AI Agents Find Lost Cities, Creators Find New Hooks

What creators are doing

YouTube912 points, 335 comments

First-person AI discovery essay: a solo creator documents using an AI agent to find or verify something that experts said was hard or impossible, then writes it up as a personal narrative with screenshots and coordinates. The Maya site post follows this structure exactly — personal backstory, the tool used, the specific finding, the verification step.

Why it works: Specificity creates credibility. 'My AI agent found an unmapped Maya site' is a falsifiable claim with a real location — readers stay to verify, not just consume.

Copy this

Pick one niche dataset you have access to (property records, old menus, flight paths, local history archives) and document your AI agent working through it. Lead with the finding, not the process.

Make this today

Make a 'My AI Agent Found Something' Short-Form Video in Under an Hour

1. 1

   Pick your dataset and run a quick agent query15 min

   Choose a dataset you already have access to — local planning records, your own analytics, a public API, a spreadsheet of old data. Run a simple AI agent query (ChatGPT with browsing, Claude with a file, or a basic Python script) and screenshot the output. You need one specific, verifiable finding — a number, a name, a location, a date.

2. 2

   Write your hook sentence5 min

   Follow this template exactly: 'My AI agent [did the thing] and found [specific verifiable claim].' Example: 'My AI agent scraped 10 years of local planning applications and found the one street where every new build gets rejected.' The claim must be falsifiable — that's what makes people stay.

3. 3

   Film or record your screen in one take10 min

   For a talking-head video: sit down, say the hook sentence, show the screenshot on screen, explain the finding in under 60 seconds, end with 'link in bio to see the full data.' For a faceless version: screen-record your terminal or browser with a voiceover reading the hook and finding. No intro, no 'hey guys'.

4. 4

   Edit and add captions in makerzz10 min

   Drop the raw take into makerzz, select your house style, let it write the caption using the hook sentence as the first line. Check that the caption includes the specific claim from your hook — that's what gets shared as a screenshot on X.

Platform signals worth knowing today

Tools that change what you can make today

Scripts to post today

TikTok

United Airlines Bans AI Agents — Here's Why

“United Airlines just banned AI agents from booking flights on your behalf. This is real.”

United Airlines just banned AI agents from booking flights on your behalf. This is real — it's in their terms of service as of this week. Here's why they did it: AI agents don't accept liability. When a human books a flight and something goes wrong, there's a paper trail. When an agent books it, the airline can't charge change fees, can't enforce cancellation policies, and can't hold anyone accountable for the booking decision. So they banned the agent, not the AI. The irony is that every airline app already uses AI to price your ticket against you in real time. They just don't want AI working for you. That's the whole story.

Why it works: Opens with a falsifiable claim, explains the real reason behind the policy (not the stated reason), and ends with a pointed observation — the contrast between AI-for-them and AI-for-you is the share trigger.

Make a video about this

LinkedIn

What Tim Peters Taught Me Without Knowing It

Profiles trending right now

pietromiotti.comYouTube

Posted a first-person account of using AI agents to locate an unmapped Maya archaeological site, earning 912 Hacker News points and 335 comments. The post is being widely shared as a template for 'AI-assisted discovery' content.

Steal this: The structure: personal backstory (2 sentences), tool named explicitly, specific verifiable finding, verification step, open question for readers. Use this frame for any niche dataset you have access to.

shadcnGitHub

The shadcn/improve repo hit 361 GitHub stars trending with a tool that uses a capable model to audit codebases and write execution plans for cheaper models — a workflow that directly addresses the cost problem in AI-assisted development.

Steal this: The product positioning: 'use your most capable model to plan, your cheapest model to execute' is a content angle that works as a tutorial, a thread, or a short — it solves a real cost problem with a concrete workflow.

addyosmaniGitHub

The addyosmani/agent-skills repo hit 787 GitHub stars trending with a collection of production-grade engineering skills for AI coding agents — a resource that is being shared heavily in developer communities as a practical reference.

Steal this: The 'curated skills list' format — a public repo of prompts, skills or templates for a specific use case is a content asset that compounds over time and drives followers back to your profile every time it's updated.

What to do today

• Film a 45-second 'my AI agent found something' video using any public dataset you have access to — use the hook template from the playbook and post to TikTok and YouTube Shorts today.
• Write a 4-tweet thread on the Jensen Huang story with a clear position in tweet 1 — post it within the next 2 hours while the story is still in the first news cycle.
• Make a 5-slide Instagram carousel comparing the 'sure' open-source personal finance app to one paid app your audience uses — lead with the cost saving, end with the GitHub link.
• Screen-record 20 seconds of the WebGL sand demo at av.github.io/safe-sand, add a one-line voiceover, and post as a YouTube Short with the URL in the description.
• Search 'how to [any Apple feature]' on Google right now — if the top results are Apple Support Community threads (which close November 20), that is your video topic for this week.