Let’s adopt shenanigans.

No. Looks clean, thanks for sharing. I find it interesting how some people embrace the word “vibe coding” while others consider it a curse word—the latter often distinguishing themselves from vibe coding by redefining it as shipping generated code to production.

I’m also considering notion’s MCP server for ingesting technical proposal documents. Each document defines a software project—its name, problem, scope, dependencies, components, endpoints etc. My intention is to connect VScode to this context for an agentic coding workflow.

@bretfisher.com some additional notes about my workshop experience.

In a workshop last week, I built a cli for desktop automation—it controlled the mouse & keyboard. Next, I made an MCP sever based on its functionality, exposing the important functions as MCP cmds. Then, I connected to the MCP sever and an LLM in a middleman script to automate a specific task.

The paper provides a formal framework for what AI practitioners have long suspected.

"A Conjecture on a Fundamental Trade-Off between Certainty and Scope in Symbolic and Generative AI"
Version 2
papers.ssrn.com/sol3...
3/3

On a serious note, is all brain engagement productive?

Miguel has a point but I'm optimistic. Painting didn't die when the camera was invented––it evolved. Photorealism emerged, abstraction flourished and new styles were born in response. This tech is in it's trail phase, in the future, we will develop new techniques inspired by these challenges.

5. Langchaingo is a great library for orchestrating an Agent's business logic, which may involve getting context from an MCP server, or chaining LLMs if desired. Langchaingo doesn't work with MCP by default––so you'll need an adapter.
github.com/dmahlow/lang...

4. An Agent is basically an LLM in a loop that has access to internal/external tools. It continues the loop until the task is complete.

3. MCP is like an agent's menu of available tools and context. It's just a unified API that exposes capabilities, such as access to files, databases or services. Setting up MCP servers in go is simple with mcp-go.
github.com/mark3labs/mc...

2. Cline is a nice plugin for agentic coding in VSCode. Honestly, I would prefer sticking with my JetBrains Goland IDE. JetBrains has the Junie coding agent––I hope it's worth the cost. cline.bot

1. Claude Sonnet 4 is expensive for a reason––it's coding quality exceeds Open AI GPT 4.1. Even when prompted to stay minimal the responses were noticeably more elaborate. I only spent $2.50 using GPT 4.1 for the 7 hour workshop while the person next to me racked up $9 using Sonnet 4.