Chat mode — one regulatory update applied to one P&P
What you’ll need to upload to the chat window: the regulatory update, saved in a machine-readable format like CSV, and the P&P you want to update. Upload both to an AI chat, and ask the AI to tell you where your P&P isn’t in line with regulatory requirements.
Is it really that easy? Yes and no.
AI and LLMs have hallucination problems. Your prompting and regulatory files need to be designed to mitigate against hallucinations. As part of this newsletter I put together machine-readable regulatory updates. The regulatory updates have assigned tracking capabilities to prevent hallucinations (see our full methodology — if you have access to Codex, Cowork, or Claude Code, I encourage you to copy and paste the methodology in, play around with making some of your own regulatory files, and have some learning-type fun).
Take a look at the regulatory update for FHA Loss Mitigation (this one already came out, so why not double-check your work). The regulatory update is a machine-readable CSV file (feel free to open it up and look around). Take that file, put your FHA Loss Mitigation P&P in the same chat, and use the prompt from the website to compare your P&P to applicable regulatory requirements. The prompt and the machine-readable regulatory files work in tandem to combat hallucinations.
Project mode — one change, many P&Ps
A Project is a bunch of chats linked together, where the chats can communicate with each other. Upload all of your P&Ps into one Project, one P&P per chat. With that setup, AI can read each policy and compare notes across all of them — so a single regulatory change becomes one coordinated review instead of forty separate errands.
Just like a single chat, that cross-chat communication needs infrastructure to combat hallucination — and at this scale, that infrastructure is a Skill. The Skill carries the verbatim-source discipline, the citation rules, and the output format, and enforces them identically in every chat (our methodology has a chapter on this; the discipline is called context engineering).
A textbook example of a good use for project mode is the Fannie Mae AI Lender Letter resource we provide (LL-2026-04). Since AI touches so many P&Ps, it makes sense to push the regulatory update to each P&P through project mode.
Again, take a look at our FNMA AI Lender Letter Survival Update Kit. The Skill itself includes the machine-readable regulations, plus the workflow to inventory the information you have, gather further information, draft your AI master policy, and then, based on the master policy, push changes back to the individual P&Ps.
Take a look, play around, and tell us what you think.
Which to reach for
| Chat mode | Project mode | |
|---|---|---|
| Best for | one new rule, one P&P | one change across many P&Ps |
| Setup | one chat, one combined upload | one P&P per chat, linked in a Project |
| Anti-hallucination | discipline in the prompt | discipline in a Skill, enforced everywhere |
| You get | redline + gap analysis for that policy | inventory, master policy, every P&P realigned |
| Example | FHA Loss-Mit kit → | FNMA AI kit → |
Start in chat mode — it’s where most P&P updates live, and it works anywhere you can upload two files. Graduate to project mode when one change reaches across your whole policy set and you need every P&P treated the same rigorous way.