Amazon wants a seat in your law office. Here's why you should think twice before you give it one.
Its new AI workspace, Quick, is pitched straight to legal teams as a time-saver. Source precedents, read long contracts, pull case law, draft the first version, and hand the saved hours back to you for the work that actually needs a lawyer.
That promise is directionally right. Here's my straight answer anyway.
A generalist tool like Quick will help a little and annoy you a lot, because it's missing the one thing that makes legal work yours: context and taste.
Amazon has never run your kind of matter. It doesn't know why you'd cite one authority over a cleaner-looking one, or how you like a brief to open.
So the time you save gets spent babysitting and correcting the output instead of hitting that "aha."
Use it. Just don't expect it to feel satisfying, and don't stop there.
The real prize is machinery with your experience baked in. I'll show you where Quick falls short, then how we built that machinery for ourselves.
What Is Amazon Quick for Lawyers?
Amazon Quick is a general-purpose AI assistant and workspace from AWS. It connects to your documents and applications, answers questions about them, runs research, and automates multi-step tasks. For lawyers, Amazon markets it as a way to research case law, review contracts, draft documents, and cut e-discovery time from one place. It launched as Quick Suite in October 2025 and starts at $20 per user per month, with a $40 tier for heavy research and automation plus consumption-based charges on top.
Read that positioning again. Every job on the list is the same move: do the mechanical first pass so a lawyer reaches the same place in fewer hours. That's the appeal. It's also where the gap hides. This is an enterprise assistant wearing a legal landing page, not a product built from the way you practice.
What Amazon Says Quick Does for Legal Work
Amazon's legal pitch covers four jobs. Each one is real work your firm pays for today, which is why it gets attention. Here is each claim next to what the task actually demands before you'd sign your name to the output.
| Amazon's claim | What the task actually requires |
|---|---|
| Surface relevant case law, precedents, and statutes in minutes | Verified citations from a legal database, correct jurisdiction, and a human check on every authority before it reaches a client or a court |
| Extract key clauses, flag risks, and track obligations across contracts | Knowing which risks matter for this client and this deal, which is judgment, not extraction |
| Generate drafts from your templates and clause libraries | Templates and clause libraries that are actually organized, current, and loaded into the system, which most firms have not done |
| Summarize filings and keep e-discovery details from slipping through | Defensible process, privilege awareness, and a lawyer who can stand behind what was produced and what was withheld |
Look down the right column. Every task leans on context and judgment that live inside your firm, not inside the software. Quick can accelerate the mechanical half. The half that makes it legal work stays with you. If you're weighing this category broadly, my guide to AI for lawyers covers where these tools genuinely help.
A general-purpose AI assistant with a legal landing page is not a legal AI product. Judge it as the former.
One objection comes up right away, so let me handle it fast. Yes, Amazon says your queries aren't used to train its models, and AWS already hosts privileged data for most firms through their practice management software. Confidentiality is answerable. A written agreement, scoped access, and the ABA's Formal Opinion 512 as your checklist (the FAQ below walks through what to ask). But confidentiality was never the hard part. The hard part is whether the answers are good enough to save you time once the security review is behind you. That's where I can speak from experience.
The Missing Ingredient: Context and Taste
Quick runs on a strong model. A generic one. It read a large slice of the internet, not twenty years of your matters. That's the whole problem in a sentence.
Think about what you carry in your head. You know the judge in your county has no patience for a certain argument. You know this client won't sign a clause worded that way. You know two cases can look equally on point, and one is still the better cite because of how a panel reasoned three years back. That judgment is taste. The years behind it are experience. Neither is written down anywhere Quick can read.
You can paste a style guide into a prompt. You can tell it your preferences. But the moment a matter gets specific, the generic reasserts itself, and you're back to editing. That's the trap with off-the-shelf AI. It gets you to generic-competent, then hands you the last and most important mile.
So the saved time doesn't fully show up. You trade drafting hours for reviewing and correcting hours. The tool is fine. Fine isn't satisfying, and "fine but not satisfying" is the exact feeling that makes a lawyer open Quick less each week until they've quietly stopped.
Off-the-shelf AI gets you to generic-competent. The value you actually want lives in a context layer, and a context layer is not a prompt.
How We Built Taste Into Marcel
We hit this exact wall ourselves. Over the past year we built Marcel, our own AI tool, trained on how our team actually works and what we know. Same category Amazon is selling. The difference is we sat inside it and refused to ship generic.
Getting from "Marcel is connected to our documents" to "I'd stake a client recommendation on this answer" took far more than any demo implies. The model was the easy part. Three things did the real work.
Prompting the context layer
We encoded how we think. What we weight, what we ignore, what a good answer looks like next to a merely plausible one. Small wording changes moved the output more than swapping the underlying model did. This is the taste, written down at last, and it's slow, unglamorous work no mass product does for you.
Unit testing the answers
We ran Marcel against real questions with known answers, caught every confident-but-wrong reply, and turned each one into a test. Software engineers have done this for decades. Almost nobody does it to an AI assistant, which is why most of them feel great in a demo and shaky on your third real question.
Iterating until it earned trust
Fix, re-run the tests, repeat. That loop is the line between a tool you rely on and a tool that impresses you for a week and burns you in week three. It never fully ends, and that's the point.
The model is a commodity now. The context you encode and the tests you wrap around it are the moat, and a $20 mass-market seat ships without either.
We took that machinery further. We built a system for criminal defense lawyers that ingests and searches thousands of pages of discovery, the kind of pile that buries a case when you're paging through PDFs by hand. Ask it who said what and when, and it returns the passages with citations. It only works because the taste and the testing sit on top of the model. Quick can't offer that. There's no forward-deployed engineer sitting with your firm, encoding your judgment and testing against matters your partners know cold, and at $20 a seat there can't be. I've watched the same arc with ChatGPT in law firms. The firms that got value treated it as a supervised assistant, not an answer machine.
Directionally Right, for Operations and Marketing
None of this means the AI skeptics are right. They're wrong, and the ones treating this launch as proof that AI is hype will pay for that read later.
AI will fold into legal practice the way email and practice management software did. Completely, then invisibly. Amazon entering the space tells you where the money is going, and Amazon is rarely alone for long. So the question isn't whether to engage. It's how to engage now, while the products are still clumsy, so your firm has the muscle memory when they aren't.
Notice this cuts two ways at once. It's true for operations, the contracts and research and discovery Quick is aimed at. It's just as true for marketing, where the same machinery decides whether an AI answer engine cites your firm or a competitor's. Firms that started publishing and optimizing early own the search rankings today. The same compounding is running right now in AI marketing for law firms. Start small now or start behind later.
How to Start Building Machinery With Your Taste Baked In
Forget Quick for a second. The prize was never a subscription. It's your judgment, encoded and testable, so the tool stops guessing and starts sounding like your best lawyer on a good day. You don't need to understand vectors or embeddings to begin. You need one narrow job and the discipline to write your rules down.
- Pick one job you do the same way every time. Intake triage, one specific contract type, or the discovery in a single matter. Narrow beats broad every time.
- Write down your taste. The rules in your head: how you decide, what you'd never do, what a strong output looks like. This is the context layer, and it's the part Amazon can't ship you.
- Feed it in, then test with questions you already know cold. Watch closely for the answers that are confident and wrong. Those are the ones that would have embarrassed you in front of a client.
- Turn every miss into a fix and a test. Then grow it one job at a time as it earns trust. The machinery compounds. A rushed dump of everything does not.
Quick can be the on-ramp for step three if you want one. Point it at a single narrow job, hand it your written rules, and test it against answers you know. Keep anything privileged out until it earns the security conversation. If it holds up, good. If it doesn't, you've spent a seat and a few hours and learned exactly what to demand from the next tool. Either way, the rules and tests you built are yours to keep and carry to whatever comes next.
The tool will change every quarter. The taste you encode and the tests you write are the assets that last.
Frequently Asked Questions
What is Amazon Quick?
Amazon Quick, launched as Quick Suite in October 2025, is a general-purpose AI assistant and workspace from AWS. It connects to a company's documents and applications, answers questions, runs research, and automates tasks. Amazon now markets it to legal teams as a time-saver for research, contract review, drafting, and e-discovery.
Is Amazon Quick a legal-specific AI tool?
No. It's a general-purpose enterprise AI workspace with a legal use-case page. It isn't built on legal research databases and carries no legal-specific citation verification, so its output needs a lawyer's review before it reaches a client or a court.
Why do generalist AI tools like Quick frustrate lawyers?
Because they're missing your context and taste. The model read the internet, not your twenty years of matters, so it doesn't know which authority you'd cite or how you write a brief. You end up correcting and babysitting the output, which eats the time the tool was supposed to save.
Is Amazon Quick safe for confidential legal documents?
Amazon states that queries aren't used to train its models and that Quick runs on enterprise-grade AWS security. Safety for your firm still rests on your own diligence under ABA Formal Opinion 512: a written confidentiality agreement, scoped access controls, and a clear answer on where data lives and who can see it.
Does Amazon train its AI on your firm's data?
Amazon's published FAQ says user queries are never used to train the underlying models. Get that commitment in your written agreement rather than trusting the marketing page, because the confidentiality duty is yours, not Amazon's.
How much does Amazon Quick cost?
Quick starts at $20 per user per month, with a $40 per user tier for heavy research and automation. Consumption-based charges can apply on top, so a heavily used deployment costs more than the sticker suggests.
Can Amazon Quick replace a paralegal or associate?
No. At its best it speeds up mechanical work like summarizing, extracting, and first-pass drafting. The judgment about what matters, what's privileged, and what's correct stays human, and under Formal Opinion 512 the supervision duty stays with you.
What is a context layer, and why does it matter?
A context layer is your firm's judgment written down in a form the AI can use: how you decide, what you weight, what you'd never do. It's the difference between generic-competent answers and answers that sound like your best lawyer. A prompt alone can't carry it, which is why off-the-shelf tools plateau.
Should my firm try Amazon Quick now or wait?
Try it small now if you have the appetite. One seat, zero client data, one narrow job, and a written set of 20 questions you already know the answers to. The learning compounds even if the tool fails your test, because the category improves every quarter and you'll know exactly what to re-check.
What should I ask any AI vendor before connecting client data?
Four things: where the data lives and who can access it, whether confidentiality terms sit in a signed agreement, whether your data is excluded from model training, and what happens to the data when you leave. If a vendor can't answer all four in writing, stop there.
Want a Straight Answer on AI for Your Firm?
We build and test AI on our own business before we recommend anything to a client, from Marcel to the discovery-search tool we built for criminal defense lawyers. And we work with law firms every day on the growth side of this: getting found, getting cited, and turning attention into signed cases. If you want an honest read on where AI fits your firm's operations and marketing and where it's still smoke, book a call with our team. You can also read more about how we work.
