Why AI forgets your nonprofit, and how to give it a brain

0:43Scott

I was having a conversation with a president of a nonprofit. We were talking about business, and they sell things. This is what we do. And so we were talking about packaging all of their things that they do that many different ways that you could just package things up. And it was, oh, man, there were so many options.

And so as we were talking through all of these different options, we were recording conversations. And then there were other assets like spreadsheets and all this kind of stuff. And so I took it all. And then I sat down with Claude and I just spoke out. All right, here's what I want to do is I want to build a builder.

I want to build a product offerings builder. There's a lot of complexity to this, Claude. And I just started laying out all the complexities. And you'll find this and this thing. And I just kind of like onboarded Claude onto the problem, the intended outcome and solution we're looking for.

But ultimately, what the output was, was a interactive and modifiable HTML that allowed them to go through service by service and select multiple different options and just make all these tiny little decisions. And then after the hundreds of decisions that each of them are fine making, then they get to see their final output and how it all puts together because then there's an output page like the final thing. And so you just build all of this calculator thing that would legitimately a year ago have taken years to build, probably, like this would have taken a lot of time to get done and you're doing it in an afternoon. And it's bizarre the world we're living in. Yeah. And I

2:22Colleen

Think that like, okay, the core assumptions that we're making right now is that you've already, like this goes, you can do this without all of these things, but this goes better and faster when you've got an AI brain under the hood, when you've dialed in like your intelligence layer under your AI so that you can just get after these things. It is really difficult if you were working just with an AI by itself because you're going to be constantly working past its own automatic output or you're going to have to build an insane amount of context every time, right? AI

2:57Scott

Slop, it gets produced by no context given before expecting good output. You know what I mean? That's what AI slop is. So there you go. But if you want good output from AI, you have to give it good context.

So then your core question is, how many times do I want to give this AI more context? One. I only want to do this one time, you know? And if it's I only want to do it one time, then it's like, well, I know these things don't remember me like they say they do. They won't remember everything. It's like short term memory.

You're dealing with a goldfish. You know, that's what AI is. But what you need to do is say, hey, that goldfish has enough connectivity to connect me to something that will always remember. And that is the brain. And so you got to get a brain or you have to build a brain every single time you

4:15Colleen

If somebody, if I had just like 10 minutes that I could spend with somebody and help them quickly get better at using AI, what would I what would I do in that moment? And I was just curious, like, what would you show them? What would you do? What an

4:27Scott

Interesting question. All right. So if you had 10 minutes to help someone get better at AI, when everything First started out, it was like you could chat with it and it was pretty amazing that everybody really started getting into prompt engineering. It was the idea that, like, if you said the magic words, you said this and then everybody's like you, you'll get this magic response out. And still those things are real.

Like if you say, hey, tell me all about myself from what you know from me. And then it comes back some pretty cool stuff like I get that. But the one thing that it doesn't do really well is this idea of long term memory. The idea that prompt engineering still is alive and well, but you actually need to bury that deep inside of this database. And so if you spend a ton amount of time just telling AI everything about yourself, number one, everything about your business, number two, everything about everything that you want it to kind of have some sort of connection with or whatever.

But here's the key is that you have to store it somewhere. You have to store it somewhere. And the one that we use is Notion. And the reason why is Notion has a very easy MCP connector. And my point in all of this, the one thing I would do, 10 minutes, I would say start a Notion account and then train your AI to always store everything in Notion.

And then over time, what really is wild, what really becomes very cool is that over time it will store and learn and see all of these insights that you could not previously see. And so what most people do in AI is they treat it like a tool. They just say, I need it for this one moment for here. I'm going to use it for 5, 10, 20 minutes and then I'm going to move on and do something else. And all that is great.

But if you have a thing, if you have something you repeatedly need to do that most people build upon the thing, like it's not like I just need it for this one little moment and I'll just tell it a few things and it'll give me excellence. No, no, no. There's so much context around doing good work around everything. And that context is what the database is in Notion. They will constantly understand what the real situation is and then perform really good work for you. But I would say a brain.

10 minutes. What do you teach somebody? It's a brain.

6:57Colleen

Just chasing those outliers saying, OK, why did this dip? Why isn't this performing the way it is? Let's go dig into all of the things and see if we can find it. It's hours and hours of work. Whereas now I feel like we get to just put our energy towards the fix.

Like what's the what's the answer? What's the next thing that we're doing? Which is really creative work. It's really like strategic work. It's fun work.

Like you said, I feel like I'm constantly moving things forward in like all aspects of my day. Yeah. And

7:25Scott

Deeper and richer. Like this is like you get better output, don't you think? Oh, I

7:31Colleen

Do. I do. Especially if you approach it right. Like I think this idea that like I can only produce slop is like honestly just a signal to me that people don't know what they're doing.

7:41Scott

Yeah. And that's OK. Everybody starts there. As a matter of fact, I felt like that for a while. And then I just realized that truly the measure of output and the greatness of the output is 100 percent correlated with your input. It is. It just is.

If you give it all the context in the world, just kind of like stuff that you wouldn't even think about telling it, just jam pack it and then you say, hey, this is what I want to do. You just give it an end goal. And then all the context in the world. And then your prompt doesn't have to be magnificent because your prompt is hidden in all of your data underneath. And if you just said, hey, do this.

Now, if it's like I'm not talking about design or anything like that, I'm just saying like this is what you can do because of that. It's it's a wild life that we're living right now. I've seen

8:31Colleen

Several people do this where they give it a spreadsheet and then they tell it how to clean up the spreadsheet rather than telling it what a clean spreadsheet should be. So it's like, don't tell it to go delete these columns and these columns and these rows and all that, like to say this is what I want. I

8:46Scott

Really love that analogy. That's exactly what it is, is that there is a process. But when you explain the process to AI that you perform, you're assuming that you are the craft. You are the master at making all processes. And this is the best one that could ever happen for this particular thing.

And you're not like AI is way better at these sort of things than humans. And so if you just skip all of that, it's almost like you have to think a bit more like you're an adolescent. Meaning like you don't I don't presume to know how to make sandwiches when I am three. I just want to tell mom I'm hungry, you know, like I'd mom I'm hungry and then she's like cool I'm gonna make you a sandwich. What's that like?

Or I mean, at three, you know, which is, you know, I'm saying like, yeah, no.

9:34Colleen

But like I wouldn't go into like a gourmet restaurant, like a five star restaurant and tell the chef how to make their food. I might think I know how to make it,

10:27Colleen

Questions that stuck out to me ultimately, I think it's this tension between feeling like they have no capacity. And yet seeing that this can get them capacity, but maybe just feeling a bit afraid about where to start. And and the first big question that we got a lot was tell me about what tech I need to buy. Is this going to be like too expensive for us to maintain from a capacity standpoint? So like, what does that look like?

So let's talk about it. Let's talk about what what are the what's the core tech stack that we think nonprofits need to get started with an AI first approach? Yeah.

11:12Scott

So first and foremost, I'll take the easy one. This is the Jesus in Sunday school answer here. You know, you have to choose what your LLM. Is, large language model. Your AI is the AI when we think AI and then most people think ChatGPT.

So that's what we're talking about is choose your ChatGPT. That's not that's like Kleenex, the facial tissue. But the point is, you have to choose whatever yours is. So for like 85 percent of what we use, 80 percent, I don't know, is Claude. And so we use Anthropic Claude.

It's the one that feels like the most on the frontier for us as it relates to work. Gemini does a good job. But our point is, is that, number one, you got to choose a large language model. If your nonprofit locks you down into one, the one you have is probably fine to do a lot of the stuff that you need to do. It may not be perfect, but we can find its edges.

No problem. But that's really where the first thing is. That's tech stack number one is you got to figure out what your AI is. Yeah. Number

12:14Colleen

Two is you need a place for an AI brain. So we like Notion for that. What's a

12:21Scott

Brain? Yeah, a

12:22Colleen

Brain. Think of it like your medical chart. So like your medical chart, when you are part of a hospital system, has all of your health information fully documented, all of your history, all the rules about caring for you in one single place. So that any part of the medical team that works with you has all of that information before they make any decisions. That's what a brain ultimately is for your LLM, your AI tool. Yeah. And

12:49Scott

The way that those two work together, I think of it like this. I don't know if you never talked about it like this. It's like there is a place where you store things and there's a place that you need to think about those things that you've just stored. It's almost like there are zippers over here and then there is a zipper factory over here that gets the zippers to put together. And they're not the same thing.

And so there is a place where you store things and there's a place that connects those things. And so the place that connects those things, that's the AI, that's the thinking. Like it actually does thinking. Like really, truly thinking like it looks and it's like here, here, this, this makes connections just like our brain does. And so that's

Where you're talking about that we need a place to store all of this information because the amount of information that holds your life, your business, your role, your nonprofit, whatever, is bigger than what Claude or ChatGPT or Gemini or whomever can hold. And so you need to have a place to continually tell your AI of choice to say, grab this and go store it.

15:10Scott

Advice is don't think about AI as something that you can get something from quickly. If you don't think of it that way, but rather, and this is going to sound bizarre, don't ask anything of AI for a while. Only you tell it things for a while. Like the more that you can spend telling AI about you, your context, everything in your life and have a place for all that to get stored, some database, a brain, you know, Notion like that, just store something somewhere. And then you just do that over time and tell it more about yourself and give it more information.

You could, one thing I did, I went and I gave it all of my health information and it is finding things that, oh, man, that is cool. Like stuff, yeah, stuff about myself that I'm like, I didn't, I had a recollection this happened that, okay, huh, that's interesting. So I'm making connections that I hadn't previously seen before because I'm giving it context. And so that's, I think probably general AI advice, that's my first one is if you can kind of have a brain of like, all right, if you can imagine that AI is a little bit like onboarding an employee. And every single time, if you're a business owner or whatever, if you've ever hired somebody, you kind of have to know that that for a period of time, you are just paying their salary for them to learn.

You're not paying their salary for that, for you to get good output from them for a while. You're just truly paying for them to learn. And that happens for a really long time, like really long time. And if you have that same thought with AI, that same mentality, you're going to go a long way. Because if you're an employee or employer to an employee, you know, inherently, I need to teach them this, I need to teach them this, I need to teach them that.

Because you can't just say, go do the work because they'd be like, I don't know what anything is. But AI, you just think, oh, well, it knows everything in the world. It doesn't know this, doesn't know anything about you. It doesn't know everything in the world. There's too much in the world to know.

It only knows what the Internet knows. That's essentially what it is. The Internet doesn't know you, Colleen, and what you ate last night and how you're feeling about your friendship with this one person. Do you know what I mean? Like there's all this information that if you just say, oh, I need to teach it about my world, it knows everything in the world, but now it doesn't know anything in my world.

And so once you teach it everything in your world, then it knows everything in the world. Then we can really start to see some major gains, some major impact here. Yeah,

17:54Colleen

And that compounds, like it compounds a lot over time as you're working with it, too. And the prompt that I put in this chat, it is not a good prompt. So help me put together an event promo package for the nonprofit AI Intensive. We just changed the date and are catching up other assets. Day one starts 9/22.

So that's a bad prompt, right? There's no context. There's like it is such a direct thing. But the reason I can do that is because of the compounding effect of lots of context over time and building that system.

18:33Scott

There's three ways that AI is going to get context. It's either from what you tell it in the prompt, from what you tell it that lives in a brain or a database, or it will make up context. It will assume context. And so it will fill in the gaps. So what you're saying here is that it's a bad prompt because it has no context.

And then you said, hold on a second, the context doesn't live in the prompt. The context lives in the brain. And the reason why that is important is because if that didn't happen, if either of those two things don't happen, you will get context. It will just be the wrong context. You

19:09Colleen

Got it. You got it. What happens between here and there is what's really interesting is it runs skills that are tied to AI employees. It reads memories. It goes into our brain and Notion. It adds tasks for itself and process.

It goes to our actual landing page and looks at like, what are we saying about this? Goes through several different Notion articles about different like strategies that we've already documented details about this. I had previously worked on a different promotion package for this, so it was like read through all that. Look at our meetings, because again, this is all part of that like training that we've given this AI employee, which is like, hey, context is important. Go look at what we've actually said about this.

And then it did all of the work without talking to me. And then it was done. And I had a really good handoff for a event promo package that I took right into Claude Design. And I was able to pretty quickly get V1 of ads drafted. I sent them over to you yesterday just to say like, OK, that was easy.

But that that is the value of what you just said, which is like spend time, spend time giving it context, spend time building context. Not just for you, but for like all of the work you're doing, because it can become very, very efficient in helping you get there, but not immediately. No. And that

20:36Scott

Thing is just having the right kind of like right expectations on it. If you can just straight up say, hey, for a bit of time here, I'm just going to be spending time just pouring into this database. And then I'm going to connect AI to the data. You know, it's just like, oh, and then that connected AI to the database is just, oh, now. Oh, yeah. Gosh, it's just fire output all the time because you've told it everything.

You dumped everything. It's just like, let me tell you everything, man. Just think of all the different areas that it would need to know. Think about I think maybe to think about is that any time that you are making a decision, if you can trace your decision criteria, think about it. How are you making a decision?

Well, I need to do it. But this person, they might hear it. OK, then I need to think about this schedule. But that might impact this. Literally what you just showed is what that is.

You know, you might have a simple thought saying like, oh, I'm going to go to this track meet tonight. But then but then you go through the well, I need to do to do this and let me search my Notion. Let me do like all of those things happen in your brain. And then you process it and then you go, here's what I'm doing. Here's my output.

And so you just showed the inner workings of how it's thinking about stuff from a simple thought. And it has all this context. Now, if it didn't have context, it would not search Notion. It would probably search the Internet for saying, oh, hey, something general about AI, how can they approach that? And so it will search all these things and it will come back and it will be absolute garbage.

Like you will look at it and be like, this isn't usable in almost any respect. And then you'll be frustrated and say AI is junk because it's AI slop. And it's all because context. Good context in, good context out. Bad context in.

No context in is made up context out. So