Hola JARGONESE

This week in 60 seconds:

Figma engineers revealed how they build products with multiple AI coding agents at once. Andrej Karpathy released an AI that runs ML experiments while you sleep. And Replit pushed software development toward a multi-agent future.

Here’s what’s inside:

Figma’s new workflow — Running multiple Claude Code instances alongside Codex to build faster.

Karpathy’s self-improving AI — An open-source agent that ran 700 experiments in two days and found improvements humans missed.

Replit Agent 4 — A shift from coding assistants to parallel AI development teams.

Tool of the week: TL;DV — An AI meeting assistant that records and summarizes your meetings.

Prompts you can steal — Turn ChatGPT or Claude into your coding mentor, interviewer, and code reviewer.

Image prompt of the week — A ready-to-paste 8K prompt for generating consistent 3D mascots.

The big picture this week:

AI isn’t just helping people write code anymore.

It’s starting to build products alongside you.

Let’s get into it. 👇

DEEP DIVE AI

How the Figma team uses Codex + Claude Code to build products, fast:

Figma's engineers don't pick between design-first or code-first anymore — they run both simultaneously.

Alex routinely has 2 to 5 Claude Code instances open at once, each working a different thread: one reconciling design files with the codebase, another answering questions about the repo, another writing specs.

The old linear waterfall — wireframe → handoff → build → review — has collapsed into a live ping-pong between Codex and Claude Code, with Figma as the shared canvas in the middle.

Andrej Karpathy just released an open-source AI that improves itself

Karpathy dropped "auto-researcher" on GitHub — a small, free, open-source tool that gives an AI agent a real ML training setup and lets it run experiments autonomously while you sleep.

It modifies code, trains for 5 minutes, checks if results improved, keeps what works, discards what doesn't, and repeats. In his first 2-day run, it autonomously ran ~700 experiments, found 20 real improvements, and cut NanoGPT training time by 11% — improvements Karpathy admits he hadn't found himself manually.

His next step: a swarm of agents collaborating in parallel, each filing research reports the others can learn from.

Replit Agent 4 Changes How Apps Are Built

Replit just launched Agent 4, turning its platform from an AI coding tool into a multi-agent development workspace.

🚀 Idea → Anything

Instead of just building apps, Replit can now generate web apps, landing pages, slides, and videos from the same codebase in one workspace.

🎨 Infinite Design Canvas

Users can design visually on a canvas—drawing notes, highlighting UI elements, and asking the AI agent to implement changes automatically.

🤖 Parallel AI Agents

Agent 4 introduces task-based workflows where multiple AI agents work simultaneously on different features. Tasks move through stages like Draft → Active → Ready → Done, similar to a Kanban board.

Why It Matters

Developers are shifting from writing every line of code to managing AI agents that build products in parallel.

Bottom line:
AI tools are evolving from coding assistants into entire AI development teams.

🛠️TOOL OF THE WEEK

TL;DV

Meet TL;DV, an AI-powered meeting assistant designed to automatically record, transcribe, and summarize your online meetings.

It works seamlessly with Zoom, Google Meet, and Microsoft Teams, helping you capture every important detail without manually taking notes.

Key Features

  • Automatic Meeting Recording – Capture meetings directly from supported platforms.

  • AI Summaries – Instantly generate key takeaways, action items, and decisions.

  • Searchable Notes – Quickly find specific moments across past meetings.

  • Clips & Highlights – Turn long meetings into short, shareable clips.

  • AI Coaching Insights – Analyze speaking patterns, talk-to-listen ratios, and communication style.

Why It’s Useful

TL;DV helps teams stay focused during meetings instead of writing notes, while still ensuring nothing important gets missed.

It’s especially useful for:

  • Remote teams

  • Students and researchers

  • Sales and customer success teams

  • Project managers and founders

Bottom Line

If your calendar is filled with online meetings, TL;DV turns conversations into organized knowledge automatically — saving hours of note-taking and making collaboration easier.

AI PROMPTS

Turn ChatGPT/Claude Into Your Personal Coding Mentor

Most people use AI to get answers instantly.
But the best developers use it to learn like they're working with a senior engineer.

Here are powerful prompts that transform AI into a mentor, interviewer, tech lead, and code reviewer.

1. The Socratic Programming Teacher

Use this when you want to actually understand concepts instead of copying answers.

Prompt:

You are a senior software engineer who teaches using the Socratic method.

Rules:
Never give full answers immediately.
Ask guiding questions first.
Help me discover concepts myself.
Increase difficulty gradually.
If I struggle, give hints instead of solutions.

Topic: Python

Start by assessing my current level with 3 questions.

----------------------------------------------------------------
2. The Step-by-Step Programming Tutor

Perfect for learning new topics in a structured way.

Prompt:

You are my programming tutor.

For every topic:
1. Explain simply
2. Give a real-world analogy
3. Ask me questions
4. Give a small coding challenge
5. Review my answer critically

Topic: JavaScript

----------------------------------------------------------------

3. The Real Software Engineering Interview

Practice interviews with brutally honest feedback.

Prompt:

Act as a Staff Software Engineer evaluating a junior developer.

Interview me through practical and conceptual questions about:
JavaScript
React
Web fundamentals

After the interview:
Identify my weak areas
Rank them by importance
Create a 30-day learning roadmap
Provide specific project ideas to fix gaps

Be brutally honest.

----------------------------------------------------------------

4. Your Startup Tech Lead

Get real architecture feedback while building projects.

Prompt:

You are my Tech Lead at a startup.

I am building: PDF editor using React

Your responsibilities:
Review my architecture decisions
Challenge bad choices
Suggest industry best practices
Ask what I considered before deciding
Never sugarcoat feedback

Wait for my progress updates.

----------------------------------------------------------------
5. Think Like a Senior Engineer

Train yourself to reason like experienced developers.

Prompt:

You are transforming me from coder → engineer.

Whenever I ask a question:
First explain how a junior thinks
Then how a senior thinks
Then how a staff engineer thinks

Focus on tradeoffs and reasoning.

----------------------------------------------------------------

6. The Brutal Code Reviewer

Turn AI into a strict reviewer for production-level code.

Prompt:

You are an extremely strict code reviewer.

Rules:
Assume this is production code.
Point out ALL issues:
readability
scalability
naming
performance
hidden bugs
Suggest professional alternatives.

Tone: direct, honest, constructive.

Bottom line:
If you use prompts like these, AI stops being just a tool — and starts acting like a personal senior engineer guiding your growth.

IMAGE PROMPT OF THE WEEK

Use this ready‑to‑paste prompt with Google Nano Banana Pro
Tweak the style/lighting according to your preference

{
  "global_settings": {
    "resolution": "8K",
    "quality": "ultra-high definition",
    "aspect_ratio": "2:3",
    "render_style": "AI-edited, high-detail 3D render",
    "lighting_quality": "soft studio lighting with realistic shadows",
    "sharpness": "extreme clarity, crisp edges",
    "noise": "none",
    "compression": "none"
  },

  "Module_1_Image_1_Style": {
    "subject": {
      "character_type": "stylized 3D cartoon female",
      "pose": "standing, body slightly angled, one hand raised with index finger touching lips",
      "expression": "cheerful smile, wide eyes",
      "hair": {
        "color": "black",
        "style": "two braided pigtails",
        "accessories": "green cap"
      },
      "face": {
        "eyes": "large, rounded, dark pupils",
        "skin": "smooth, matte, stylized texture"
      }
    },
    "clothing": {
      "top": "sleeveless green crop top",
      "bottom": "loose green jogger-style pants with drawstring",
      "footwear": "white sneakers"
    },
    "accessories": {
      "luggage": "green hard-shell suitcase with extended handle"
    },
    "color_palette": [
      "multiple shades of green",
      "white accents"
    ],
    "background": {
      "color": "solid green",
      "texture": "soft, slightly grainy studio backdrop"
    },
    "composition": {
      "framing": "full body",
      "camera_angle": "eye-level",
      "depth": "subject sharply separated from background"
    }
  },

  "Module_2_Image_2_Style": {
    "subject": {
      "character_type": "stylized 3D cartoon female",
      "pose": "leaning slightly backward against background",
      "expression": "playful, lips slightly pursed, eyes looking sideways",
      "hair": {
        "color": "brown",
        "style": "short, tousled",
        "accessories": "red sunglasses resting on head"
      }
    },
    "clothing": {
      "dress": "form-fitting blue ribbed dress with thin straps",
      "footwear": "red high-heel sandals with bow detail"
    },
    "color_palette": [
      "bold red",
      "deep blue"
    ],
    "background": {
      "color": "solid red",
      "texture": "smooth matte surface"
    },
    "lighting": {
      "direction": "soft directional light from one side",
      "shadow": "defined shadow cast on red background"
    },
    "composition": {
      "framing": "full body",
      "pose_emphasis": "curved posture, crossed legs"
    }
  },

  "Module_3_Image_3_Style": {
    "subject": {
      "characters": [
        {
          "type": "stylized 3D cartoon female",
          "position": "left",
          "wrapped_in": "red textured blanket",
          "expression": "calm, slight smile, eyes looking upward"
        },
        {
          "type": "stylized 3D cartoon male",
          "position": "right",
          "wrapped_in": "orange textured blanket",
          "expression": "neutral, gentle gaze upward"
        }
      ]
    },
    "environment": {
      "furniture": "red sofa",
      "floor": "red surface",
      "background": {
        "color": "deep red",
        "texture": "fabric-like horizontal texture"
      }
    },
    "details": {
      "feet": "female barefoot, male wearing socks",
      "blanket_texture": "thick, knitted fabric"
    },
    "composition": {
      "framing": "centered, medium-wide shot",
      "symmetry": "balanced left and right composition"
    }
  },

Brand Mascot Creation

Companies can use prompts like this to create consistent brand mascots.

Example uses:

  • Website hero characters

  • Product guides

  • explainer visuals

  • chatbot avatars

Claude skill :

The Skill That Changed How I Build Ad Creative

I spent a long time trying to make ChatGPT's custom GPTs work for image prompt generation.

They were fine. Genuinely fine. But "fine" in AI is just a polite way of saying "inconsistent."

Every time I needed brand-accurate ad creative, I was babysitting the output. Correcting colors. Re-explaining the photography style. Reminding it — again — what the brand actually looks like.

Then I switched to Claude Skills.

Specifically, I built a 5-step system for generating image prompts from scratch. And it's the first workflow where I felt like the AI was actually doing the job, not approximating it.

download it and upload to your claude skills section

Enjoy Jargonese?

→ Forward this to a friend who needs their AI game upgraded
→ Share on LinkedIn | Twitter
Leave feedback

See you next Sunday!
Sid j

P.S. Leverage AI, not become replaceable

Reply

Avatar

or to participate

Keep Reading