Teaching AI to Remember

Dear Negotiation Explorer,

Welcome to issue 30 of our NegoAI series.

Last week, I shared a folder system for context engineering — four folders that make every AI negotiation session better. If you built it (even a basic version), you already felt the difference.

But here's what I didn't tell you: that system has a limitation.

It's static. And more importantly — your AI still doesn't remember anything.

Here's a truth that most AI advice glosses over: large language models have no memory.

None. Zero. Every time you open a new chat — ChatGPT, Claude, Gemini, doesn't matter — the AI knows nothing about you. Not your negotiation style. Not the counterparty you've been dealing with for six months. Not the brilliant analysis it gave you yesterday.

This isn't a bug. It's how the technology works today. Every session starts blank.

Now, ChatGPT and Claude have started adding "memory" features — small snippets saved from your conversations. It's a step in the right direction. But look at what it actually stores: fragments. A fact here, a preference there. You can't control the structure. You can't decide what's important. You can't build a systematic profile of a counterpart or a debrief of a deal. It's a Post-it note on a whiteboard — better than nothing, but not a system.

What we're building is different. You decide what AI remembers. You structure it. You control it. And you can feed the same memory to any AI tool — not locked into one platform.

So how do you give AI real memory?

The only way: you bring it yourself.

You provide documents. Files. Structured information that you feed to AI at the start of every conversation. That's the entire mechanism. There is no other way.

This is actually what I do right now when I work with Claude on this newsletter. Not one file — five. A file that explains the project. A file that captures who I am and my background. A file that describes how I like to work. A file that accumulates what we've learned across sessions. And a file that transfers context when we switch from one conversation to the next.

Five documents. Each one small. Together, they mean the AI doesn't start from zero. It starts from me.

That's not a workaround. That's the solution to a fundamental constraint of current AI. And it's exactly what you can build for negotiation.

Last week's system gives you a folder structure. Knowledge Base, Relationship History, Active Deals, Prompts. That's essential — but it's a filing cabinet.

A filing cabinet stores documents. Memory is what you learn from them.

Here's the difference. You had a negotiation call last Thursday on Teams. The transcript is sitting in your folder — 47 pages of back-and-forth. That's the filing cabinet doing its job.

But you remember something from that call. The CFO anchored aggressively in the first round — then softened by the third. The procurement lead said "we're flexible on timeline" but their body language said otherwise. Their team deferred to the CTO on technical scope, not the project manager who was nominally leading.

That's memory. Not the 47-page transcript — the patterns, the insights, the intelligence that makes your next session sharper.

An experienced negotiator carries 20 years of this in their head. Your AI carries none of it — unless you write it down and bring it to the conversation.

Here are the seven documents that make that possible.

Each of these is a document (or a type of document) that you build, maintain, and feed to AI when relevant. Together, they give your AI a memory it doesn't have on its own.

Your Knowledge Base captures what you know. Your Identity captures who you are at the negotiation table — and how you want AI to work with you.

This is the document that tells AI:

Think of it as the difference between a textbook and a personal brief. The Knowledge Base is the textbook — frameworks, principles, methodologies. The Identity is the brief — "here's how I think, here's how I work, here's what I need from you."

When you feed AI both, it doesn't give you generic analysis filtered through standard frameworks. It gives you analysis filtered through your frameworks, delivered the way you think.

Your Relationship History folder from last week collects raw data: meeting notes, emails, deal records. Essential — but it's a filing cabinet, not a brief.

Counterpart Memory is the intelligence layer on top. For each major counterpart, a single evolving document that captures what you've learned about how they negotiate:

Here's where your meeting transcripts become powerful. That Teams or Zoom transcript from last Thursday? Feed it to AI along with the existing Counterpart Memory and ask: "Based on this transcript, what should I update about how this person negotiates?" The AI reads the 47 pages you'll never revisit and extracts exactly what matters — new patterns, changed positions, reveals you might have missed in the moment.

The document grows after every interaction. Not by adding more raw notes — by updating the intelligence. The transcript is the input. The Counterpart Memory is the output.

When you feed this to AI before your next session with the same counterpart, the AI doesn't start fresh. It starts with everything you've learned. It can warn you: "Based on your notes, this counterpart tends to re-anchor when you make concessions early — consider holding firm in the first round."

This is the single most powerful component in the memory system. Everything else feeds off it.

After a negotiation (or a significant round), you write a structured debrief:

The debrief and the transcript work together. You debrief from memory — your impressions, your gut read, your strategic assessment. But the transcript catches what you missed. Feed both to AI and ask: "Compare my debrief with the transcript. What did I miss? What patterns do I not see?"

Here's the key: the debrief doesn't just sit in a folder. It feeds back into the system.

The debrief is the engine that keeps the entire memory system growing. Without it, your folders collect dust. With it, every negotiation makes the next one better.

Individual deal debriefs capture what happened in one negotiation. Cumulative Lessons capture what you've learned across all of them.

This is the meta-layer:

This document compounds over time. After 5 deals, it's useful. After 20, it's invaluable. After 50, it's a competitive advantage no one else in the room has — because no one else has systematically captured their own patterns.

When you feed it to AI, the AI doesn't just help with today's deal. It helps with today's deal in the context of everything you've ever learned.

A practical problem: AI chat windows have limits. A complex negotiation spans weeks or months. You can't keep one conversation going forever — eventually the AI loses track of earlier context, or you need to start a new session.

Session State is a short document you update when you pause or end a session:

Next time you open a new chat, you feed this in. The AI picks up exactly where you left off — no re-explaining, no lost ground.

The negotiation equivalent of saving your game.

Over time, your debriefs and lessons will reveal recurring situations. The Playbook is where you codify your responses:

These aren't generic tactics from a textbook. They're your tactics — tested, refined, proven in your specific context. When you face a familiar situation, you pull the playbook, feed it to AI, and get advice grounded in what's actually worked for you.

Not every AI output is worth saving. But some are. The BATNA analysis that surfaced a risk you hadn't considered. The creative options brainstorm that broke a deadlock. The counterpart profile that predicted their move accurately.

Save these — not in the deal folder (that's for deal-specific outputs), but in a separate archive organized by type. Over time, you build a library of your best AI-assisted thinking. Patterns emerge. You see which prompts + which context = which quality of output. The archive becomes a feedback loop for improving your entire system.

Here's what the full system looks like — building on last week's foundation:

Everything from last week stays — we're building on it, not replacing it.

You don't need all of this on day one. Start where you are:

Here's what makes this different from just "being more organized."

Each layer feeds the others.

A deal debrief updates your counterpart memory. Your counterpart memory improves your next deal brief. Your cumulative lessons refine your playbooks. Your playbooks shape your prompts. Your prompts generate better outputs. Your best outputs get archived. Your archive reveals what works. And the cycle continues.

After 5 negotiations, your AI knows your style, your counterparts' patterns, and your preferred tactics.

After 20, it has more organized institutional memory than most negotiation teams.

This isn't about making AI smarter. AI is already smart. It's about giving it something to be smart with. We're overcoming a fundamental limitation of every LLM — and turning it into a system that gets better with every deal you close.

Take your system from last week and add two memory layers:

Step 1 (5 minutes): Create a Negotiator Identity document. Write down your negotiation style in 2-3 sentences, your top 3 non-negotiables, and how you want AI to advise you (challenge you? support your direction? play devil's advocate?).

Step 2 (10 minutes): Pick your most recent negotiation and write a Deal Debrief. What was the strategy? What happened? What worked? What didn't? What would you do differently?

Then run a prompt with your Knowledge Base + Identity + Debrief attached. Ask AI: "Based on my debrief and my negotiation style, what patterns do you see, and what should I do differently next time?"

If you have a meeting transcript from that negotiation, attach that too. Ask AI to compare your debrief with the transcript — what did you miss? What patterns do you not see?

That's the memory system in action. One debrief, one identity file, and your AI already knows more about your negotiation practice than it did yesterday.

Next week: The complete workflow.

You've built the prompts, the system, and the memory. Now I'll walk through an entire AI-assisted negotiation preparation — end to end. One scenario, every step: from opening the folder to walking into the meeting. Everything from the last six issues, working together.

Your AI is already smart.

The question is whether it has anything to remember.

#Negotiation #NegotiationSkills #NegoAI #AI #Memory #ChatGPT #Claude #NegotiationPreparation