Should you read the code, is RAG dead, and did Skills kill MCP?
Source
github.blog
Author
GPS
Date
Context
Skills and MCP address different parts of an agent's work. GitHub's explanation describes MCP (Model context windowThe maximum amount of text a model can consider at once — its working memory for the current conversation or task.Full definition → Protocol) as a standard connection to tools and data, while a agent skillA reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.Full definition → supplies instructions about a task, project or working method. Providing access does not, by itself, tell an agent how a particular team expects that access to be used.
Retrieval adds another function: finding relevant supporting information outside the model's training. A workflow can therefore use an MCP connection to reach a resource, a skill to guide the procedure, and retrieval to select the useful material. Adopting one does not make the others redundant.
Key takeaways · AI-distilled
GitHub's podcast hosts push back on 'you don't need to read AI-generated code': the reviewer is still accountable, so review depth should match risk (a production auth refactor vs. a CSS tweak), not treat every AI-generated change identically.
On AI and hiring, the argument is that companies increasingly ask candidates how they use AI, but the stronger signal is judgment: being able to explain when to use AI, when to work manually, and how you review generated code, not blanket adoption or refusal.
Skills didn't kill MCP: MCP standardizes how agents connect to tools and data, while Skills package process and context knowledge (often in readable Markdown) — the piece argues they solve different problems and compose together rather than compete.
RAG isn't dead, just less discussed: without good retrieval, models either rely on stale training data or spend extra tokens searching for context; groundingTying a model's answers to checkable sources — retrieved documents, live data, tool results — instead of letting it answer from memory alone.Full definition → retrieval narrows the search space and combines with agents, skills, and MCP in the same workflow.
On 'if you need to fine-tuningTaking a trained model and training it a bit more on your own examples so it gets better at one specific job.Full definition →, your code is bad': modern models have seen enough common patterns that struggling to parse a codebase is itself a maintainability signal — code confusing to a model is likely confusing to a new human teammate too.
Terms in this piece · Glossary
agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
grounding — Tying a model's answers to checkable sources — retrieved documents, live data, tool results — instead of letting it answer from memory alone.
fine-tuning — Taking a trained model and training it a bit more on your own examples so it gets better at one specific job.
