AI has zero taste

@angusthenontechnical

36.4K followers · usually 1.7K views a post

Views

12.3K

Likes

663

Comments

18

Shares

57

Selling · SaaS › AI Tool

The Loop

angussewell.com ↗

Why it worked

The video effectively uses a native mention strategy by explaining a complex AI concept ('The Loop') through clear visuals and relatable analogies, positioning human creativity as essential in the age of AI. The creator's personal take and energetic delivery make the content engaging and shareable.

Summary

The video explains that AI, specifically Claude, is moving beyond prompt engineering towards self-prompting systems. It highlights the importance of human input and creativity in building and managing these AI systems, emphasizing that human oversight remains crucial.

Structure

1Introduction to AI's limitations and the shift from prompt engineering
2Explanation of 'The Loop' system and its components
3Demonstration of how to build and manage AI systems with human input
4Emphasis on the continued necessity of human creativity and oversight
5Conclusion on the value of human involvement in AI development

Product placement

The video discusses and visually represents 'The Loop,' an AI system that improves itself, and mentions Claude by Anthropic. It also shows diagrams of databases and skills, implying a SaaS product for building and managing AI systems.

Original caption

stop trying to become a "prompt engineer" lol. even the team at anthropic says the goal is to build loops where the AI prompts itself. basically, you build the system, and the system handles the prompting. #claude #agenticai #techtok #LearnOnTikTok

More from @angusthenontechnical

Their other posts in the index, biggest breakout first.

Call to action

Try it for yourself.

On-screen text

AI has zero taste LEARN AI FASTER Anthropic says writing code is no longer the main job THE LOOP A system that improves itself HUMAN INPUT - what to change AGENT INPUT diagnostic agents flag problems THE DATABASE tracks issues, builds, workflows, skills SCOPE decide what matters PLAN design the build BUILD write it TEST prove it works new issues logged Every output becomes the next input. GIVE YOUR AI A DATABASE STORE YOUR DATA - CONNECT YOUR AI PICK WHERE YOUR DATA LIVES YOUR TABLE connected CLAUDE asks & answers in plain English reads + updates rows for you Your AI now works directly with your data. THE BUILD REGISTRY One database for everything you build TOOLS name stage WORKFLOWS name stage SKILLS name stage APPS name stage ISSUES problem points to Every tool, workflow and skill - one row each. THE SKILLS THAT RUN THE LOOP AGENTS THAT BUILD AND FIX STRATEGIZE plan what to build BUILD write the tool or workflow TEST prove it works HEALTH CHECK detECT WHAT BROKE Each skill does one job - extremely well. THE LOOP A system that improves itself HUMAN INPUT - what to change AGENT INPUT diagnostic agents flag problems THE DATABASE tracks issues, builds, workflows, skills SCOPE decide what matters PLAN design the build BUILD write it TEST prove it works new issues logged Every output becomes the next input. Nothing runs without me I AM THE LOOP YOU STRATEGIZE RUN BUILD RUN TEST RUN RESOLVE RUN I see the problem, I run the skill, I run the next.

Transcript

AI has zero taste. Turns out Claude just killed all of prompt engineering. Look at this clip up dude. This is Boris, the head of Claude. Boris says writing code is no longer the main job. You don't prompt Claude anymore. You build systems that prompt your AI for you. Most people shouldn't be doing that, dude. Sorry to break it to you. Try it for yourself. And see. Your AI now works directly with your data. One database for everything you build. Every tool, workflow and skill - one row each. The skills that run the loop. Agents that build and fix. Strategize, build, test, health check. Each skill does one job - extremely well. You can create these skills. You can have a series of cron jobs detect if there's anything needed to be changed. And then run each skill, scoping and planning the change, building the change, testing the change. And doing that again until it reaches the human goal. You can even build a cool front end like this. But I realized that I don't want my system to look like this. I want it looking like this. Nothing runs without me. I am the loop. You. I see the problem, I run the skill, I run the next. I can't fucking read your mind, bro. Especially if you're doing something go-to-market related. That requires so much human creativity. Yes, Boris definitely has things figured out that I don't even remotely have a software development. My use case is literally different. I don't see a world anytime soon where human creativity won't be absolutely necessary to have constantly in the loop. It's not a software development.