AI won’t replace you, but a human with AI will

Here's what I built last week (espresso-fueled and very real):

We're no longer in the era of "I don't know how to do this." It's not about technical skills or mastering a specific tool anymore, it's about how you think. If something repeats in your workflow, or you need to solve a problem, you can absolutely build a solution around it. There is no excuse.

Here's what I did last week:

1. Podcast automation tool (non-commercial, but a sanity saver): I produce a podcast with a friend, I help with guest outreach, post-production, reels, YouTube uploads, etc. After a few episodes, I realized it was becoming repetitive. So I built an internal tool with Cursor AI (not a public-facing app, just raw utility).

Now I drop the audio file in, and it outputs:

→ Chapters
→ YouTube/Spotify descriptions
→ High-engagement moments with timestamps
→ Keywords
→ Even SEO copy and snippets

This used to take me hours. Now it's 5 minutes. I didn't even start with code. I just described my idea in ChatGPT, asked it to ask me clarifying questions, made it draft an implementation plan, then moved it into Cursor. We iterated until it worked. That took ~30 minutes. Now it saves me hours. No web interface, no design, just local workflow.

2. Researching podcast guests (at scale): We needed a list of potential podcast guests. I described the kind of people we want in ChatGPT. It gave me a prompt. I ran it in Perplexity Labs (great research tool), then did the same in Gemini, Claude, and ChatGPT.

Output? Over 2,000 names. Then I sorted everything in Cursor AI and formatted it into a clean spreadsheet. Next step is outreach automation.

Could I pay for a SaaS tool to do it? Sure. But I built my own. It works for my use case. And maybe in the future, this becomes a product.

Bottom line: If you have an idea, there are no blockers. You can build it. I'm currently exploring n8n to expand this system. It's an automation engine where you can also plug in agents.

Execution is no longer about knowing everything — it's about knowing what you want to build, and letting the machines help you get there.

Let me know if you want me to open source any of this.

P.S. If you’re curious what the podcast sounds like after I duct-taped AI to half the workflow — check it out here.

What Caught My Eye This Week

Looks like Cursor’s next update will include a native To-Do List system for agents.

I’ve actually been doing this manually since day one—had a markdown doc with tasks + specs, gave it to the agent, and made it update the plan line by line.

Why? Because agents forget. You don’t want them drifting into side quests.

With this new update, it seems they’re finally productizing that workflow. Your AI engineer now has a checklist—no more “wait, did we finish that feature?” moments.

How Anthropic Teams Actually Use Claude

Anthropic just dropped an internal doc on how their own teams use Claude Code. Forget “build me an app.” Claude isn’t your personal code fairy. Anthropic’s internal teams cracked how to actually make it useful:

• Security team feeds stack traces and gets fixes in 5 mins (instead of 15).

• Product devs use “auto-accept mode” — Claude writes, tests, commits. They review later.

• Inference team writes tests with edge cases, translates code across languages, learns ML on the fly.

• Growth team (1 person, 0 engineers) built ad dashboards + a Figma plugin that creates 100 ad creatives in seconds.

• Non-tech teams write plain text like “get data → export to Excel” — Claude runs the whole flow.

For solo builders, PMs, designers — this should reframe how you ask for help from AI.

Microsoft dropped a free AI agent course

The whole thing’s wrapped around one use case — building a travel agent that evolves into a multi-agent setup. Real infra, real stack (Semantic Kernel, AutoGen, etc).

If you’re even half-serious about working with agents in 2025, start here.

It’s not AI that will replace you, it’s the person who knows how to use it without crying over a broken prompt.

Motivation of the week (because AI is already demotivating)

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