# AI hype is diluting what actually matters

Canonical: https://brew.new/browse/templates/email/pt1_k97pqnpa3pqy7q86an6b85555s8e5x4q

Brand: vueschool.io
Category: newsletter

![Preview of AI hype is diluting what actually matters](https://cdn.brew.new/email-preview-2e0526f6297b1483-tracking_r57g23y9hg4p1p5h7ggkj3q9h18dva9d-1789016507254.png)

## Email content

Vue School website

Hi John,

Every week there's a new tool, a new protocol, a new "this changes everything" post. You're supposed to have an opinion on all of it. And if you're being honest, most of those opinions are based on other people's opinions, not actual understanding.

That's not on you. The AI hype cycle is moving faster than anyone can absorb, and it's doing more damage than the things it hypes. Developers aren't just confused. They're adopting things in the wrong order, without understanding what problem each tool actually solves for their workflow.

MCPs are the clearest example.

Anthropic releases Model Context Protocol as an open standard. "The USB-C port for AI!" Everyone rushes to build MCP servers.

Then the backlash hit. "It promised more than it delivered". "Forget MCP servers, just let agents run shell commands directly". And now most developers can't clearly explain what MCPs actually do, when a direct command is enough, or where plain APIs still win.

The fix isn't more hot takes. It's slowing down long enough to understand what you're actually using.

MCPs, CLIs (bash commands), and APIs aren't competing. They're three ways your agent gets things done. Same problem, different approach. Let me show you.

Say you want your agent to check what's failing in your CI pipeline…

With an API:

you write a fetch call to GitHub's REST endpoint

handle auth

parse the JSON

and extract the failing step

It works, but you wrote the integration yourself and you'll maintain it when the schema changes.

With an MCP:

you install a GitHub MCP server

your agent connects and sees tools like `actions_list`, `search_pull_requests`, and `get_job_logs`

picks the right tool for your question

No manual integration code. But the agent loaded every tool definition into context just to answer one question.

With a CLI:

your agent runs `gh run list --status failure`

One command. It already knows `gh` from training data, so there's no schema to load and nothing to configure.

Three approaches, same result. The API is manual work you maintain. The MCP is powerful but heavy for a simple question. The CLI is the obvious call here.

But what if you need to browse open PRs, check review status, AND see CI results in one go? Now the MCP makes sense because the interaction is richer than a single command. The CLI can't browse structured data the same way.

That's where the skill comes in: A skill file is a full procedure your agent follows automatically. Not a one-liner, but a workflow.

In the skill file, you talk to your agent and describe a full workflow of steps:

When checking CI failures, start with `gh run list --status failure`.

If the failure needs deeper context, connect to the GitHub MCP and pull the job logs, related PRs, and review status together.

Summarize what broke and why.

The agent doesn't just pick a tool. It follows a process, the same way you would, without you re-explaining it every time.

The procedure: turn your images on to see the details.

Without that layer, you're guessing. And that's the whole MCP backlash in one sentence. The tools were never the problem. Using them without knowing when they fit was.

Everyone is rushing to adopt the next thing. The developers actually shipping well are the ones who slowed down. They understand the layers, so they're not debugging tool choices they never should have made.

That's what we built Unlearn around. Not hype, not "this changes everything" posts, but the opposite. We take one piece of the AI toolchain at a time, strip the noise, and break it down until you have clear vision on what you're using and why.

This email is one example. Every workflow inside the platform works the same way.

The developers already inside are learning MCPs, rules files, and every other part of the AI toolchain the same way you just did in this email. We're closing enrollment today to give them our full attention and make sure the experience is incredible before we let anyone else in.

Less than 80 seats left. Early Access closes at 10:00 AM ET (16:00 CEST) today, and after that we're closing the door to focus on the members already inside.

Lock in $149/yr $299 →

Best,

Alex GS

Tech Education Lead at BitterBrains

vueschool.io

Facebook

X

GitHub

BlueSky

LinkedIn

YouTube

Instagram

TikTok

Vue School is proudly part of BitterBrains – Making Developers’ Lives Easier

Become a BitterBrains Community Supporter here.

Courses | Vue.js Master Class | Workshops | Tutorials | Plans

🔕 Not interested? Mute Unlearn Updates & Announcements here.

You received this email because you are subscribed to Marketing Information at thomas+vueschool@brew.marketing.

Manage email preferences

2093 Philadelphia Pike #9442, Claymont, Delaware 19703, United States

[Open and remix this design](https://brew.new/browse/templates/email/pt1_k97pqnpa3pqy7q86an6b85555s8e5x4q)

[Browse email designs](https://brew.new/browse/templates)