I «Hired» AI Agents v1.0 (GPT-4) - Coming For Your Job Next?

An early “AI agent employee” setup is now generating YouTube video concepts end-to-end—pulling fresh AI news, mining a creator’s own channel performance, debating ideas between two agents, and then producing both a proposed title/outline and a thumbnail image via Stable Diffusion—before emailing the results for review. The system’s core value is automation: it turns messy, unstructured inputs (news headlines and YouTube stats) into actionable creative decisions on a recurring basis, with the added twist that two AI agents critique and refine each other’s proposals.

The workflow starts when a collector script triggers two data-gathering branches. One branch searches Google News for AI-related headlines. The other connects to a YouTube channel through an API to collect details from the last 15 videos—titles, descriptions, likes, comments, views, and other performance signals. Those two streams are then fed into ChatGPT to produce a conversational summary, stored as text. That summary becomes the shared context for two named agents, “AI chris69” and “AI chris420,” which enter a structured back-and-forth discussion.

The agents’ goal is to converge on a “winning” YouTube idea for a specific target slot (the transcript uses “week 2 in May 2023” as the example). Their interaction is driven by system prompts that assign roles and enforce a workflow: one agent acts as a strategist focused on digital marketing and YouTube strategy, while the other participates in critique and iteration. After the agents agree on a concept, the system splits again. One branch sends the agreed thumbnail prompt to the Stable Diffusion API to generate an image. Another branch packages the thumbnail idea, the AI-news summary, and the proposed video plan into emails.

In testing, the first run produced a video concept centered on building an “AI-powered personal finance assistant” using GPT-4 prompt engineering and autonomous AI agents. The resulting deliverables included a title, a script outline with step-by-step development, and a thumbnail concept described as a laptop showing a personal finance dashboard with an AI bot. A separate email summarized current AI developments, including items such as a White House initiative to reduce AI risk, researchers transferring brain activity into words using LLMs, Microsoft’s AI-powered Bing, and concerns raised by Jeffrey Hinton after leaving Google.

A second test generated a noticeably different direction: a video about “AI in creative writing,” focusing on GPT-4’s impact on script writing and storytelling, with a split-screen thumbnail concept featuring a robot and an author holding pens. The transcript emphasizes that thumbnail outputs can be quirky—useful as prompts even when the first image idea isn’t perfect—while the broader creative pipeline still lands on coherent, publishable video structures.

Looking ahead, version 1.1 plans include an Azure Function to run daily autonomous updates, expanding beyond views and engagement into richer creator data (like earnings, requiring an API that may take weeks), pulling additional trend signals from sources such as Reddit and Twitter, and iterating system prompts to improve idea quality over time. The project is framed as a practical experiment in whether autonomous agents can function like “employees” for content ideation—delivering daily creative inputs via email and continuously refining what gets produced.