well... it happened

hey friends — Miron here

this issue is in partnership with the Fundraising Bootcamp led by my friends Dasha & Matviy. If you’re raising in the next 3–6 months and tired of vague “come back later,” this is the most tactical program I’ve seen: live deck rewrites, momentum playbook without warm intros, and real pitches to funds (Khosla, Geek Ventures, Berkeley SkyDeck, Unshackled, Antler, more). Cohort is starting—use code MIRON10 for 10% off.

Now, onto the thing everyone’s asking me about: what GPT-5 + the gpt-oss drop actually change for builders. Also, OpenAI quietly shipped a proper “how to actually use GPT-5” hub.

Fundraising Bootcamp

Why I said yes to partnering: it’s practical. You get live deck tear-downs, outreach that works without warm intros, and real VC rooms—Khosla Ventures, Geek Ventures, Berkeley SkyDeck, Unshackled, Antler, others. If you’re raising soon and want less guessing, jump in. Use code MIRON10 for 10% off:

thefundraisingbootcamp.com/course

THE SHORT VERSION

I used to split my stack: plan with o3 in Cursor, then code with Claude Sonnet. After a week of pushing GPT-5, I can plan and build in one place. It’s faster, less fussy, and good enough that I’m not bouncing between tabs. Not magic, just smoother.

The real game isn’t “whose model is smarter” anymore. It’s who ships the best ecosystem that moves your idea to working product with the least friction—builder UX, tools, agents, integrations, cost, and speed. Quality will keep creeping up; the near-term edge is throughput and reliability.

WHAT I TESTED / NOTICED

• Code gen: GPT-5 feels 30–50% snappier than GPT-4o/Mini-High in practice. It handled multi-file edits, added missing features without hand-holding, and didn’t choke on follow-ups.

• Structure and writing: better flow and briefs. Ask for a “script” and it returns a mini-production plan with scene directions, not just paragraphs. Less hand waving.

• Images/design: for technical prompts (mockups, layout-precise assets) it’s clean and literal. For raw style or unexpected angles, 4o still has more “soul.”

• Fewer dumb mistakes: hallucinations are down. Not zero, but I’m not babysitting every line.

A few public notes lined up with what I saw: people calling it an “iPhone 5 → 5S” upgrade, a stronger creative partner with better taste, and unified model behavior that “just does the thing.” I agree. It’s not a moon-landing, but it quietly removes friction.

THE OTHER SHOE: OPEN-WEIGHTS GPT-OSS

OpenAI also dropped open-weights (gpt-oss-120B and 20B) with Apache-2.0. Headlines: runs locally, offline by default; the 20B tier is laptop-friendly; fine-tuning allowed; integration paths across HF/Ollama/vLLM/TensorRT/PyTorch; and the benchmark chatter says 120B hits o3-class reasoning and 20B beats o3-mini on coding/math. Translation: regulated and privacy-sensitive workloads just got viable without cloud dependency. Expect a lot of “runs on your GPU, no API bill” tooling.

GitHub link

OpenAI Cookbook: GPT-5 guides are live

OpenAI quietly shipped a proper “how to actually use GPT-5” hub. It’s not fluff—there’s a prompting guide, new params/tools, prompt-migration notes, and an end-to-end frontend example.

Why it matters

• Faster ramp: copy the patterns, don’t reverse-engineer them.

• Migration help: what changes from GPT-4 → GPT-5 (thinking controls, function/tool calling, response objects).

• Shipping fuel: a ready frontend coding example you can lift into your stack.

What to read first

• GPT-5 Prompting Guide

• GPT-5 New Params & Tools

• Frontend coding with GPT-5

• GPT-5 Prompt Migration (what breaks, what improves)

Link

• OpenAI Cookbook → GPT-5 section: https://cookbook.openai.com/

ECOSYSTEM MOVES

Lovable and Cursor already wired GPT-5 into a “build me a SaaS” flow—auth, payments, admin, email automation. Demo reels show two-minute, ~700-line apps from a single prompt. Ignore the hype gloss; the direction is right: multi-component builders, not code snippets.

Strategy note I buy: speed is survival. Clones arrive in days, not months. Moats shift from single features to platform breadth, distribution, and community. If you’re building thin wrappers, the water’s rising. If you’re building fundamentals—workflows that own a pain with or without model swaps—this helps you, not hurts you.

WHAT I’M CHANGING

• Default to GPT-5 for both planning and coding. Keep Claude on the bench for creative swing tasks where tone/style matter.

• Design for swap-ability: abstract model calls so I can flip between cloud and local gpt-oss when cost/privacy demand it.

• Optimize the loop, not the prompt: fewer steps from idea → deployed. Scripts, scaffolds, CI.

• Focus on products that get better from speed/cost curves (ops tools, agents tied to data, B2B workflows). Leave pure novelty to Twitter.

If you want a deep dive on any of this—how I set up the project scaffold, or a clean “local vs cloud” stack—I can write it next week.

SCENES FROM THE WEEK

• Honest GPT-5 reviews: faster, more reliable reasoning; tighter creative tone; unified model that “just picks the right tools.” Still, 4o feels more alive for artsy images.

• Platform angle: “Build me a todo app” → “Build me a SaaS with auth, payments, admin, email.” The gap between idea and working product keeps shrinking.

• Open-weights ripple: regulated/edge use cases go from theory to plan. Expect on-prem copilots and serious enterprise privacy pitches.

• Safety reality check: the wider and faster we go, the more guardrails matter. Treat models like junior devs with sharp objects—review, test, log.

🎟️ VIBE CODERS CIRCLE (15 seats are already taken)

I’m opening a private room for 50 builders and 15 seats are already taken. Monthly live teardown calls, 24/7 focused chat (no noise), lifetime access to every current and future playbook, and member-only agent/prompt recipes. If you want a place where people ship in public and keep each other honest, this is it. Join monthly or yearly; early codes went to the first wave—if you need one, hit reply and tell me what you’re building.

more coming next week.
until then:

stay caffeinated.
lead the machine.
launch anyway.

☀️ Miron