoning and prevents premature conclusions.
  • Playwright MCP: Automates browser testing and UI debugging via accessibility snapshots.
  • Memory MCP: Stores persistent project knowledge locally in a JSONL file.
Automation and Hooks

Hooks enforce technical standards that CLAUDE.md cannot.

  • PostToolUse: Auto-formats edits and runs Prettier, ESLint, or tsc for immediate self-correction.
  • PreToolUse: Prevents edits to protected files and blocks dangerous shell commands.
  • SessionStart: Re-injects necessary instructions after context compaction.
  • Skills: Use Skilz for package management. RGR-skill enforces TDD discipline, while n8n-skills generates production workflows.
Advanced Workflows
  • Plan-First Discipline: Use claude --new in Plan Mode to explore the codebase and create a TODO.md before starting implementation.
  • Git Worktrees: Execute 3-5 instances simultaneously on the same repository using claude worktree create.
  • Headless Automation: Use the --headless flag with GitHub Actions or n8n for unattended PR reviews and issue implementation.
  • Agent Teams: Use claude --team for multi-agent coordination, shared task lists, and dependency tracking.
Setup Progression
  1. Tier 1: Short CLAUDE.md, basic MCP stack, and quality hooks for solo developers.
  2. Tier 2: GitHub MCP integration, custom slash commands, and parallel worktrees for concurrent tasks.
  3. Tier 3: Event-triggered headless scripts, Agent Teams, and sandboxed environments for autonomous execution.
Total Play: 0

Users also like

300+ Episodes
Data Science .. 300+     20+
400+ Episodes
Revolutions 2K+     50+
2 Episodes
Anxiety & De .. 20+    
100+ Episodes
Fisicast 800+     60+