👾 AI Nuggets #14: CEO Fired His Staff and Hired AI Instead

Welcome to edition #14 of the MadebyAgents newsletter. In this issue:

• OpenAI releases ChatGPT deep research

• ByteDance (TikTok) releases OmniHuman-1

• Google releases Gemini 2.0

• AI investment success of a small business owner

• AI-assisted coding to build software prototypes quickly

• Launch of the MadebyAgents YouTube channel

Latest News

🔬 ChatGPT Deep Research: The AI Analyst You Can Hire

OpenAI launched Deep Research - a ChatGPT tool that acts like a tireless research assistant. It scours hundreds of online sources, analyzes data, and generates detailed reports in 5 - 30 minutes. Designed for professionals in finance, tech, and startups, it costs $200/month (Pro tier) and handles complex tasks like market analysis or strategic planning.

What’s New

Deep Research uses OpenAI’s o3 model, trained to browse the web, interpret PDFs, and run Python code for data analysis. Unlike basic ChatGPT, it builds multi-step reasoning paths: pivoting searches, cross-referencing sources, and citing every claim. Reports include bullet points, tables, and source links for verification. Future updates will add charts and connect to databases.

It scored 26.6% on Humanity’s Last Exam - a test with 3,000 expert-level questions - outperforming GPT-4 and Google’s Gemini in accuracy - by far. But it’s not flawless: occasional errors or misjudged rumors still occur.

Business Impact

For small teams, this cuts research time from hours to minutes. Example: A founder received a 15,000-word marketing strategy report in 12 minutes, with only two minor data inconsistencies. Use cases include:

• Market analysis: Track trends, competitors, and pricing.

• Financial planning: Audit 10-K filings or model investment risks.

• Product decisions: Compare specs, reviews, and lifecycle costs for purchases.

Limits: 100 queries/month for Pro users, not for the mobile app yet, and outputs need fact-checking. But for $200/month, it’s cheaper than hiring a consultant.

🎥 ByteDance’s OmniHuman-1: AI That Animates Reality

ByteDance launched OmniHuman-1 - a tool that turns a single image into lifelike videos using audio, text or pose inputs. It animates humans, animals, and even cartoons with precise gestures and lip-syncing, rivaling OpenAI’s Sora and Google’s Lumiere. Not yet public, but demo videos show TED Talks by AI Einstein and singing avatars.

What’s New

OmniHuman-1 uses a Diffusion Transformer architecture and 18,700 hours of TikTok video data to train. It merges text, audio, and pose inputs into one model. Unlike older tools, it handles full-body motion, any image ratio, and weak signals like audio-only input. Outputs avoid robotic flaws - wrinkled clothes, stiff hands - common in AI videos.

It scored 9.1/10 for lip-sync accuracy and beat rivals in gesture realism tests. A single image can now create a dancing avatar or a CEO pitching in 12 languages. But ethics loom: deepfakes look real enough to fool detection tools.

Business Impact

For startups, this slashes video production costs. A bakery could animate its logo mascot for ads. A founder could film a pitch in minutes without cameras. Use cases:

• Marketing: Custom avatars for social media campaigns.

• Training: Animated instructors for online courses.

• Entertainment: Virtual influencers or game characters.

Risks exist. A scammer could clone a CEO’s voice and face. ByteDance promises safeguards if released, but regulations lag. For now, it’s a demo - no release date or pricing is available.

See OmniHuman-1’s talking Einstein demo here.

♊️ Gemini 2.0: Google's AI Gets Multimodal Superpowers

Google released Gemini 2.0, a major upgrade to its AI system. The new model generates text, images, and audio natively. It connects directly with Google Search and Maps. Early tests show better code generation and factual responses than its predecessor. Google plans to integrate it across its products, starting with AI Overviews in Search.

What's New

Gemini 2.0 brings three key advances. First, it creates images and audio without external tools - a leap from text-only models. Second, it uses Google's tools natively, searching the web and accessing maps in real time. Third, it has a reasoning mode. The release includes Gemini 2.0 Flash, a fast version for everyday use.

Business Impact

For businesses, this means faster and cheaper AI tools. Google's native tool integration cuts development time and costs. Watch for the rollout in Google Workspace and Cloud. Early access comes through Google's AI Studio and Vertex AI platform. No pricing details yet, but Google aims to compete with OpenAI's GPT-4.

AI Investments - What to Watch¹

📈 Why My Friend Fired His Staff and Hired AI Instead

Last week, I had a coffee chat with Chris. He runs a real estate business. The numbers were good, but his eyes looked tired. "I'm done hiring people who can't handle basic tasks," he said. He just automated his fifth workflow that month.

The Expensive Learning

Chris built his business the old way. He hired fast when work piled up. Then came the mistakes. One employee filed the wrong tax forms three months in a row. Another mixed up client records so badly it took weeks to fix. These errors cost him a lot last year alone.

"I used to think throwing more people at problems would fix them," he told me. "Now I know better." He showed me his phone. His new AI system had just processed 400 property documents without a single error. The same task used to take two full-time employees a week.

The Three Types of Employees

Chris splits his staff into three groups:

• A-players grow the business. They bring in deals and innovate.

• B-players keep things running. They're reliable but don't drive change.

• C-players drain resources. They make costly mistakes.

He learned to keep his A-players, support his B-players with AI, and replace his C-players with automation. "The math is simple," he said. "AI doesn't call in sick or make typos at 4 PM."

Start Here: The Smart Way to Automate

Chris shared his method:

1. Track where your team makes the most mistakes

2. List tasks that follow clear rules

3. Start with one small automation project

4. Measure the error rates before and after

5. Scale what works

He automated document processing first. Then tax filing. Then client communications. His error rates dropped by 90+%. His best employees now focus on growth instead of fixing mistakes.

"Here's what I wish someone had told me earlier," Chris said as we finished our coffee. "Look at your current tasks first. Fix what's broken. Then hire if you still need to. Most businesses hire first and optimize never."

AI Use Cases and Tools

👨🏼‍💻 Code at Lightning Speed: Everyone is a Programmer Now

Recently I built an AI-powered scraping dashboard in 1 week. 2 years ago, this would have taken a 5 person team two weeks. The secret? I paired with an AI coding assistant. I wrote the app documentation and focused on the core business logic. The result? Production-ready code with fewer bugs than my solo work.

The New Programming Stack

• CursorAI suggests whole functions and fixes bugs in real-time

• Replit makes even deploying super easy

• Claude Projects feels like the Google Docs of coding, always has the context

Democratization of Software

But here's the real shift: you don't need a computer science degree anymore. Now everyone can build simple tools. You can write in plain English, and the AI translates it to code.

Quick Tips to Start

1. Start with a simple app idea and easy AI tools

2. Learn to write clear instructions for AI

3. Keep security and testing in your control

The future isn't about replacing programmers. It's about multiplying what one person can build. Whether you're a seasoned developer, just starting, or about to hire one. AI tools are the new paradigm.

MadebyAgents Updates

🎉 I Finally Did It

I published my first YouTube video - and it's about something I'm truly passionate about: How AI is completely changing the way we collect web data.

Make your decisions based on real data with AI-powered web scraping

Traditional web scraping is dead. The new way to collect web data combines scraping with AI processing. I'm sharing:

• Analysis of 6 modern scraping tools

• Why AI is changing everything

• A look at my custom AI scraping dashboard

• Real comparisons of old vs. new methods

Watch the full breakdown:

That’s It for This Week

✨ Before You Go:

We’d love to hear what you think. Please share your opinion.

See you next time!

Tobias from MadebyAgents