In scaffold · Q3 2026

Vibe Coding with KAOS

An AI Coding Guide for Legal and Finance Professionals

A working guide for lawyers, paralegals, analysts, compliance officers, and legal-ops managers who already use Claude Code or a similar agent and want to push it further than chat. The book walks the reader from a first useful win, sorting a folder of PDFs into contracts, memos, and court filings, through the failure modes that bite real practice, into a partner-reviewable diligence pass, an intake bot with a conflicts check and UPL disclaimer, and the ABA Model Rules questions that come with running an AI inside a matter. Every chapter ends with commands the reader runs themselves and a three-item "what you verified" checklist.

What the reader can do at the end of each chapter:

  • Chapter 1 — classify a folder of PDFs into contract, memo, court-filing, and other, then visually diff against a ground-truth folder.
  • Chapter 3 — produce a citation that looks correct, then unmask it as fabricated using kaos-citations extract and kaos-citations verify.
  • Chapter 5 — turn a deal-room of mixed PDFs, Word, and Excel plus the target's most recent 10-K from EDGAR into one queryable corpus that answers a grounded question with citations.
  • Chapter 6 — run a six-category buy-side rubric (change of control, IP assignment, non-compete, indemnification, limitation of liability, termination) across twenty contracts in one sitting and produce a partner-reviewable flagged-issues CSV with random-sample-audited verdicts.
  • Chapter 8 — stand up an intake bot for one practice area with a conflicts-check gate, a jurisdiction-appropriate UPL disclaimer, and a transcript-to-record audit step.
  • Chapter 10 — defend the work to a partner under the ABA Model Rules and the named jurisdictional opinions (NY 2024-1, CA COPRAC, DC 388, FL 24-1, plus the federal standing orders) using the audit trail and the engagement-letter template in Appendix C.
  • Chapter 11 — turn a one-off conversation into a recipe a colleague can run on their own corpus without losing the verification habit.

Bommarito · Katz · Bommarito · Q3 2026 · outline available

The arc through the chapters

A first useful win in chapter 1 sets the habit. Five core capabilities in chapter 4 give the reader the moves. Chapter 5 turns a deal-room of mixed PDF, Word, and Excel into one queryable corpus. Chapter 9 makes verification the default. Chapter 11 turns a one-off conversation into something a colleague can run.

First Win CH 01 Five Capabilities CH 04 Documents + Data CH 05 Trust + Verify CH 09 Script → System CH 11 classify a folder of PDFs visual diff vs ground truth classify · extract summarize · compare redline mixed PDF, DOCX, XLSX plus EDGAR 10-K one queryable corpus random-sample audit cited-quote check red-team habit turn a chat into a recipe a colleague can run it OUTCOME

The full chapter table of contents and a sample chapter will appear here as the book moves out of scaffold. The conceptual reading the book builds on is at ai4lf.com.