Building Expertise: Deliberate Learning for AI-Assisted Coding

Learning Opportunities is a plugin for Claude Code and Codex that integrates science-backed learning exercises into the AI-assisted coding workflow. It counteracts the passive habits encouraged by generative AI by prompting developers to engage in retrieval practice and architectural reflection. The tool ensures that as developers use AI to build projects, they are also building their own deep technical expertise.

Key Points

• AI coding tools risk decreasing long-term retention by creating a 'fluency illusion' where users mistake generated code for their own understanding.
• The skill uses evidence-based techniques like prediction, generation, and retrieval practice to force active mental processing during the development flow.
• Exercises are designed to be interactive and intentionally 'slow,' requiring the user to provide input before the AI provides answers.
• The tool includes a 'Measure This' playbook to help organizations quantify the impact of learning interventions on developer thriving and team effectiveness.
• It provides specific modules for repo orientation, helping developers strategically sample and comprehend new codebases based on empirical research.

Sentiment

The Hacker News community is broadly supportive of the core idea that AI-assisted coding creates deskilling risks and that deliberate learning interventions are needed. However, there is notable skepticism about this particular implementation — commenters question the lack of demos, evals, and measurable outcomes. The discussion is more enthusiastic about the general concept of skills and deliberate practice than about this specific tool.

In Agreement

• The generation effect is real — passively accepting AI-generated code skips the active processing that builds understanding, leading to skill atrophy
• Developers experience tangible 'skill debt' when blindly accepting agent output, losing familiarity with their own codebase
• Short, targeted learning exercises during coding sessions can help counteract deskilling from AI assistance
• Skills as structured workflow abstractions are genuinely useful for capturing repeatable processes and maintaining consistency
• Building your own skills incrementally is more empowering than consuming pre-built ones, as it forces deeper understanding
• The value lies in seeing how others build with AI tools and adapting those patterns to your own workflow

Opposed

• The skill amounts to an overly decorated prompt in a bash script — the actual functional content is minimal compared to the repository's complexity
• Without benchmarks, evals, or sample output, there is no way to verify the skill produces better results than alternatives
• Skills are less reliable than clear instructions in AGENTS.md — they get skipped too often by agents to be dependable
• The project lacks a demo or example output, making it hard to evaluate without installing and running it yourself
• Outcome tracking is the missing piece — without measuring which exercises actually closed learning gaps, 'deliberate' practice is just aspirational