SecondBrain
Ask the Brain
Index/Sourceupdated Sat May 30 2026 08:00:00 GMT+0800 (Philippine Standard Time)

How To Use Claude Better Than 99% Of People (theMITmonk)

claudeai-toolingpromptingmcpprojectsbrand-foddersponsored-content

How To Use Claude Better Than 99% Of People (theMITmonk)

Third theMITmonk video (Sandeep Swadia) in this wiki. A practitioner walkthrough of Claude as a five-surface product stack (Chat / Projects / Cowork / Code / Chrome) plus the PRIME Framework for getting good output from any of them. First source in this vault that treats Claude as a product family rather than just the Claude Code CLI. Sponsored by Higgsfield (creative-AI tool) — disclosed inline.

The five-surface mental model

Sandeep's frame: don't think of Claude as one chatbot — think of it as a team of five working surfaces.

Surface Activity When to use
Claude Chat Think Messy problems, fuzzy ideas, rough drafts. Working through ambiguity.
Projects Remember When work spans multiple sessions and you want persistent context (files, instructions, tone) without re-priming each time.
Cowork (desktop app) Execute When the task needs local files, multiple steps, and output written back to your disk. Available on paid plans only.
Claude Code Build Build something. "If you can type in English, you don't need a computer science background to code and build something cool and useful."
Claude Chrome (extension) Browse Claude reads what you're looking at — tailoring outreach to job listings, pulling 3 insights from a 40-page industry report — and acts inside the browser workflow.

Underneath: Claude Skills (repeatable workflows) and MCP connectors (other tools the agent can call into).

Three "power moves" inside Chat

When the polished default response is too generic:

  1. Give it rich context — connect Google Drive, email, calendar, Notion. Attach files. "It will read it. It'll get context from it."
  2. Ground with research — ask Claude to do web research, then ask "is this verified?"
  3. Build the skill while flying the plane — open a parallel chat tab to learn the concept (IRR, free cash flow, spreadsheet feature) you need to use it. Do-and-learn loop.

Stronger prompt example (vs the typical "write me a LinkedIn post about leadership"):

"You are a business school professor. I am trying to explain why some leaders sound credible and others don't. I'm attaching my notion page with a rough idea... Your job is to push back on what's weak and sharpen what's real. Find me research that helps me create a strong insight and then we can write a LinkedIn post."

That's PRIME Framework in action without the acronym.

Projects — the "ongoing relationship" pattern

"A project is where you give the chatbot your files, your instructions, your tone, your ongoing work. You're not starting from zero every time you open it."

Worked example: a job-search project drops in the user's résumé, LinkedIn profile, target job descriptions, writing samples, plus a custom system instruction like "You're the world's most badass recruiter and resume writer... I write in a direct confident tone. I don't like corporate fillers...".

Sandeep's preference: customize Claude at the project level, not the account level. "Using AI to plan your vacation is very different from using it to plan your retirement." Per-context personas beat one global persona.

Cowork + MCP — the "team of agents" pattern

"Cowork is where it gets done. You start in a prompt window but finish it on your desktop."

Cowork's twist: MCP connectors = "giving Claude a set of keys to other AI tools." The video's worked example — a small business pluging Claude into Hicksfield (Higgsfield, video sponsor — disclosed) to generate 20 ad creatives for a candle product line in a single session that "would have taken you 3 weeks of a professional shoot, a photographer, a stylist, and weeks of post-production and of course thousands of dollars."

Two practical caveats:

  • The output is a starting point. "You'll still have to go through each and every slide and make sure it's good."
  • Sandboxing. Give Cowork access only to a specific subfolder. "Claude is very conscious about security, but please be very mindful about what you give access to."

Claude Code — vibe coding for the non-engineer

"If you can type in English, you don't need a computer science background to code and build something cool and useful."

Worked example: "I want a dashboard where I can drop in my sales pipeline and meeting notes and instantly see which deals are at risk, what's stuck, and what my next move should be."

This is Vibe Coding (Karpathy's term) in the IC-business-user register — the consulting-side-hustle owner using Claude Code to build operational tooling without learning to engineer. Adds a 5th source to the Vibe Coding cluster.

The PRIME Framework (canonical contribution)

Sandeep's 5-element prompting rubric — "the easiest way to direct Claude the way you would direct a smart person you've just hired":

  • Purpose — precise goal for the work
  • Research — external research to ground the response, or background material from the user
  • Interview — let Claude ask clarifying multiple-choice questions before responding
  • Mechanics — output shape (bullets vs paragraph, document vs table, concise vs detailed, strategic vs conversational)
  • Examples — what good looks like

See PRIME Framework for the dedicated page. The "interview" step is the distinctive piece — most prompt frameworks (including Sandeep's earlier AIM Protocol) don't surface clarifying-question-elicitation as a named step.

Why this matters to the vault

  • Fills the "Claude as product stack" gap. The vault's Claude Code page is CLI-centric (Boris Cherny's worldview); Sandeep's stack frame is the consumer/IC-business-user complement. The two views compose — Claude Code is one surface in a larger family.
  • MCP gets a sponsored worked example. CLI vs API vs MCP already discusses MCP's pros/cons in the engineering context; this video is the first to show what MCP looks like for non-engineer users (small-business owner connecting a creative tool).
  • Vibe Coding gets its 5th source. Sandeep's "type in English to build a dashboard" is the IC-business-user translation of Karpathy → Fu → Boris.
  • PRIME Framework is the vault's first 5-element prompting rubric. Augments AIM Protocol (drafting-focused) and Intelligent Hill (Prompting Camps) (technique-hierarchy).
  • First sponsored video in the cluster that's also brand-relevant for the user — Higgsfield's "$20-a-month creative agency" framing is on-thesis if the user ever covers the AI-democratization-of-marketing angle.

Editorial pattern (3rd theMITmonk source — pattern is now stable)

  • Solo monologue, 15–20 min
  • Named acronym framework (PRIME here, after AIM, TRAP, DRAG previously)
  • Worked-example heavy — every framework gets a paste-worthy prompt
  • Sponsor integration tight with framework (here: Hicksfield as MCP demo + sponsor)
  • Closing on the human note: "the real edge is the human using it" + "you were the one you've been waiting for"

Practical takeaways for this vault's user

  • Projects pattern is directly usable. The vault itself could move into a Claude Project (instructions = the CLAUDE.md schema; files = past wiki pages). Worth considering as an alternative to the current file-system + Claude Code setup if the user ever wants a web-only workflow.
  • PRIME as a coaching artifact. Cleaner than AIM for senior-leader audiences who need to delegate to AI well. Could be a near-ready Medium piece on "how senior leaders should prompt" if framed around the Interview step (the leadership-translation: ask AI to interview you = the same skill as briefing a smart new hire).
  • Conflict-of-interest note — the Higgsfield sponsor pitch is on-thesis advertising. If the user writes about MCP or AI-democratization-of-marketing, treat Higgsfield as one specific tool, not the category endpoint.

Cross-links

Source

  • Original transcript