This website uses cookies

Read our Privacy policy and Terms of use for more information.

Claude for Legal landed on May 12, 2026, and four weeks later most of us were still sorting out what it actually was. A lot of new language arrived with it, and even people who had been using AI for a while found the terms unfamiliar. Plenty of lawyers love playing with AI and use it in daily work, then hit a wall the moment things get past the simple stuff.

This How to Contract webinar, hosted by Laura Frederick, Founder and CEO of How to Contract, set out to clear that wall. The guest was Colin Lachance, CEO of LawQi, an education company that uses AI to teach lawyers about AI. Colin has spent around 15 years in legal innovation and describes his whole career as working an arm's length from an engineer, which is exactly why he is good at translating this for the rest of us.

Colin walked through the vocabulary that makes Claude for Legal make sense, where the lawyer engages and where IT belongs, how data moves through the model context protocol, and how to get started without writing code. Laura then pulled audience questions on subscriptions, security, surgical redlines, how Claude compares to tools like Harvey, and what to do on day one.

Here are our top ten takeaways from the speakers' comments during the webinar:

  1. Learn the building blocks before you start building. Claude for Legal runs on a handful of concepts, and getting them straight early is what lets the rest click. Colin walked through how the model talks to other tools, how saved instructions work, how the system hands work to helpers in the background, and how a house-style file keeps everything consistent. You do not need to code to understand these. You need to understand them well enough to direct the work.

  2. Treat your Claude Projects experience as the on-ramp. If you already use Projects, you are closer than you think. The instruction file you put in a project does the same job as a skills file in a Cowork session, and the material you preload behaves a lot like pulling in outside data. Colin framed the jump from Projects to Cowork and Skills as changing the frame, not starting over. That reframe takes most of the intimidation out of it.

  3. Know which work is yours and which belongs to IT. You can do a surprising amount yourself in chat, Cowork, and the Microsoft apps. The platform and managed-agent layer, and most firm-wide rollouts, are where you bring in your tech people. Colin drew that line so you do not stall out trying to build infrastructure you were never meant to build alone. Knowing the boundary keeps you moving on the parts you actually own.

  4. Lean on what you already know, which is delegation and supervision. The skill that makes this work is the one you use with junior lawyers every day. You tell the AI what to do, what to flag, and what to escalate to a human, and the skills files just write that down. Colin liked the line that the agent extracts and the lawyer commits. You stay the one who decides.

  5. Keep your data where it lives instead of copying it everywhere. The old way left copies of sensitive material scattered across tools. Colin explained how the model context protocol lets the AI inspect information where it sits without pulling a copy out. For legal research especially, connect your real sources so the system works from actual court information instead of guessing. Source attribution gets better and the conclusions stay tied to what you fed it.

  6. Get comfortable in Cowork before you chase skills. The fastest way in is to open the desktop app, point Cowork at a folder of non-confidential files, and ask it to do something genuinely challenging. You will watch it plan, ask permission, and work through the back-and-forth that teaches you how delegation actually feels. Colin called this the real first step. Skip it and the more advanced pieces will not land.

  7. Start a new build with the cold start interview. You do not have to design a whole system from scratch. Colin described how the cold start interview asks you what you need and assembles the pieces, getting you productive in a couple of minutes once you have worked up to it. Your first build is slow, your second is faster, and it compounds from there.

  8. Pay for access and match the plan to your usage. Do not run real legal work on the free tier. The reason to pay is privacy, since paid plans let you turn off training on your data. Colin suggested starting at the entry paid tier, moving up when the system starts telling you to slow down, and considering Teams when you want to share securely across people. On these plans you are not billed by the token, so the surprise-bill stories you read about are mostly developers building apps.

  9. Watch for the surgical redline trap. Asking Claude chat to redline a document usually produces a visual mess, because it tends to rebuild the document from scratch rather than tracking changes. Colin pointed to AI built into Word as the better path, since it is constrained to the document you are already in. If clean redlines matter to you, that is where to look.

  10. Use Claude for Legal as a lens, not a verdict on the future. Even if you never build a tool yourself, understanding what is possible makes you a sharper buyer of legal tech and a better judge of what vendors are selling. Colin made the case that your clients will start showing up with more sophisticated work too. He was careful to say this may not be the tool you use in the future, only the clearest window into what is coming. That is the reason to spend a few hours with it now.

Subscribe to Stay in the Loop

Our weekly newsletter brings you the next round of How to Contract webinars along with write-ups like this one. Subscribe now to keep the practical takeaways coming, whether or not you can make the live sessions.