The Agentic Engineering Bootcamp: Beyond Prompts & Vibes

Explore the five levels and five mindsets of AI coding skills and why you are stuck at level 2, with research and expert insights that challenge your developer identity.

Examine the productivity paradox of AI coding, map the five levels from level 0 to 5, including level 2 junior dev, and learn why AI can slow you despite speed.

Level 0 to 1 shows how AI assists coding without taking control: auto-completing variable names and delegating discrete tasks like writing tests, while you remain the architect.

Level 2 introduces AI-assisted coding that speeds feature delivery but keeps you as a bottleneck, reviewing every diff and not transforming your workflow.

Level 3 shifts you from coding to reviewing AI-generated diffs, focusing on architecture, patterns, and edge cases. Describe needs in paragraph, like adding a caching layer with Redis; assess correctness.

Level 4 requires you to write a plain-English specification in Markdown, describe what should exist, argue with AI about requirements rather than implementation, and verify with tests.

Level 5, the dark factory, automates coding from spec to production with AI generating code, writing tests, and shipping, while humans approve outcomes and measure business value.

Shift from coding speed to precise specification, describing what should exist, edge cases, and success criteria, and move toward roles of reviewer, product manager, and architect.

Follow a practical progression from coder to architect: write specs before code, trust AI for implementation, test, ship when tests pass, and measure business outcomes.

examines ai coding tools and productivity, citing METR's controlled study showing 20% higher perceived productivity but 19% longer task completion due to reviewing and steering ai outputs in complex codebases.

The five levels of AI coding skills framework, inspired by self-driving factories, maps AI roles from spicy auto-complete to the dark factory, guiding teams toward autonomous software creation.

The lecture presents StrongDM's level five mastery and its no-human-code-review manifesto, plus threshold, then explains building a digital twin to test security software with AI-generated code guided by human intent.

Dark code is functional but incomprehensible, accumulating silently as generated AI code passes tests while hiding behavior and security risks, making debugging feel like archaeology.

Adopt a three-layer defense in software development: spec-first development, self-describing code and docs, and a comprehension gate to ensure review clarity before shipping.

Learn how to separate fuzzy human requirements from precise specifications, leveraging PRD level detail, edge cases, and institutional (tacit) knowledge to guide AI and avoid misalignment.

Promotes maintaining clean, non-spaghetti code and strong observability, while guiding AI with explicit constraints—such as function length, parameter limits, and loose coupling—to ensure quality without human code review.

Explore loop engineering and its hype, including automated agents, parallel work trees, and ai running ci overnight. Assess the real costs and failure modes when automation misleads on call.

Explore loop engineering in the agentic engineering bootcamp by designing automatic cycles that prompt agents, assign work, verify with a second agent, and surface only exceptions.

Identify the six building blocks of loop engineering: automations, work trees, skills, MCP plugins, sub-agents, and persistent memory that track what ran and what remains open.

Learn when loop engineering boosts productivity through stable tasks, repeatable triggers, and self-documenting code, and why incident triage and chaotic legacy code break the loop.

Spot the silent cost leak in prompt caching caused by dynamic elements. Move the cache breakpoint to the end of the static prompt and monitor reads cost 100% vs 10%.

Explore what makes an effective AI coding assistant, from context-aware completions and debugging to refactoring, tests, and documentation, and weigh security, integration, and team workflows.

Discover top AI code assistant plugins for your IDE, including GitHub Copilot, Cody, Windsurf, and Gemini, offering onboarding, code comments, unit tests, and soc two type two compliance.

Gemini Code Assist offers an IDE plug-in that streamlines adding comments, implementing unit tests, and improving readability, integrating with GitHub/GitLab and automated PR reviews across enterprise tools.

Explore top ai assistant plugins for your ide, including GitHub Copilot, Gemini Code Assist, Windsurf, and Kodi. Onboarding highlights core capabilities like adding comments and generating unit tests.

Windsurf is an AI code assistant platform acquired by OpenAI, offering test generation, broad IDE support, and enterprise self-hosting. It lets developers choose from multiple providers for tailored performance.

Explore Zed, a high-performance Rust editor for speed and collaboration, with an integrated ai assistant and flexible model integration from Anthropic, OpenAI, Google, or Llama models.

Explore cursor, a vscode fork that deeply integrates ai into the development workflow with mode controls, context sharing, and background agents for code completion and chat.