Building a Personal AI Agent for a TV Anchor in 4 Hours: How Claude Code Turned 21 Hours of Research Into 10 Minutes

The segment set out to build a personal 'do-my-job-for-me' AI agent for anchor Tsuyoshi Nojima in four hours, opening with the instructor simply saying 'let's launch Claude Code' and walking through the screen live. Nojima admits he already uses 34 different AI services day to day, but says none of them go beyond chat-style interaction into automating genuinely complex work.

The goal was to take the show's routine 'research and pitch-building' workflow and formalize it into a reusable Claude Code 'skill.'

After having Nojima verbalize his actual workflow -- researching pitch ideas, comparing angles, sourcing experts, booking them, and writing scripts -- the instructor asked Claude Code to 'draft an execution plan for building a skill.' Rather than accepting the AI's first proposal at face value, he stressed checking whether it's actually feasible, calling that verification step the key to avoiding a common trap.

A 'skill,' he explained, is essentially an instruction manual a human hands to an AI -- the AI-era counterpart to a program, which is an instruction set a human hands to a computer.

Seeing the generated pitch ideas -- including one on 'the US-China moon race' -- Nojima couldn't hide his shock that work which used to take 21 hours now took 10 minutes. The hosts contrasted this with ChatGPT and Gemini: cloud-based, browser-chat tools can't directly read local files and require manual uploads each time, whereas Claude Code, 'living inside the computer,' can read and write files directly and even run on a schedule.

In the four hours of filming, the segment went from building and running the pitch-generating agent to producing a web page that visualized the results.