Hola JARGONESE
This week in 30 seconds:
OpenAI is coming for GitHub. A new model just unified coding and creativity into one beast. And one founder ran $1M through an AI system he built in a week — across 4 companies, from his phone.
Here's what's inside:
OpenAI eyes GitHub's throne — Frustrated by outages, OpenAI is building its own code repository. The real question: will developers trust them with their source code?
GPT-5.4 is OpenAI's most complete model yet — Coding + personality + 1M context window, finally in one model. Benchmarks look great. Real-world results? More nuanced.
The AI Operating System running entire businesses — Forget one-off prompts. AIOS wraps AI around everything — data, tasks, decisions — and delivers your business briefing before you wake up.
Tool of the week: Napkin AI — Paste text, get presentation-ready diagrams in seconds. Free while in beta.
Prompts you can steal — 15 slash shortcuts that replace full sentences and make every AI response sharper.
The big picture this week:
AI is quietly moving from tool to infrastructure.
The founders building systems around it are pulling ahead fast. Everyone else is still copy-pasting prompts into a browser tab.
The gap is only going to widen.
Let's get into it. 👇
DEEP DIVE AI
OpenAI Is Building Its Own GitHub
OpenAI is reportedly developing an internal code repository to replace GitHub, frustrated by recent outages on the platform. The tool could eventually be sold to developers — and paired with OpenAI's Codex coding agent to stand out against GitHub Copilot. But the bigger play here isn't just reliability — it's reimagining how code gets managed in an AI-first world, where AI writes the code and humans oversee evolving systems.
Quick hits:
OpenAI is building internal source control, possibly launchable within a month
Integration with Codex could be the key differentiator over GitHub Copilot
Big companies may hesitate to upload source code to OpenAI's platform — trust is a hurdle
Microsoft, which owns GitHub and funds OpenAI, is quietly watching this rivalry unfold
GitHub's own former CEO just raised funding to reimagine code collaboration in the AI era
The AI Operating System Running 4 Businesses at Once
AIOS
Forget AI chatbots and one-off prompts — a new concept called an AI Operating System (AIOS) wraps an intelligent layer around your entire business. Built using Claude Code (no coding required), it connects your data, automates recurring tasks, and delivers a daily business briefing to your phone before you wake up. One founder used his to go from idea to $1M webinar launch in 7 days — solo.
Quick hits:
Layer 1 — Context: AI learns your business like a co-founder, no re-explaining every session
Layer 2 — Data: All dashboards (CRM, revenue, marketing) unified into one live view
Layer 3 — Intelligence: Auto-generated daily briefing with SWOT analysis across your business
Layer 4 — Automate: Recurring tasks (proposals, reports, follow-ups) get crossed off permanently
🚀 Layer 5 — Build: Freed-up bandwidth gets reinvested into new growth initiatives
GPT-5.4 Is Here — OpenAI's Most Complete Model Yet
OpenAI merged its best coding model (5.3 Codex) with its best general model (5.2) into one: GPT-5.4 — a unified flagship with a 1M token context window, native computer use, and strong agentic capabilities. On benchmarks like GDPVal and OSWorld, it edges past Claude Opus 4.6. But real-world tests tell a more nuanced story — Opus still wins on creative writing, teaching, and following instructions precisely.
Quick things :
GPT-5.4 = 5.2's personality + 5.3 Codex's coding, finally in one model
First general model with native computer use baked in — no workarounds
Beats Opus 4.6 on GDPVal (83% vs 78%) — real-world knowledge work benchmark
Opus 4.6 still wins on creative writing, explanations & instruction-following
Pricier than Opus at $2.50/M input tokens vs $5/M — but outputs cost more
🛠️TOOL OF THE WEEK
Generated on Napkin AI
ITurning research or data into presentation-ready visuals used to take hours. Napkin AI converts plain text into clean, customizable diagrams and figures in seconds — no design skills needed. Paste your content, pick a visual style, tweak colors and layout, and export as PNG or PDF. Currently free in beta.
What it does:
📝 Paste any text → instantly generates multiple diagram styles
🎨 Customize colors, line thickness, layout, and frame style
📤 Export as PNG, SVG, or PDF — high-res conference-ready
🔁 Works with any LLM output — run your content through ChatGPT first to get 3–4 crisp bullet points, then drop into Napkin
💸 Free while in beta (3 napkins free forever on basic plan)
Best for: Researchers, students, founders, and anyone building decks from dense content.
AI PROMPTS
ChatGPT prompt shortcuts
Most people waste time re-explaining what they want from AI on every single prompt. These slash shortcuts cut that friction entirely — just append them to any message and instantly control the tone, format, depth, or role of the response. Once you internalize a handful of these, your AI workflow gets noticeably faster.
/BRIEFLY → short answer, straight to the point
/TLDR → compress a long text into a few lines
/ELI5 → explain in simple, beginner-friendly language
/EXEC SUMMARY → high-level summary like you’re briefing a manager
/STEP-BY-STEP → break it down into steps
/CHECKLIST → turn the output into a checklist you can follow
/AUDIENCE → tailor the answer to a specific audience
/TONE → change the vibe (formal, funny, bold, calm, dramatic)
/FORMAT AS → force a structure (table, bullets, JSON, template)
/REWRITE AS → rewrite in a specific style (academic, casual, Gen Z, etc)
/ACT AS → make ChatGPT adopt a role (coach, editor, professor, PM)
/MULTI-PERSPECTIVE → show different viewpoints on the same topic
/PITFALLS → highlight common mistakes + what to avoid
/NO AUTOPILOT → ban generic filler and “safe” answers
/EVAL-SELF → ask it to critique its own response and improve it
Why this is the best:
One word replaces a full sentence — no more re-explaining what format you want
Works on any prompt — just append and go
Saves mental energy — less thinking about how to ask, more focus on what to ask
Consistent results — same shortcut, same behavior, every time
Stackable — combine
/BRIEFLY+/NO AUTOPILOTfor tight, honest answers
IMAGE PROMPT OF THE WEEK
Use this ready‑to‑paste prompt with Google Nano Banana Pro
Tweak the style/lighting according to your preference
Make an Ultra-wide fisheye lens perspective shot from ground level looking straight up at a young woman use image for face reference crouching mid-air between brick buildings forming a circular frame around the sky. She reaches one hand dramatically toward the camera, fingers spread wide, creating strong foreground distortion and depth. Dynamic action pose, confident expression, casual streetwear outfit: fitted white t-shirt, blue straight-leg jeans, red and white sneakers. Urban European neighborhood setting with brick houses, tiled roofs, windows visible around the edges. Bright daylight, clear blue sky, natural lighting, sharp focus, high detail, cinematic composition, exaggerated perspective, 8k resolution, photorealistic, adventure photography style.
Style: Cinematic photorealism — combines technical camera specs (fisheye, ground-level, 8K) with scene, subject, outfit, lighting, and mood in one clean block
Pro tip: Layer your details in this order → shot type → subject → pose → outfit → setting → lighting → quality tags. The more specific each layer, the less the AI has to guess — and the closer your output lands on the first try.
YOUR TURN
GPT-5.4 just closed the gap with Claude Opus 4.6 on nearly every benchmark — but real-world tests still hand the edge to Opus on writing, teaching, and following instructions.
Question of the Week:
Is benchmark performance becoming meaningless — and should we only trust real-world use case testing to pick our AI models?
Hit reply and let me know - I read every response!
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. The model that wins your benchmark might not be the model that wins your workflow.