Agentic AI for Lawyers: Moving Beyond Enhanced Search, with Drew Bloom

Episode Notes

Transcript

Stephanie Everett:

Hi, I’m Stephanie.

Zack Glaser:

And I’m Zack. And this is episode 5 81 of the Lawyers podcast, part of the Legal Talk Network. Today I talk with Drew Bloom about artificial intelligence and a agentic artificial intelligence. Now don’t worry, we don’t get too nerdy in this and it doesn’t get too specific, but there is some real hands-on information on how you can use ag agentic, artificial intelligence in your office today and how you can prep yourself for using ag agentic artificial intelligence in the near future.

Stephanie Everett:

Awesome. I can’t wait to listen to that. I am learning every day.

Zack Glaser:

Well, speaking of learning, every day you are on a work trip right now out in Vegas, which there are worst places to take work trips, but you’re out in Vegas at a conference. Well, you tell ’em to kind of keep your learning, keep your education going.

Stephanie Everett:

Dean and Conrad have been on our show before. They have the Lunch Hour Legal Marketing podcast and they’re doing a conference for the first time all about legal marketing and I am here as a student. I’m here. I’m not speaking, we’re not a vendor, we’re not a sponsor. I just called Gee and I was like, can I just come and participate and learn? And he was like, you can come any way you want, Stephanie, of course join

Zack Glaser:

Us. Oh, that’s fantastic. So why legal marketing? Why that one? You’ve got, there are tons and tons of things to go to. Why go to that one for you? For us,

Stephanie Everett:

Yeah, I mean I guess two things kind of to back up about two weeks ago or three, I can’t keep up, but several weeks ago I was actually in San Francisco at HubSpot’s inbound conference, which is about marketing in general. It’s not industry specific. It’s about everything that’s happening in marketing. And I got to tell you, it was drinking from a fire hose for sure. Marketing is changing, search is changing, and that was the one thing we kept hearing over and over again. The predictions are by 2028, Chad, GPT will outrank Google in search. More people will go to chat to ask questions and look for information than they will go to Google. And so for years, all of us have been learning about SEO and content marketing and all the things we need to do to show up in Google search. Well now it’s turned to what they’re calling a EO answer engine optimization. How do you show up for perplexity and Claude and Chat GPT? Because people are just going to go there and say, what family? I’m getting a divorce? What lawyer should I hire?

Zack Glaser:

Right?

Stephanie Everett:

And it’s going to tell them an answer and you want it to be you.

Zack Glaser:

That’s a good point. It’s going to give them an answer. Yes, it is. Absolutely. I’ve also heard this referred to as GEO, but I think a EO is the one that’s catching on, but either way it’s the getting yourself shown up in these generative engines, I guess. Yeah. So you found that at a HubSpot and just very quickly, HubSpot is a client resource manager that we use that a ton of people use. It’s a really, really big one, but just for anybody that was wondering what HubSpot is, it’s a CRM.

Stephanie Everett:

Of course. Just like anything, you don’t have to be a HubSpot user to go to the conference, but obviously if you do use it, it’s got lots of, you learn a lot more. So I did that a couple of weeks ago and now I’m here for this conference because it is focused on legal specifically, and I really feel like our job, part of our job, I have a lot of jobs, but one of our jobs

Zack Glaser:

Have lots of hats for Stephanie I

Stephanie Everett:

Know, but for our lab community, for the lawyers that we work with most closely, I always see our job as taking all the information out in the world and trying to distill it down to what they need to know next for their business. They’re busy running their business, they’re practicing law. Do they have time to go figure out all this AI stuff and which tools they should be using? Do they have time to figure out now a EO or whatever it’s going to end up being called. That’s part of the benefit I think of working with us is that you’re figuring out all the AI things and I’m figuring out the marketing thing. We’re figuring out these what is on the horizon for businesses and then distilling it down in a way that makes sense to lawyers specifically so that it’s not just generalized advice. It’s like, no, this is actually how this is going to show up and you’re going to use it in your law firm. So I love, obviously Stay Curious is one of our core values. I feel like I’m always reading, always learning, but I’m really excited. Really I feel like we’re tapping into something new and it’s kind of fun. Okay, now we got to figure out this new thing and how are we going to attack that both for our company but for the lawyers that we work with.

Zack Glaser:

Yeah, I mean that’s the idea of the strategic. The strategist is helping you figure out what’s next, where to go. So you drink from the fire hose so other people can actually take it in consumable measures.

Stephanie Everett:

Yes, I had a weird vision of me is the funnel right? Us as the funnel, but that is what we’re trying to do and give people frameworks. I read on LinkedIn last night on the plane I read somebody was saying, oh, business coaches just help you feel good and we don’t want to do that, right? We’re really trying to say no, we’re here to help you implement frameworks in your business, in this case marketing frameworks. What do you need to know and what do you need to today that’s going to impact your business tomorrow?

Zack Glaser:

I look forward to hearing all the things that you learn at the Lunch Hour Legal Marketing summit. I know we already are behind in unpacking all that you learned at the HubSpot conference, so I really look forward to hearing all the things that you’re picking up there

Stephanie Everett:

And I can’t wait to hear your conversation with Drew. So let’s check that out. Now

Drew Bloom:

I, I’m Drew and I build AI things for law firms. I work at Affinity. I am working really kind of in the background of a lot of things we’re doing for law and my goal is always to stay at the forefront of what’s new in the AI world.

Zack Glaser:

Drew, thanks for being with me. It’s difficult to really get you a proper on the nose sort of introduction in this, I think, hi, I am Drew and I build AI things for law firms is probably the closest. You do a lot of things and you’re saying that you’re in the background of a lot of the implementations and stuff like that. I see you scurrying around getting information, putting things together, making sure that the right pieces are in the right place a lot of times for implementations that we do, but also staying on the forefront of what in the hell can we use this stuff for? I learn a lot from you on that area of what can it do and how easily can we make it happen. So thank you for being with us and for being willing to share your knowledge with our listeners here.

Drew Bloom:

Thanks for having me.

Zack Glaser:

I wanted to ask you specifically, because I think you’re a good person for this about ag Agentic ai. For our listeners, I want to take a step forward. We’re not talking about large language model necessarily. We’re not talking about how do I input something into chat GPT and get a result, get an answer out of it. We’re talking about something a little bit more. So we we’re in step two, we’re artificial intelligence 2 0 1 I’d say, or maybe at least one 50 here, but enough from me on that. Can you help us define or understand what I mean or what we mean by agen artificial intelligence?

Drew Bloom:

So I think you’ve set me up to walk into a trap here because if I go in the weeds a little bit, I think the difficulty is that if you see definitions of this from different companies, different providers, you’re going to hear a hundred different things. So I’ll give you my best take on this, but I’m wading into some unsure waters when I try to make any kind of definition.

Zack Glaser:

I think that’s a good point and lemme just kind of interject a little bit of like we’re trying to get the basics here. We’re trying to help people move forward on it, so it’s not necessarily going to be perfect, but yeah, what are we talking about generally?

Drew Bloom:

Well, when you say that an AI is an agent, when you designate an agent, even a real human one, you are saying that that thing can take action on your behalf. The weird thing is that we’ve been in these waters for a while where people have talked about assistance and more and more of these AI assistance have gained the ability to take actions. So I think the big thing that I ask myself is, well, what makes an agent an agent?

To me that is a system where the AI can go back and forth between you, the user and a set of tools that helps it accomplish tasks for the user that goes beyond simply let’s chat back and forth. I think once you’ve added tooling that allows it to reach into a software system or talk to the internet, you’ve really gone, I think beyond the model that first came out in 2022 and 2023 and you’ve entered kind of this new phase where the AI can go and take steps on your behalf. And I would even say it becomes more agentic the more steps it can take on your behalf before it comes back and says, what’s next?

Zack Glaser:

Let me get into explain it like I’m five sort of place here, which I think is a great, this is where a lot of good conversations come with you and I is you helping me understand things. When I first think, when I think of LLMs or chat GPT or Anthropic or Claude or something like that, I think, well, it is connected to the internet. It is going out to the internet and getting these answers because that’s what Google does when I’m, or at least that’s how I perceive what Google does when I’m asking a question there. What do you mean? What’s the difference in an agent going and connecting with the internet or an agent going and connecting with the tool and chat GPT or an LLM just kind of querying and using that, the internet at large, let’s say.

Drew Bloom:

I mean that’s a great distinction. So when you go into chat GPT or you go into Claude and you talk to it and it decides that it’s going to, let’s say search the internet, it comes back right away and it says, here’s what I found Zack, how else can I help you? And that’s sort of the rhythm that you feel with a clot or a chat GBT. But if you use one of the new tools that’s become, I think really common inside of these, which is called deep research, that’s actually what I would call more ag agentic Deep research just asks you, Hey Zack, in general, what would you like me to go find out? And then it just disappears, right? It goes out, it talks to the internet, it comes back later, Zack gets a cup of coffee, maybe you can even finish a POP two or make some lunch and you come back 15 minutes later and it’s got a 12 page presentation for you based on its research that’s maybe the agency piece versus what you’re used to seeing in your assistant tools.

Zack Glaser:

So then what’s behind the curtain there? Because from my perspective, it’s just taking longer and it feels like it’s doing the same thing, it’s taking longer and it’s getting me a deeper answer. I mean, that’s the deep research thing, but what does that mean? What’s it doing? And where I’m trying to go here is obviously how can we as attorneys start using that concept, that sort of stuff in our office. So yeah, pull back the curtain here. What’s it doing?

Drew Bloom:

So in the background, that deep research agent is actually a piece of software, and it’s not just one ai, but it’s as though the AI had a harness or an orchestrator almost let’s say that if a person could tell the AI, go search this, then go search this, then go search this. They just decided to automate that piece with software. Now how that translates to law and outside of the deep research context then is you’ve now got more and more of these product offerings in the legal space where you have an AI that you can say, go do this, and they’ve built their own little custom harness that tells the ai, you know what? Here’s the task. Let’s go take 10 steps on our own and see if we can solve this on the behalf of the user. Then we’ll come back instead of you having to drive the car the whole way, I use that analogy a lot as the autonomous vehicle idea is I want to be able to tell it we’re going somewhere, I’ve got to put it in the GPS, but the car can handle most of what I need to do, and if I need to take the wheel at any point I can, but that’s a difference from conventional driving where in the last 50 years you’re used to maybe cruise control, but you have to still steer the wheel.

That’s the assistant versus the agent. I would say format.

Zack Glaser:

That makes more sense to me, but it seems like the agentic tools then would need to be kind of built for purpose, right?

Drew Bloom:

Yes, that is exactly the problem is now you’ve got this almost open world of all the possibilities of different agentic tools you could build and different, well, you could almost consider them servers really where the AI can go and talk to this program that helps facilitate that agency. The AI is actually learning from the program. It’s connected to how to best serve your interests, and you could even have multiple of these functionalities connected to one ai.

Zack Glaser:

It feels like then it’s kind of like layered artificial intelligence even.

Drew Bloom:

That’s exactly right, and I don’t want to get too far in the weeds here, but there’s something I know you’ll find interesting and our listeners will find interesting is when they compare the performance of the top models in humanity’s last exam, how well can the new GPT Gemini Claude models solve problems that humans can’t even take on

When they compare some of that performance and accuracy to instead an approach where we use a more limited model, a standard GPT five or a standard Claude model, and we just give it a thousand chances to accomplish the same task. What we find is that we still have a high accuracy rate if we can pick the best of those completions, and what I’m saying there is what we could do instead of having to pay for larger pricier models that could do a bunch of reasoning and have a ton of training time devoted to them is instead we could allow the model to do a lot of different steps, small steps,

Zack Glaser:

Break

Drew Bloom:

Things down for it and help it accomplish tasks for us and get that same high level of accuracy. That’s sort of what’s happening inside of this sort of agent architecture where we’re actually breaking down the steps using a platform that it can reach out and let’s say it talks to your Microsoft Word, I’m very excited for this, for all firms is the agent that can talk back and forth with your Microsoft Word document and make edits in real time because it knows what you’re talking about and actually how to read the document as a machine, but in a way that corresponds with the way that we read it as humans.

Zack Glaser:

What in the hell do you mean by that, drew? That’s okay. So now we’re getting into how this actually affects our listeners here because the concepts are, we’re breaking things into chunks, and if we can tweak those chunks, those steps, I think steps is probably the better word. If we can kind of provide feedback in each individual step, then we’re going to get a better result as we go than just saying, Hey, go do this a million times and we’re going to give you a ton and ton and ton and ton of feedback. We’re a little bit more focused. Okay, so what does that look like? What does that actual interaction look like in your mind for me, interacting with an agent, interacting with my Word document?

Drew Bloom:

So what’s going to feel really good, and this should be out in the next 12 to 24 months in a variety of different programs.

Zack Glaser:

I wanted you to say hours. Let’s say that so bad,

Drew Bloom:

Sorry.

Zack Glaser:

In our programs we’re just 12 to 24 hours. I was like, that be great. That’s how fast AI is moving. Okay, fair enough. No. Okay, so 12 to 24 months hopefully.

Drew Bloom:

Yep. And you’ll expect to see it everywhere because right now it’s really in the avant-garde products. The ones that are really, really ahead on their game is you could come into a Word document, open up an add in, and basically tell that agent what here’s looking at, I’d like to upload a file and I want you to edit this contract based on this term sheet. Think about this even being multi doc, if you wanted to, I’m going to upload these five other docs and I want you to use these as sort of a heuristic way to think about my new document. Maybe these are five NDAs that are approved by me as gc, but I want to review this new one and I say, go ahead and review this based on the ones that I think I’ve approved. What this takes away, I never had to sit down and write a playbook. The agent used the context to look through the word document. Then the really cool part that’s going to feel very seamless is that it can actually go and find pieces of the document and offer suggestions like those red lines, the way that attorneys actually work. Whereas before it’s been, well, let me just print for you in plain text, here’s the things that you should copy and paste back into your document. Now agency is like, I’d like to make these 17 edits. Would you like to approve them?

Zack Glaser:

Okay, so I think this is coming together in my brain here. The idea of the artificial intelligence being able to use the tool redlining would be one of those sorts of advancements there of alright, instead of just because when I go into copilot right now and I talk to copilot on one of my documents or I talk to copilot on one of my Excel spreadsheets or something, it can tell me what to do, but it doesn’t do it. I can’t say make a table of this or I can’t say redline this. Let’s stay with that example. I can say, do you have any suggestions on what this paragraph should look like? And it would give me an example of that. Now I’ve got to go in and redline the thing. So the next step, the age agentic step would be, Hey, do you have any examples? Here’s the example, here’s the red lines I would make. Do you approve them?

Drew Bloom:

That’s exactly right. And now the amount of time that you have to spend is cut. I mean I would say by three quarters in that kind of setup, which is really, really nice. It just feels very seamless. And that’s exactly the agency that’s enabled by having sort of a way for it to engage with Microsoft Word-based tools. But you can imagine those tools can exist for any kind of software as well.

Zack Glaser:

The mind reels, right? I’m like, all right, well, can I go into Clio? Can I go into my email? Can I go into, yeah, just pretty much anything at this point, okay, as a attorney who wants to get in front of this wave, how do I start to play with this or think about this, I’m not going to go build a Microsoft Word agent, am I?

Drew Bloom:

We’re at a point where sit and wait isn’t the worst advice for a lot of these. When I say 12 to 24 months, I actually am not going to be surprised if copilot and chat GBT and Gemini launch some functionality like this inside their core products in the next 12 to 24 months.

Zack Glaser:

So now we get into one of my favorite weird concepts. If we want to go into deep space, we actually shouldn’t ever build the spaceship because our technology is going to advance so much that every spaceship we build as our technology advances is always going to outpace the one before it. And so if I build one now and we send it out with our technology in it and we’re trying to get to deep space, and then I build one in 10 years, that 10 years, the technology is going to advance so much that’s going to catch that. So sit and wait is great in that sense, but at some point I have to build the spaceship, otherwise I’m not going to get out there. And I don’t have a tell me when to stop sitting and wait question here, but I just wanted to throw that out there of I like that, but I got to strike at some point, right?

Drew Bloom:

Yeah, I think there’s an alternative pathway if they’re not going to be part of the tools themselves that you have already in your toolbox, the alternative is going to be what’s called an MCP. There’s a lot of hype around these, and I can explain them a little bit more if we need, but I think the core thing here is that inside of your favored AI platform, you can connect custom tools to that ai. So Chachi, BT didn’t even have to build them, Google didn’t have to build them. You can, and it will get easier and easier to do this. You can grab ones that are available on the internet and add them, pay for them, and add them to your own AI clients. So that’s going to be a new marketplace that opens up as well.

Zack Glaser:

Okay, so let’s see. If just telling me what MCP stands for helps me understand what MCP is.

Drew Bloom:

I unfortunately will not help you with that because it means model context, protocol, and that means nothing to anybody. And then we’re going to get too far, I think a field in the technicals and it’s not going to help anyone at all. But basically, just to give you a little history on this, because I think that’s sort of helpful, is that Claude was trying to come up with a way that people could use open source technology to make Claude be able to do more things, and they ended up making a model context protocol for ways to give an AI a chance to understand its instructions, the different resources it can use and the tools it has access to. And they wrapped all of that together into a little piece of software that they call an MCP server. And now the really cool thing that started to happen is that people are excited to provide new toolings that correspond with whatever software, YouTube, Clio, it doesn’t matter whether it’s legal or not, if it has those internet capabilities where you can interact with the tool securely, people can build around that and offer you those access points so that your AI of choice can come in and be able to work with that tool on your behalf.

Zack Glaser:

So in a way, it’s kind of the same concept or close to the same concept as an A PIA standardized sort of way for one product that has its own code, has its own data structure, things like that, to talk to another product that has its own code and data structure and being able to say, alright, well this is the way that we’re going to talk to each other.

Drew Bloom:

You’re absolutely right. And I’ll add to this, when MCP first came out, I think a lot of people, including myself, dismissed it as just another API wrapper, something that we make fun of in the software world, just saying like, look, it’s just an a i, it’s wearing a different set of clothes.

And I felt like the emperor would have no clothes by the end of it. But I think the novel thing that it did is that it also wrapped that together with instructions for using those tools and the tools themselves can take an API that let’s say it’s very mature. It’s got, there’s maybe over a hundred ways to interact with a software like Clio or with net documents as a DMS or SharePoint, and how do I take those hundred things which are completely overwhelming to an ai, and how do I just make it five tools? That is actually the real value of an MCP is being able to wrap it down and compress it in a way that an AI can actually use it. Well, and then you can also add some steps like Zack, I can similarly think you and I are very security minded too. I want to make sure the updates I’m making are legitimate and I’m not sending a hallucination into my billing software. And the MCP server can allow us to add that, right? We can add our own verification steps to make sure that we’re not doing anything crazy with our tools either.

Zack Glaser:

Okay, so with the possibility of going off base here, is this something that I as a law firm can build, or is this something that I need to make a informed determination as to whether or not I should use somebody else’s MCP that they’ve built?

Drew Bloom:

I think for most firms that decision is going to be the latter. I think it’s about deciding when to use someone else’s, and in many cases, you’re going to see legal AI providers, these vendors create their own, and that’s going to be an exciting emergence. I think over the next, again, probably zero to 24 months, a lot more of these vendors are going to come into this space and say, here’s my MCP and you can use it and I’ll take care of all the setup, and it’ll just be another product that you can then have your AI first sort of work experience where you’re not in the actual front end of their software as much.

Zack Glaser:

So yeah, let me go down that path for a second just to make sure that we’ve got this in everybody’s imagination here. So if I were to use an MCP with say, net documents, I would be communicating through Claude and Claude or my whatever AI of choice that was able to connect to that MCP. Claude would then be going and taking actions in net documents on my behalf or not taking actions in net documents on my behalf because we were able to put in appropriate security protocol, but my user interface would not be going into the net documents dashboard. In that case, it would be going into Claude and, okay, so here’s where I’m getting very excited. Could I have multiple CPS connected to where I could interact with net documents, but I could also be interacting with my Azure environment, my own Microsoft environment?

Drew Bloom:

Yeah, absolutely. And I think that is a future. That agency really opens up a lot of possibilities. Let’s think about even, I need to make sure that my data comes from this practice management or time billing and accounting system, and it needs to end up somehow in net documents. If you had an MCP for each of those, then you could just say, Hey, Claude, I need to make an update. Just find this matter number and could that stuff please go from here to hear those kinds of things become possible from that AI chat client as opposed to ever having to touch anything inside of the systems yourself by opening different tabs and windows. And I’ll even go very, very Star Trek on you and push into the sci-fi realm. I think the exciting part may be for a lot of attorneys then is if I had a secure AI app on my phone and I had those MCP servers, I could just hit dictate and be like, Hey, I need you to do X, Y, Z for me in my systems, and it does that for me without me having to be at my computer.

I feel the pain that many of us do of I’m stuck sitting all day. How can I get out? Or if I’m in the middle of something, how can I take care of something that otherwise would take me sitting down in my office and opening a bunch of windows to do? Okay.

Zack Glaser:

So that gives me kind of the basics in my mind, at least me, the basics of agentic and then kind of extrapolating out into where we’re going. That excites me, makes me, yeah, if I can sit there and talk to an agent, talk to the artificial intelligence while I’m writing my Peloton or driving down the road or whatever it is, because that’s actually the way that I want to interact with something for the most part. So this is coming, this is out there, it’s on the horizon. What do I need to do to make sure that I’m not left behind on this? Because I imagine that this takes a little bit of getting your house in order, getting my data in order. What are some things that I can be thinking about in my law office to be able to strike on these things?

Drew Bloom:

I think the most important thing that comes to mind when I think about this is you’ve got to have programs that allow you to access their data on the web. A lot of us have already built systems where I find a way, let’s say using my SharePoint to kind of integrate across systems because I already use copilot and I’m sort of treating that as my single source of truth. The closer you can get to having integrated data sharing in some way on some platform or across multiple platforms is going to give you that inroad to be able to then work with those programs. Agently, the ones that are I think going to be a lot more difficult to do this with are ones that don’t have that sort of public web access or secure web access that let you work with those utilities.

Zack Glaser:

Okay, so I need to get my data in the right place that is accessible by these sorts of things. Do I need to structure my data in any sort of way? Do I need to get that data already talking to each other? I have been reading and looking at a lot of things that are kind of saying, I see both sides or I hear both sides of like, you don’t have to super structure your data right now because artificial intelligence can go in there and kind of suss it out. But then on other sides it says if you would structure it, if you would just structure it basically, and by that I’m talking about structuring your case files in folders that have the appropriate matter as opposed to just dumping everything into one place. So just even that kind of small structuring will, the tool can extrapolate things off of that, that is still giving it more data. Do I need to be thinking about getting my data in order?

Drew Bloom:

I think there’s a lot of cases where putting your data in order is going to be hugely beneficial. Now, you’re right that on a surface level, AI doesn’t necessarily need the data structures that we used to have. If it can search things using these new kind of semantic search offerings, which is just to say that it can use natural language search and go find stuff inside of a really, really large set of data. The issue that runs into where we still need traditional data management is that what if you have multiple things that match that search query? And inevitably, I would imagine every single law office is going to run into that problem. How many different documents of a certain type do we have so many, and are they labeled in such a way that the AI can distinguish among them? And are they set to be retrieved in a way that the AI can even keep track of whether this comes from this document or this comes from this document. So there’s absolutely a lot of structuring that has to be done to know where the data’s coming from and know whether or not you’re finding a bunch of stuff that’s not really what you’re looking for or stuff that’s actually right on the head of what you need.

Zack Glaser:

So that actually makes me think more about metadata as opposed to the actual, what I would think about as the file cabinet structure, it’s more metadata. Because one of the things I think about with this, when I think about cleaning up my data, it’s when I was running my law practice, we had tons of leases, tons of leases, and most of them were outdated. Well, I knew which ones were outdated, I updated my template or whatever. But if I were to set a thoughtless AI product against that data and didn’t have a way of indicating that this was outdated, I mean, I’m talking about just having a date of creation even on that data because yeah, I could say anything that’s created before blank date don’t take into account, but I still have to have a data creation. So the more metadata I can have on this about the document, then I think probably the better off we are.

Drew Bloom:

That is exactly true. Now, the interesting thing here that we have to think of is many of the vendors we work with in software, they already do a lot with metadata.

Stephanie Everett:

And

Drew Bloom:

So when they’re integrating AI into their own products, they’re often leveraging that structured data that they already have. So I think maybe the important question we have to ask ourselves, is my data in a place where it’s getting some kind of structure from the vendor that I’m using? If not, then I might need to apply my own structure. I might need to bootstrap that myself. I might need to ask myself if this should go to a place where a vendor can help with structuring that data. But at the same time, an interesting thing that’s possible as well is what if I just took this data and I structured it myself? It’s another thing that AI can help with, and there’s a bunch of products on the market that do just that.

Zack Glaser:

Lots of options here, lots of options, and frankly, to me, lots of excitement as to where things are going. Unfortunately, for me, unfortunately for me, we’ve got to wrap up. I don’t know how everybody else feels, but I can’t talk to you for three hours here. So if I were an attorney that was just trying to get my head wrapped around agents and agentic ai, where could I go to play? Where could I go to mess around that’s safe that I could just kind of figure it out? I had Sam on Sam Harden on a couple of weeks ago, and he was saying, we were talking about vibe coding, and he was saying, make games. What could I do to play with this?

Drew Bloom:

I think to understand the concept, to understand what is going on. If you have a preferred AI client, be it Claude, be it Chachi, bt, be it copilot, see if you can start using some of their connectors. They already offer connectors that they’ve built themselves, and they often call these connections or connectors

Under the hood. These are usually cps. And so what you can do is if you’ve got a connection that actually goes to a place you want, I know Chat, GBT offers SharePoint. I can think of Claude offering just about anything you bring to it. I think that Copilot offers a lot of their own Microsoft ecosystem connections, but playing with those connections and experimenting with what they can do, even something as simple as asking copilot, can you find my X, Y, Z? And teams helps teach you how those systems are actually operating and how to optimize the work that you’ll do with them. Because the underlying skill is going to stay the same. How well can I send a succinct natural language query to a model that causes it to pick the right tool, find the right results, and then return them to me in a way I can understand? And that is the kind of eternal challenge that every piece of software is going to try to solve, but it’s never going to take away the underlying fact that we as users, we’ve got to be the Kickstarters that give the right instructions for it to go and accomplish that task. And that, I think, is a skill that you build over time with practice. I

Zack Glaser:

Like it. I like it. Well, drew, thank you for helping me understand this stuff. I know that I actually am coming away from this with a better concept of agentic ai, what it is, and then how we can potentially use it in our offices. I really appreciate your help today.

Drew Bloom:

Thanks so much for having me on Zack.

Continue Reading