The Pre-Move Thesis: Why Legal AI Should Start Work Before You Ask

In chess, grandmasters don't wait for their opponent to move before thinking. They've already calculated dozens of variations, anticipating threats and opportunities several moves ahead. By the time their opponent plays, the grandmaster has already "pre-moved." Their response is ready the instant their opponent makes a move, because they anticipated what was coming and prepared for it in advance.

At Mage, we believe legal AI should work the same way. Instead of waiting for attorneys to ask questions in a chat interface, the best AI will anticipate what they need and have the work ready before they even know to request it.

The Problem with Chat-Based Legal AI

The dominant paradigm in legal AI today is the chatbot. You open an interface, type a question like "summarize this contract" or "find cases about indemnification caps," and wait for a response. This is what we call reactive AI. It waits for you to make the first move.

But this model has a fundamental problem: it requires attorneys to break their flow.

Research on knowledge worker productivity shows that it takes an average of 23 minutes and 15 seconds to fully regain focus after an interruption. For M&A attorneys reviewing hundreds of documents under tight deadlines, every context switch represents cognitive friction that accumulates into hours of lost productivity. Alt-Tab to a chat window, compose a prompt, wait for results, copy-paste back. Each step pulls attention away from the core work.

Psychologist Sophie Leroy calls this phenomenon "attention residue." When you switch from reviewing a contract to typing a prompt, part of your attention remains stuck on the contract. You're now processing the AI chat with diminished cognitive capacity. When you return to the contract, you burn mental energy reloading context. Studies show this kind of fragmented attention can increase error rates by up to 50%.

The In-Flow Principle

The solution isn't better chatbots. Chat should be secondary. The primary interface should be tools that automatically do the work without you having to explain what you need.

Look at the most successful AI tools for knowledge workers. GitHub Copilot doesn't make you describe the code you want. It infers your intent from your cursor position and suggests the next line in real-time. Superhuman doesn't require you to manually sort emails. Its AI categorizes them in the background so your inbox is already organized when you arrive. Notion AI lets you highlight text and transform it with a single keystroke, never breaking your writing flow.

These tools share a common design philosophy: the AI should be invisible. It should work within the environment where attorneys already spend their time, not require them to visit a separate destination.

We call this the "in-flow" principle. The goal is to preserve the psychological state of flow, that peak cognitive efficiency where the brain's full processing power is deployed on the complex problem at hand. Any tool that breaks flow with extra clicks, context switches, or waiting spinners is working against the attorney, not for them.

From Co-Pilot to Autopilot

But being "in-flow" is only half the equation. The bigger opportunity is moving from assistance to anticipation. From AI that helps when asked to AI that acts before you ask.

In the legal tech industry, this represents a paradigm shift from what analysts are calling the "Agentic AI" approach. Unlike a chatbot that waits for prompts, an AI agent is given a broader goal and autonomously plans, executes, and monitors the necessary sub-tasks.

Consider the M&A due diligence process. The traditional workflow looks like this:

1. Documents are uploaded to a data room
2. An attorney reviews each document manually
3. The attorney takes notes on key terms and risks
4. The attorney asks a paralegal to organize findings
5. The attorney drafts a memo summarizing the diligence
6. The partner reviews and requests revisions

Each step requires the attorney to initiate action. Even with AI assistance, they're still the ones clicking "summarize this document," copying results into a spreadsheet, and manually compiling findings into a memo.

Now imagine the pre-move approach:

1. Documents are uploaded to the data room
2. AI immediately classifies each document, extracts key terms, and flags risks
3. A draft diligence memo is auto-generated, organized by category
4. The attorney reviews work that's already done, making corrections and additions

The attorney never had to ask for any of this. The AI anticipated that uploaded documents need classification, that key terms need extraction, that a memo will eventually be required. It pre-moved. The attorney's role shifts from doing the work to reviewing the work.

The Chess Grandmaster Model

This is why we use the chess metaphor. A chess grandmaster doesn't just react to the position on the board. They've studied thousands of games and know what usually comes next. When documents hit a data room, we know with high probability what work will follow:

• If it's a contract, attorneys will need to know the term, liability caps, change of control provisions, and governing law
• If it's a set of employment agreements, attorneys will need a comparison matrix
• If it's a litigation matter, attorneys will need to understand exposure and status
• If it's IP documentation, attorneys will need to trace the assignment chain

We can anticipate these needs because the M&A diligence process, while complex, follows predictable patterns. The right move is knowable in advance. We just need AI sophisticated enough to execute it before being asked.

Why This Matters for Attorneys

The attorneys we work with didn't go to law school to manually populate spreadsheets. They went to law school to exercise judgment, negotiate strategy, and counsel clients. But the reality of modern M&A practice is that junior attorneys spend the vast majority of their time on mechanical tasks: reviewing documents, extracting data points, formatting deliverables.

Our thesis is simple: AI should do the first draft of everything.

Not the final product. Attorneys will always need to review, refine, and exercise judgment. But the initial pass? The data extraction, the comparison matrices, the memo drafts, the disclosure schedule population? That work should be done before the attorney even opens the file.

This isn't about replacing attorneys. It's about returning them to the work that actually requires a law degree. When AI pre-moves the mechanical tasks, attorneys can spend their time on strategy, negotiation, and client relationships. That's the high-value work that drew them to the profession in the first place.

Building the Future

At Mage, every product we build follows the pre-move thesis:

• Tabular automatically extracts and categorizes key provisions across hundreds of documents the moment they're uploaded. No prompts required.
• Memo generates comprehensive diligence memos that are ready for review, not waiting to be requested
• Schedules auto-populates disclosure schedules by cross-referencing the data room against the purchase agreement

We believe the future of legal AI isn't a better chat interface. It's AI that thinks several moves ahead, anticipates what attorneys need, and has the work ready before they know to ask for it.

Like a chess grandmaster, we're pre-moving. That way, attorneys can focus on the game that matters.