Agentic Barometer Dashboard

[00:00] Hey folks, I'm happy Tuesday, but I'm off tomorrow. So it's kind of like a Friday. So great so Wanted to kind of quickly show you this. This is a tool that I'm making internal for us

[00:10] I think it's gonna be really helpful just to get a sense of kind of where everything is and what's working and how the whole

[00:20] How our agentic barometer AI stack is actually running So This is just a mock-up. It's a mock-up made with Claude

[00:30] So I can you know or with Figma make via Claude so We can click in and I can sort of show you how it's intended to work But it will not work until we have the mini set up and then once we do I can kind of install this and have

[00:40] Claude code connected to all of our actual data. So all the stuff in here too is all nonsense There's no actual real methodologies or goals or voice and tone or anything because this is just pure pure pure mock-up land

[00:50] But wanted to kind of give you a sense of what it will do so the idea is that you would log on And then there'll be some sort of basic user thing and kind of like, you know

[01:00] So so each person is so it knows who you are and what changes you've made But basically you have the overview we have research

[01:10] Sources the agents and then settings right? Those are sort of the tabs across the top so on the overview

[01:20] This is showing basically like the number of reports run right sources covered the average words blah blah blah These are basically nonsense This would be the first thing that we need to figure out like what would be the most relevant things to actually show

[01:30] Up here at the top the idea is it's sort of high-level takes on what what has been going on say The past month past 24 hours like we could kind of pick

[01:40] Then we have this map piece And so this is really where I think the majority of like our interactions will live This the idea here is to kind of give you a sense of how the the the actual stack is built

[01:50] So it's sort of laid out kind of vertically Such you can see how things are being ingested sort of what knows about what? So what does that mean?

[02:00] Obviously, you know at the bottom here is the whole thing is going to be hosted on cloud code using the app locally And it's going to be running Opus 5

[02:10] We're going to need Opus 5 high at a bare minimum for all this just for the sheer number of content or the number of tokens that are there Built on top of that are these six MD files, right?

[02:20] We have the voice and tone file We have the goals methodology so on and so forth and then we have truths truths are treated slightly differently And these are basically they're more harsh, right? So the guardrails and trusted sources

[02:30] Trusted sources is where we would define what those sources are Explicitly point them out or connect API's or MCP's later. So, you know note there

[02:40] And then the agents on top and all of our agents are our MD files, right?

[02:50] That's important because this dashboard is not going to control the agents. It's because they are not alive All of these are basically scheduled runs that are always happening and we're calling them agents because they're making a few

[03:00] Agentic decisions, but they're not an always-on open call type situation yet, right? That's where we went ahead to but we need to kind of define all this other stuff first to make sure that

[03:10] We trust it running once at a scheduled time instead of all the time So each of them of course then have outputs that we can then drill in on

[03:20] So the idea is that you can kind of click in on one of these and see what it is And then edit it. Obviously, there's like nothing in this if I were to grab

[03:30] our actual Work and tone. Let me just so you can see what it would look like your voice in tone Copy

[03:40] So if I were to paste this in and save you can see that Way to go that it, you know Renders it out and renders out the tables and stuff and then allows us to kind of edit it

[03:50] So if you're like, okay, I want to change what the thing is Then you can we can go in and edit it idea being is that it saves it's going to save that empty file

[04:00] Such that the next time that whatever agent runs, that's the version And so the same thing applies for the the drafting agent So obviously this is like what it put in for the draft prompt ours is is much much longer

[04:10] So I can show you what ours looks like It's closer to that

[04:20] So you can see that there's a lot of stuff in there and it's calling out the other The other pieces of content in here as well, right? So that's the idea and then that way you'll be able to kind of see what's going on and then of course drill in on

[04:30] The research it made so that's sort of the next piece is jumping over to the research. So Every time that a research report is created or any output is created

[04:40] It will basically show up in one of these and you can kind of drill in on them and take a look at what they Are at the top we have the ability to pin it

[04:50] So this would be something that is, you know pinning for us for later like oh, this is a good thought or whatever We can determine if there's any functionality

[05:00] AI perspective after that later. Included, basically this means that this piece of research is going to be included in the drafting or in the methodology. So if we're like, oh,

[05:10] this is not the right thing, we just exclude it, and then it'll be ignored when we kind of keep going. And then there's a memo, and this is a memo for us humans. The AI does

[05:20] not read these memos. For those memos, the ones we do want it to read, we can leave feedback here. Now, what's going to happen is it's going to add feedback into that research report

[05:30] MD file such that when it's being read again, the feedback will take into consideration. So if we're like, oh, this is actually not that relevant, but this part works great or

[05:40] whatever, that would be stuff to put in here. And that way we can kind of apply some feedback to the actual system so that it's getting a little bit better. We could also just look

[05:50] at which agents are doing what and kind of drill in on what those are. And of course, our research reports are very, very long. This is just kind of a simple little thing. There's also a list view and a calendar, so you can sort of see when the stuff was actually

[06:00] run, which I think will be helpful to get a sense of like, you know, when are we actually doing these? And this is based on when the MD file was made. It's really quite simple.

[06:10] Sources, this is going to be where we can add all the sources. So this is just a fancy wrapper for the MD file. Basically, it's a list or a table in the MD file. And to add

[06:20] another line to the table, we can do it basically here. And then we can, it's already built to do web, feed, or an API. So we'd be able to put all that stuff in and then have it

[06:30] can continue. Future state would be that to show a history of sources here as well. You

[06:40] basically be like, here's all the sources that have been pulled for other research listed out. They would all of course be trusted resources, but they would be specific instead of having

[06:50] to kind of drill in. And then the agent files. So this is where, again, we can kind of change what the actual agents do. The global rules are there, and then like the drafting agent.

[07:00] You can see I pasted the drafting agent on the other side. It didn't save here because it's a mock-up. But again, this is the same thing. You can change the color if you want. And then that changes it for the research itself. Very fancy. But again, another place

[07:10] that if you wanted to edit the way that the agents are running, we can edit it here, save. That will save the MD file. And then the next time that they run, that would make the change.

[07:20] All of those things that I just went through are accessible here. Editing the drafting agent can also be done there, or done here. Same thing. So it's really just a way of kind

[07:30] of like drilling in and looking at the different pieces. So anyway, that's where I have so

[07:40] far for this. Obviously it's a dumb mock-up. I'll share with you so you can kind of click around and play with it or whatever. But that would be the overall, that's the overall idea. If there's other stuff that we want to add to it, that would be really cool. One of the

[07:50] things I'm going to have it do is, you can see it riffed on the spelling based on my voice note, but one of the things I'm going to have it do is allow us to change the project

[08:00] folder because we can have any number of agents running against different research sort of

[08:10] folders and stuff. So instead of us just looking at the MD files in one, we could say, well, we have this set running and we also have this one running. We want to compare and be

[08:20] able to kind of switch between those two. Yeah. So, and you may be asking like, why is this so much faster to build than like the viewer and all that? And it's because this is internal. We don't worry about like any sort of security. It's going to be the

[08:30] most unsecure thing basically because it lives on a mini, behind a private network, behind a VPN as opposed to having to make it kind of like web accessible and all that. It's

[08:40] also going to look like crap on mobile. Like there's just going to be a whole bunch of things that not going to spend time sort of working on just to kind of give you a sense of lift and how these things tend to go. But anyway, let me know what your thoughts are.

[08:50] Sorry for the very long video as usual.