AI Builds: Ground Zero

Every ad promises you can build an AI agent in 20 minutes. Ron spent over 12 hours building the simplest one he could imagine — a bankruptcy motion to extend time — and tested it across Claude, ChatGPT, Gemini, and Copilot to find out which platform actually delivers. The winner surprised him, and so did the lesson about what "regenerating a document" really costs you.

Listen/Watch On

Spotify • Apple Podcasts • YouTube

Show Notes

Are "build an AI agent in 20 minutes" ads lying to you?  Ron spent way more than 12 hours trying to find out.

Ron set out to build the simplest possible AI legal workflow — a motion to extend time to file bankruptcy schedules — and discovered that "easy" and "AI-assisted" don't always mean the same thing.

In this episode:

  • Why Ron decided to test AI workflow-building himself instead of buying a course

  • The 10-step workflow development cycle ChatGPT walked him through, from defining objectives to platform optimization

  • Testing the same workflow across Claude, ChatGPT, Gemini, and Copilot

  • Why Google Docs + Gemini unexpectedly produced the best output of any platform tested

  • Why Copilot's deep integration with Word actually made it harder to use, not easier

  • ChatGPT's pushback on regenerating an entire document — and the hospital-acquired-infection analogy Ron uses to explain AI "drift"

  • The lesson that mature AI users isolate and fix specific problems instead of regenerating whole documents

  • Why Ron avoided using Claude for this project due to a prior usage-limit experience on the Pro tier

  • The "airport test" from Ron's prior Field Note, Confessions of an AI Hallucinator, and how it applies to building workflows

  • The free "Motion to Extend Lite" workflow Ron is releasing publicly

We also discuss:

  • How persistent AI workflows (skills, agents, StrongSuit-style systems) represent a brand-new product category

  • The decision points Ron had to map for an "omnibus" extension workflow covering any deadline type

  • Quality control checks built into the workflow — missing dates, missing deadlines, missing cause language

  • How Ron accidentally discovered this entire approach through his own podcast post-production process

  • Why bare-bones bankruptcy petitions create the exact problem this workflow solves

  • The risk of inconsistent details (names, dates, captions) when reusing forms across cases — and how AI reduces that risk

Key Takeaway

Building a workflow that actually works — across platforms, with minimal user friction, passing real-world testing — is a fundamentally different job than writing a clever prompt. The platform you choose matters as much as what you ask it to do, and the AI tool that's "supposed" to be best for the job (Copilot in Word, Claude for heavy drafting) isn't always the one that delivers.

For Flintstones lawyers, this episode is proof that a usable AI workflow can exist without them building anything — Ron's free download does the work. Simpsons lawyers will recognize the platform-testing process as the real work of AI adoption. And Jetsons lawyers will appreciate the granular lesson on regeneration risk and isolating fixes rather than reprocessing entire documents.

Mentioned in This Episode

[email protected]