Another week of AI chaos — and why prompting isn’t enough

The world is moving pretty fast right now and soon, we’ll have to think less about the technical stuff (GitHub, infra, whatever) and more about how to prompt. But here’s the kicker: to prompt well, you have to learn to think first — and not just jump into building mode.

That’s why I always advise starting with the basics: a Product Requirements Doc (call it whatever you want) — a simple doc where you outline your idea and the philosophy behind it, whether it’s a product, a feature, or just some workflow hack for your 9-5 so you can work less.

Everyone’s obsessed with building right now. Truth is that building is easy these days, but distribution, workflow design, actually shipping something useful — that’s where the game is won.

If you’re lost on where to start:

→ First, optimize your own life or workflow.
→ If it works for you, it’ll probably work for others.
→ Congrats, you’ve got a product idea (and a story to tell).

So in this issue, I’m sharing what I’ve been learning this week — what’s new in the AI space that can actually help us build and grow — plus, of course, a meme from your espresso-fueled solo founder in Italy.

And a personal thought: I can’t compete with the giant AI newsletters run by 8 people and an army of interns. But I can share real lessons from shipping solo, getting things live, and learning in public.

That’s the goal here.

What Caught My Eye This Week

Cursor 1.0 is out

Game-changing release. BugBot, Background Agents, Memories, one-click MCP installs, Jupyter support — and a ton of “small” but high-impact upgrades.

Here’s the quick version:

• BugBot → finds bugs in PRs, one-click fixes. (Finally, something for my chaotic GitHub tabs.)

• Memories (beta) → Cursor starts learning from how you code. Yes, creepy. Yes, useful.

• Background Agents → now async + Slack integration. You tag Cursor, it codes in the background. Wild.

• MCP one-click installs → no more ritual sacrifices to set them up.

• Jupyter support → big win for data/ML folks.

• Little things:

  • parallel edits

  • mermaid diagrams & markdown tables in chat

  • new Settings & Dashboard (track usage, models, team stats)

  • PDF parsing via Link & search (finally)

  • faster agent calls (parallel tool calls)

  • admin API for team usage & spend

Advice from Sequoia Capital

Sequoia Capital — one of the largest VCs in the world — recently dropped a 28-min keynote on building with AI. Here’s the distilled version (so you don’t have to watch the whole thing):

• The AI wave is bigger than cloud — eating both services and software.

• Adoption speed? Insane. Global attention + no distribution barriers.

• Application layer is where the $ lands — infra players will fight margin wars.

• Start with customer outcomes — not “let’s build an LLM.”

• This is a gold rush — timing matters more than polish.

• Early “vibe” metrics ≠ durable revenue. Look for real behavior shifts.

• Margins will migrate up — build towards outcome-based businesses.

• Data flywheel or die — your product must learn and compound.

• Agent economy is coming. Not ready yet — but multi-agent networks are the future.

• Building with AI is probabilistic, not deterministic — embrace the mess.

• Solo-founder leverage is rising fast. One-person unicorns? Possible.

More news

• OpenAI Codex + Google Jules → multi-step autonomous agents wired into your GitHub and workspace → debugging, refactoring, async code actions.

• Perplexity Labs → One of my favorite recent finds. Preplexity have launched Labs — basically, it’s a powerhouse for early product research: Ask about your niche, map the market, create a competitive analysis — all with actual graphs and insights. I feed that into GPT with my project ideas, and it’s like chatting with a power-brain. Works surprisingly well.

• Veo3 → Saw a crazy viral clip on X this week — AI-generated Bible scenes. We’re getting closer to a point where we won’t be able to tell real vs. generated. Huge opportunity here for creators and builders — distribution is key.

• Nucleus Embryo → wild experimental project. You can now choose traits/characteristics for your future kids (yes, really). Insane — and a glimpse into where biotech + AI is heading.

Meme of the week (because AI is already weird enough)

We’ve just crossed 3,000 subscribers — wild. Huge thanks if you’ve been here since Issue #1. That’s it for this week — stay caffeinated, ship fast, and I’ll see you next Friday.

If you’ve got feedback, or want to share what you’re building — just reply. I read every email (yes, even the weird ones).

—Miron