

Get featured on the show by leaving us a Voice Mail: https://bit.ly/MIPVM
This episode with Joris de Gruyter explores how advanced AI agents are reshaping individual productivity and technical work. The conversation covers coding agents, personal agent systems, automation beyond code, and the risks of burnout, privacy, and memory misuse. Real-world examples show how AI can turn notes into workflows, handle testing and reporting, and operate as a team of specialised agents. The key message is clear: applied AI skills now combine technical fluency with strong boundaries around focus, context, and trust.
👉️ Full Show Notes
https://www.microsoftinnovationpodcast.com/821
🎙️ What you’ll learn
- How coding agents are expanding into everyday knowledge work
- Practical ways to turn notes into repeatable automation
- When to use multiple AI agents and when to limit them
- How memory features can introduce risk and bias
- Why APIs matter for an AI-driven future of work
✅ Highlights
- “I realised that I had incredible power to create anything my mind can conceive digitally.”
- “Tokens have become my money.”
- “The productivity gains is causing this tremendous burnout for some reason.”
- “I’ve put the guardrails in place so they’re not allowed to talk to me after 9pm.”
- “It started with notes, and it’s turning into automation.”
- “The internet is a very hostile place to AI.”
- “Memory is the first thing I turn off everywhere.”
- “I don’t want anybody else’s code in my mix.”
- “It allows you to become almost like superhuman in a weird way.”
🧰 Mentioned
- AI Brain Fry: https://hbr.org/2026/03/when-using-ai-leads-to-brain-fry
- GitHub Copilot: https://github.com/features/copilot
- Work IQ: https://learn.microsoft.com/microsoft-365/copilot/extensibility/workiq-overview
- Claude:https://claude.ai/
✅Keywords
ai agents, github copilot, coding agents, automation, productivity, ai burnout, ai memory, privacy, workflows, devops, markdown, applied ai
Microsoft 365 Copilot Adoption is a Microsoft Press book for leaders and consultants. It shows how to identify high-value use cases, set guardrails, enable champions, and measure impact, so Copilot sticks. Practical frameworks, checklists, and metrics you can use this month. Get the book: https://bit.ly/CopilotAdoption
If you want to get in touch with me, you can message me here on Linkedin.
Thanks for listening 🚀 - Mark Smith
00:00 - From Occasional AI Use to Total Workflow Takeover
01:15 - The “February 14th” Inflection Point: When Productivity Went Vertical
03:05 - Token Burn, Creative Addiction, and the Rise of AI Brain Fry
05:10 - Building a Personal AI Army (and Forcing It to Respect Your Life)
08:05 - When AI Leaves the Screen and Starts Optimizing Real Life
11:30 - The Invisible Layer: Turning Notes into Autonomous Systems
19:19 - Power vs. Privacy: Why Memory Is the Most Dangerous AI Feature
00:00:01 Mark Smith
Welcome to the Copilot Show, where I interview Microsoft staff innovating with AI. I hope you will find this podcast educational and inspire you to do more with this great technology. Now, let's get on with the show. Welcome back to the Copilot Show. I'm here with Joris. It's great to have you back. Let's discuss what's been happening in the last three months in AI land.
00:00:26 Joris de Gruyter
Three months. So yeah, I guess you always have a bit of a delay. So we're end of March right now, right?
00:00:31 Mark Smith
Yeah.
00:00:32 Joris de Gruyter
Yes. I'll start with, you know, because I want to hear from you as well. But for me personally, I think all of the coding agents, Claude, Codex, GitHub, Copilot, I feel like there's been a huge leap there. And it's not just the models, but it's also sort of the software and how they use the models, like their workflows, their to-dos. And so I found beyond coding, a tremendous use for all of these tools for just any type of work I'm doing. And that's, you know, that's how I pinged you, right? I was just looking at myself. I remember our previous conversation, like, man, I got to ask Mark what he's doing, whether he's seeing the same. So I'm glad you said three months because that is exactly sort of my timeline as well.
00:01:16 Mark Smith
Yeah. So if you're listening to this and you haven't heard your before on the show, there's two or three other podcasts with him on. So go back to the back catalog on those just to get his background. The defining day for me is February the 14th.
00:01:30 Joris de Gruyter
Okay.
00:01:30 Mark Smith
As in, that when things really started to accelerate away. And the reason I can know that is by looking at my GitHub activity. It goes, I've never been on GitHub in my entire career. I've had an account, I've done some little things around the edges. I'm not a developer. And now I'm hard, you know, dark green squares and I am maxing them out like tokens have become my money.
00:01:57 Joris de Gruyter
Yes.
00:01:58 Mark Smith
The amount I'm burning a day is ridiculous. And it's not because I feel I'm wasting them because I've created a whole token optimization system for myself. But it's the closest thing I can describe it to has been when you're a teenager and you're gaming and you could just burn hours gaming. Like it was just all of a sudden it's 1:00 in the morning, it's 2:00 in the morning and you're just doing, I just got to, I just got to do this one more. I want to deploy this, see how it works. Like, and it's, that's the, that's what I feel like I've been in. And although I say February the 14th is when GitHub really started to light up, I was going from about late December where things really started to change for me. And then I started to, put the rigor in place, which is why it really started to light up from, I realized that I had incredible power to create anything my mind can conceive. digitally. And that's, and that wasn't there in the start of December last year.
00:03:01 Joris de Gruyter
No, agreed. I, the same actually. I probably know February is probably the same for me when like January, I was seeing all this stuff and I saw some internal demos and online stuff. And I'm like, man, I need to get my hands on this because this is looking really, really good. And that same like early February, I started using it. And it's at this point, it's an addiction thing. I agree with you where you're like, all of a sudden you're like, oh my God, it's past midnight. I got to go to bed. In fact, I wanted to ping this and you need to put this in your show notes. Have you seen that there was an HBR research called about AI brain fry? Have you heard this term at all? Yeah, if you Google it, you'll find a bunch of articles, but they'll refer to the Harvard Business Review. They did this research study that talks exactly about this, where basically the productivity gains is causing this tremendous burnout for some reason. I mean, some of it is just because you have some people have multiple agents running at the same time and there's constantly context switching and you have to sort of architecturally or whatever, keep everything in mind. And they say as soon as you get to like 3 or more agents, people just completely burn out. But to your point, even with just with one project, it's gotten to the point where I've actually the last, literally the last week, it's been a focus of mine to sort of put limits on not just saying like, I'm just going to work on this. It's like, I need this and this done. And if I get it done in 15 minutes, then great. And that's this weekend as well. I don't know if you recall this. I love making games as a hobby. And there was a couple of things. And I finally decided to sort of put it in practice there as well. And so I pointed at this huge code base I have and like in 5 minutes, it's like, oh, here's the architecture, figured it out. Okay, what do you want to do? I was like, I need to do this, but it's a huge refactor, which would have probably taken me a couple hours to do. And like 10 minutes of back and forth and the whole thing was done. And I was like, really debating, do I move on to the next thing or?And I decided, I decided not to and I went outside. So it's really gotten to the point where you have to watch yourself if you're not careful.
00:05:06 Mark Smith
Yeah. I've put the guardrails are put in place for the agents is that they're not allowed to talk to me after 9 P.m.
00:05:12 Joris de Gruyter
Okay.
00:05:13 Mark Smith
So I've said, that's it. You get cut off. You've got to. So I've made them all go quiet between that and they're not allowed to start talking till 7 A.m. in the morning to me. And that's their proactiveness. And I'm currently over 23 agents built in my team. I've created a couple of things, orchestrated a couple of things. I've got one agent's called Dobby.
00:05:38 Joris de Gruyter
Yeah.
00:05:39 Mark Smith
And I've got an acre and a half property and it's ingested everything from the government records on this property to the typography of the property to he accesses over 1200 sensors that I have on the property, which are, you know, smart home devices, weather stations, my solar power system. And now he is learning it and then coming up with suggestions for optimization. power, energy consumption, mind-blowing stuff like some situational awareness stuff with the war and stuff happening, right?New Zealand at the bottom of the world, guess what? We're running out of fuel.
00:06:18 Joris de Gruyter
Right.
00:06:18 Mark Smith
So I've been running now for about four weeks where it models where things are at in the world, the war, the government websites in New Zealand that show inventory as in we're talking about jet fuel, diesel, and petrol on hand and how many days we have left. And then even how many ships are on the water coming to New Zealand with fuel and like over six of them have already been diverted away because the company sending the fuel can get a better price somewhere else. And so I'm sorry, laters. And so it's allowing me to optimize.
00:06:55 Joris de Gruyter
Your other agent is a commodities trading agent or.
00:06:58 Mark Smith
No, all I'm doing is going. Is my family protected as much as possible? So gone out and, stockpiled fuel, all sorts of things, because what's modeling at the moment is very similar scenario to March 2020, pre-COVID, right? The government's already started midday announcements of restrictions coming. And it's just sound, it's just, it's mirroring COVID, but not as a, an outbreak. It's now an energy shortage crisis that's causing it. So that's just one of, that's just one agent that I have now monitoring all that stuff. Last night, It already had worked out with my solar to go in and change a whole bunch of settings on my solar system because I pump back into the grid as well. And we get quite good money for that and just optimize it even better to protect me in case of power failures, but also optimize my income generation off that.
00:07:59 Joris de Gruyter
Nice.
00:08:00 Mark Smith
That's just one.
00:08:02 Joris de Gruyter
Yeah, to be fair, it's on my list. Most of what I do is while at work and then just like personal projects. I haven't, some of the personal projects are, tied into personal life, I guess. But I have like, I'm a huge user of Home Assistant. I've had home automation for many, many, many years since it was like crapshoots. But my system's super stable. I have a pretty high spousal approval rate, which is high, difficult, difficult to get here. So that's all great. But I have been thinking like, what could I do there as well? So I haven't really looked into that stuff. Another thing like this weekend, talking about, crises and whatever, like inflation. I mean, Seattle's gotten so crazy expensive. Actually, were you here for the MVP summit? No, it didn't count. Okay. Because you would have seen like prices here are just insane. And one of the things we were comparison shopping the other day and we found on their stores like, oh my God, this stuff is a lot cheaper there. I was like, I should build this, just a way to track it, like from my phone, take a picture, have it track, like, oh, how much did we pay for this last time when we were at whatever store, And that's the other thing, right? We're talking about working late nights, but also the brain's constantly thinking of what else could I do, right? Every, yeah.
00:09:21 Mark Smith
Yeah, that example you just gave, I've been working on it this morning, which is all my grocery shopping, which is Dobby's job now. And like what I've realized is that the internet is a very hostile place to AI.
00:09:39 Joris de Gruyter
Okay.
00:09:40 Mark Smith
I would say it's adversarial to AI. And a lot of it came from Gartner Publishing that you need to lock down and don't let any AI crawl your site. Don't, you know, because it could prompt inject you, blah, blah, blah. So what's happened is people have invoked, let's say, CloudFare on all their server infrastructure. And if you smell like an agent coming in, boom, you're just pinged. The problem is it's now hurting e-commerce type businesses, I think, that actually want to genuinely shop on that website, but they are blocking all type of agents. Amazon blocks every, you know. I buy a lot of stuff out of China, like components. Like if I want to do, because I'm big on home assistant, right? I've got all those, I only know I've got that many sensors just because I had, two weeks ago, I had Dobby rebuild Home Assistant from Bear, but with it fully controlling everything. And I'd already got the architecture of like my floor pans, my spaces, all that, but it scaffolded that all up, then all the devices, then got really crisp on all the IDs. I even gave it access into my whole network. And it did MAC address matching to IP, sorted out IP of blocks, decided what needs to be fixed at Home Assistant, could add, just amazing stuff. And then the interfaces and dashboards, my wife, I just put her on it three days ago and she was like, this is amazing. And you know, for spousal approval, I just say. Just incredible because there's kind of like no limits. Even if stuff doesn't have an API, it goes well, let's create one.
00:11:20 Joris de Gruyter
Yeah.
00:11:21 Mark Smith
No, I agree. Let's work out what we can sniff there and let's see how we can integrate to it. And it's just blowing, just honestly, blowing my mind. But like, as I say, the internet is fundamentally at first real and I think that's going to change in the next 12 to 24 months as businesses realize everybody's going to end up with a dobby, right, that can go out and do their grocery order for the week, order whatever type of supplies, But they're going to make it their, I mean, ultimately, they need to create APIs to their website. If you're an e-commerce product catalog type business, and you want to make it super easy for people to acquire your products without forcing somebody to go through a UI.
00:12:06 Joris de Gruyter
Yeah, and I think that's probably to your point of the API. I think it needs to get to a point where we find the middle ground, right? I understand some of the scraping or botting that they don't want that to happen. I mean, think of, you talked about COVID or whatever. Remember like all these news reports of like people, all the title paper was, sold out and that kind of stuff. Like you'd have the exact same thing, they'd be gone in seconds, right, instead of hours. Because, so I understand that there's issues with that need to be figured out.
00:12:41 Mark Smith
At the moment, they've just blocked it, everything, right? So now, I'm learning, of course, now it's in a cat and mouse game, right? So what I've been scripting all morning here on another device, because I've got multiple nodes now with different agents on nodes. So my main infrastructure is all in Azure. And then I've got some physical devices. And you have to go through a process of even going, hey, how do we do a login credential that still looks very human in nature? doesn't enter passwords and things in microseconds, looks like they've been typed in and getting to that point of then going, okay, can we now use the search function now that we're in? And because I don't attach my credit cards permanently on these things, I go, I'm almost at the point now that I'll get everything in the shopping cart that I want, and then it will ping me and say, hey, it's ready for you to submit the order.
00:13:34 Joris de Gruyter
Nice. I have one thing and we'll talk about the soft line because I don't want people to start doing this and then they block it as well. I have found one thing that's, but I haven't used it at the volume that you do. So I'm curious. So we'll have to, we'll have to, I'll send you an e-mail if you don't have time after this.
00:13:49 Mark Smith
So what else are you touching? What are you seeing that's like, what has been that kind of light bulb moment in the last three months for you? What are you, what are you finding for yourself?
00:13:59 Joris de Gruyter
Yeah, I think coding in itself is the obvious one. And I feel like that's where this revolution is starting, because it's the obvious one. But what I've come to realize is that the coding also augments how you use the AI. You've probably seen that as well, where you're doing things like, oh, you know what? I'm going to just vibe code a little script. that then the AI can call later. It's more efficient. It's more sort of, you know, it's going to be correct. Same with like an MCP server, like, oh, they have like an API or something, but like, let's just vibe code and MCP over it. And now it's super clear. You can scope it to just a couple of tools. So it's also vibe coding the tools to support what you're doing with AI.
00:14:42 Mark Smith
Yes.
00:14:43 Joris de Gruyter
And then at work, I started basically, and this seems like ridiculous, but I started with just using it to track my thoughts during work, right? So the work I do, we have sort of these projects. It take like a week or two or so that we're doing this product. And I basically started with, hey, like. create a project folder. obviously, AI loves MD files. So it's like, hey, can you create, like, let's start with, here's the things that are top of mind for me that I want to make sure we test, right? And then, okay, now create a note for every day that I'm working on this. And then every day I was like, okay, I tested this that was on the list. Here's my observations, blah, blah, blah. So I started keeping notes. And then I was like, actually, I have to file all of these things in our system because we track it in Azure DevOps. I'm like, Azure DevOps is an MCP server, but that MCP server is huge. There's too much in it. So I basically, I was like, okay, I added the Microsoft Learn website. They have an MCP server. I'm like, hey, go on the Learn website, find the REST API for Azure DevOps, and just five code me an MCP server that just does this and this. I did that. And then I figured, we have a bunch of custom fields. So it's like, okay, you need to create one more tool to grab all the custom fields.And then we sort of went through basically in a conversation, like, okay, create a new bug for this, like what are the fields we need, blah, blah, blah. And then at the end, they're like, okay, now create an instruction file for yourself as a skill so that going forward, you know exactly how to do it, right? Because we just spent like 10 minutes going back and forth, figuring it out. But then boom, now you have a skill. And then I was like, okay, all these notes we've had, just file a bug for each of these. Boom, takes 2 minutes, goes through the whole thing, great. And then at the end, I'm like, well, I have to write these reports. right? But basically, they're based on my notes and on the Bugsby file, because you have the listing of the files. Like, okay, well, they're in Word. Okay, so then I've had coded a Word to Markdown. I know there's some existing MCPs, but our template is very specific and you lose a lot of fidelity going to Markdown. So I wanted to make sure some of that. So then we started tracking that. And at the end of the week, I created the report, which is great. I had to do a lot of editing, but it was saved me quite a bit of time. And then the week after, I was like, I could do more with that. So I started really leaning into the note-taking and sort of the, during the day, talking to it. And then...
00:17:05 Mark Smith
Are you using audio to talk to it?
00:17:07 Joris de Gruyter
I've done that a little bit.
00:17:09 Mark Smith
Okay.
00:17:10 Joris de Gruyter
I want to do more of it because it's just a lot of typing right now. And I... I have, but not a whole lot. But then I was thinking, okay, now I'm having conversation with product teams as well about the things I'm finding and I have to sort of report to our leadership. So then I augmented the report the next week and it's like, okay, you have my notes, which turn out a lot more descriptive. There's more in there.But also I want you to go look. So I added the work IQ MCP from Microsoft. You can get all your stuff. I was like, hey, go look at this conversation I had with the product lead and with my leadership. Because from that, you can distill the worries that I have and sort of put that stuff in the exec summary, right? And so now that report was like 85% what I needed, right? And so it just snowballed into this. And then, I started doing, okay, I can connect Playwright to my browser. I'm like, okay, now that you've seen, can you just read the chat that I had with, say, Copilot or whatever? Can you just distill the basic parts of what I did here. And then maybe let's repeat it next week, right? What if they make fixes or whatever. So now I'm starting to have basically participate in some of my testing as well, or reading what I've done in the browser and just taking notes without me even asking like, hey, go check the conversation and grab what we did. And it's just ***********. And it's just started with notes. And it's turning into automation.
00:18:37 Mark Smith
So are you using like GitHub for maintaining all that? Or are you using Azure DevOps or something else?
00:18:46 Joris de Gruyter
It depends. We've always done that where if we write specific scripts or something for like these projects that we do. we sort of have an internal Azure DevOps where we stored at so that we can always go back if another colleague wants to test the same product, they can go look at, you know, but we have to, you know, obviously there's a lot of NDA stuff. Sometimes there's like bad content that we have to worry about or legal issues, you know, those things happen. So we have to worry about that.
00:19:12 Mark Smith
But that'd be part of all your context, right, that you'd want to feed into the context?
00:19:17 Joris de Gruyter
It kind of depends. So the skills right now, we just started a new repo internally for just for our team to just start making sure we track it somewhere and then people can share. And we plan to, I set up now a weekly meeting and there's only a handful of us, but it's sort of increasing. People are more and more interested.See, hey, if some of these skills turn out to be a lot of people are using them, let's promote them to a marketplace. Because now GitHub Copilots allows these private marketplaces. It's a preview feature. So I've been playing around with that to say, okay, now we could promote specific skills, say like a playwright skill to interact with M365 Copilot or like my Azure DevOps MCP that's very specific for our team on how to file bugs. We can put it in there and then anyone on the team who wants to get started can just pull it down from that private marketplace. It's easy to find. So Again, it's just tons of experimentation going on right now, but it makes sense. I mean, the fact that you asked the question, right?
00:20:20 Mark Smith
Yeah, it's so interesting. How are you thinking about things like, or are you thinking about things like memory management as in particularly around, you talked about across the week, right? You've got a lot of things happening and there would be an element that you'd want your agents to start. memorizing some things, but not everything. like you talked about, you dip into specific cases where confidentiality, et cetera, and you don't want that to jump frame. How are you thinking about things like memory, short, medium, long-term in anything that you're doing?
00:20:57 Joris de Gruyter
You know, just by nature of my job, I have a huge dislike of memory features for multiple reasons. I still feel like it's a security problem for the most part. I think pretty much across the whole spectrum of companies that have AI, there's been all sorts of interesting memory security things where the memory gets poisoned in some way. And then, you know, the other thing.
00:21:22 Mark Smith
Easy to happen.
00:21:23 Joris de Gruyter
Easy to happen.
00:21:24 Mark Smith
I'll add one phrase that's poison, like, and it wasn't a bad thing, but I noticed it persisted. which was, I gave it an example and I said, there be dragons there, which is an old mapping term, right? Yeah, And then a week or so later, and it goes, here's a dragon. And I'm like, what? They're like, you're still holding that context of, yeah.
00:21:45 Joris de Gruyte
So I mean, there's been some interesting cases. I don't know, it's starting Microsoft. This is the general issue with memory. that I have a little bit of a problem with. But actually, my main thing from a pure practical perspective is LLMs are all about context, right? And I need the ability to start a chat that has zero context. I don't want it to be tainted. And it mainly started for me with, obviously, in my work when I'm testing, you know, I cannot have cross-contamination because, you know, I'm trying to get it to do something, it refuses. Now if that leaks into the next prompt, like I'm done testing, right? So I, that's one thing. I have to turn it off just from, but also I feel like.
00:22:31 Mark Smith
How do you start that? How do you start with, you know nothing? Like, I wouldn't even know how to do that right now.
00:22:38 Joris de Gruyter
Each and every. AI product out there that has memory has a way to turn it off.
00:22:43 Mark Smith
Yeah, that would be wholesale turn off, right? Not just this conversation. I mean, I know they've got private mode, but yeah.
00:22:49 Joris de Gruyter
Okay. Yeah, but then private mode, you typically lose some features, right? There's usually like things that don't work. I think like in ChatGPT, if you do temporary, you can't generate images or something. Like there's a couple of things like that just... And also, I wouldn't say a privacy nut, but at least I do care about privacy. And especially here in the US, that's always been a problem because there's just not a lot of legal framework for that. And I'm still concerned about that as well. Like all these companies are blowing through billions of dollars a year. They're going to have to make money. And think about that, right? That data that they're mining there, that is gold, my friend. It is gold right now for training purposes, but it's going to be gold for advertising and everything, right? So.
00:23:30 Mark Smith
I'm holding up and I'm showing my aura ring, which I now haven't worn for maybe six months.
00:23:38 Joris de Gruyter
Okay.
00:23:39 Mark Smith
I was very, very good at collecting data and analyzing my health. And up on my shelf, I also got my wife's one. Until they sold it to Palantir.
00:23:51 Joris de Gruyter
Oh, they did. I didn't even know that. How did I use that?
00:23:55 Mark Smith
Man, and I'm just like, hold up.
00:23:58 Mark Smith
And now I'm, you know, you talk about privacy. Now I look at every small startup and I'm like, how long do you have to sell? Do you have to make money? And then that data set now becomes, we don't know who's. And I'm just like, the day Palantir bought that was the day, the last day I wore it and the last day my wife wore it and we're done, accounts closed, et cetera. Because, you know, who knows?
00:24:23 Joris de Gruyter
Yeah, these, I mean, these memories, like you talked about short-term and long-term, usually there's different types where they track. just specific facts, right? Like, okay, Mark is a male and he wears glasses and you mentioned New Zealand, so he probably lives there, that sort of thing.Those are facts, it's one thing. But then there's a lot of other stuff that goes into like long-term memory or whatever, where it's just basically profiling you, right? It's like, okay, based on these chats, we know that Mark is like a Microsoft guy, like he, you know, works in that ecos, that sort of thing. And that scares me as well, right? Because If you think about all the stuff that Facebook and Google, right, they're probably the two biggest data gatherers, at least from in big tech. I mean, they profile people as well. But if you think about how LLMs can feed into that semantically, right, because now it's no longer like words have to match to make it work. You sort of infer things.
00:25:22 Mark Smith
Yes.
00:25:23 Joris de Gruyter
And so I'm huge, like memory is the first thing I turn off everywhere. Like I use all the AI, maybe not all the AI products, but I use a lot of AI products and I have an embarrassing amount of subscriptions on them as well. But like Gemini, Claude, Copilot, like first thing I do is go in and turn off memory everywhere.
00:25:40 Mark Smith
That's so interesting. That's so interesting. This is probably why I've lent into building my own agent army. That it's not, you know, everything's through API and I'm not giving they're only getting part of what they're doing. Like I have nine different providers connected into my environment. And like my dev team, so for example, my developers on my dev team, which I think at the moment I switch between OpenAI and Anthropic for that. But the test team are totally different models and totally different providers because I'm of the opinion you can't mark your own homework, right? But that's all the way through that cycle. And so because of this API, they have a much limited view. I feel like I'm never now giving them the full picture of what I'm doing. I'm only giving them their little snippet of what they're working on. Because the memory has been held in my MD files on my server infrastructure, not out there. And I think that's, like, I can do that because I understand the tech. The layperson's not going to be able to do that unless Maybe by the end of the year, it has just been mainstreamed and that they don't have to worry about that infrastructure. But the thing for me, not worrying about the infrastructure is like, there's the risk. I do want to know where it's been stored. Like right now, I'll see somebody has built a skill. You talked about marketplaces and stuff. I've created an aversion to marketplaces. because I just think there's so much opportunity for people to do nefarious things in the cover of good.
00:27:12 Joris de Gruyter
Oh yeah, for sure.
00:27:13 Mark Smith
And so I will look now at a skill that everyone's talking about in the market and everything, and I will start a conversation with AI and say, hey, go check that out. Understand the architecture of it. Don't take anything of it. Just understand kind of What are they thinking, how they're working on this? And then I'll go and say, okay, now go out and find five other products that are doing the same thing in the market. Understand them. What would you do differently?And then once it's done all that research, because I'm really big on doing planning mode as opposed to building mode. And then I'll get it go, okay, now look for the last six months, any white papers and any academic research that has come out on these specific areas that are in the public domain. And it's only at that, bringing that all together, then I'll say, now let's build our skill.
00:28:03 Joris de Gruyter
Yes.
00:28:04 Mark Smith
Right, because I don't want anybody else's code in my mix.
00:28:08 Joris de Gruyter
No, I agree. That's where, like, what I was talking about, there's GitHub Copilot now supports these private marketplaces. Private market, which is great, because it's essentially, again, it's GitHub Copilot. I don't know how it works in other agents, but I'm assuming.
00:28:21 Mark Smith
Yeah.
00:28:22 Joris de Gruyter
They're stealing features from each other constantly. So I'm kind of assuming if one has it, the other will. If not already, they'll have it soon. And basically, it's like a Git repo that has to be structured in a specific way and has some like JSON metadata to drive it. But that's the way you can control it, right? Where we say, look, we have an internal and then sort of by committee with the side to promote something, a skill or whatever somebody builds. that we feel like is good enough for others to use and ready for that. So that's.
00:28:51 Mark Smith
And not everybody's going to have the ability to build skills at a level of sophistication, and therefore they need the marketplace, right? You're going to have colleagues that just, their brain's not wired that way. And so they need somebody like you to build something that, and being that you're in, I like this idea of private marketplaces, because at least you're now saying, hey, this is everybody that's in the, you know, in the team.
00:29:15 Joris de Gruyter
Yep, and it goes, I haven't really played it much yet with the agents and the swarms and that sort of thing. But that's sort of on my list for the next couple of weeks as well is like what I'm doing, can I scale that out? Especially for the stuff where it's helping me test or helping me retest. Like today as well, like before product team came back and said, hey, we made a fix for this thing. And I had actually had the co-pilots build a set of test. Test scenarios that I knew were going to have to retest. So basically I did like one-on-one with it, like testing it on the fly. And then in the end, I was like, OK, now write this down exactly what we did, where the issues were so that you don't run into this again. And so the team came back to me this morning and they're like, hey, we think we fixed it. Can you retest? So just spun up the.I use GitHub CLI, and that's just because I'm a nerd. But you CLI, I was like, hey, run this particular playbook. It went out and it's like, Well, it looks like it's not fixed yet, so I went back to the team and they're like, Are you sure? can you test this? And then I basically just copy-pasted their text back to get a call. I was like, Hey, here's what they're telling me. Can you go check that? They came back like, No, so I took a screenshot, sent that to the team, like, All right, and then as I went in our meeting, during the meeting, they pinged me. It's like, Hey, you know, we found the issue, restarting all the services. Can you retest? So then, basically during the meeting, I'm like, hey, they said they fixed it, just start a new chat, try again. Yeah. But just thinking about the fact that none of this is difficult, right? The testing that it did for me, it's not like nothing crazy, because I figured out exactly the steps on how to reproduce it. But the fact that it could do that while I'm grabbing a cup of coffee or while I'm on a meeting and don't have to pay attention to it, and just when I get back, it just tells me like, yes, it's fixed or no, it's not fixed. And here's a screenshot of the conversation. I mean, that is just crazy. And like I said, this is not, this is nothing to do with code. Even though I'm using a coding agent, I'm using this, you could argue for a technical task, but it's, I'm just interacting with websites, right?
00:31:25 Mark Smith
Yeah.
00:31:26 Joris de Gruyter
And that to me is, so I feel like, you know, you're seeing these things, co-work, right, cloud co-work, and now Microsoft's gonna have to co-work. I'm really excited to see where that goes, because I'm assuming it's going to bring sort of that whole workflows, to-dos, stuff that the coding agents can do, is going to bring that into the workplace in a manner that hopefully non-technical people can do this as well. And I'm excited to see where that's going to go.
00:31:53 Mark Smith
You know, and I've used Claude Cowork for, I don't know, maybe the last three to four weeks, quite a bitAnd I can see why that is gonna be successful because when it finds a break, it just constantly goes back and goes, okay, let's test, why did it break? There's much less giving up or much less, can I take the next step, sir? You know, type thing happening. It's been, and it's going, no, I've made a mistake here. And like, oftentimes, it'll go, hey, It'll give me an answer and I'll go, no, that's wrong. And I'll go, this is a better answer.
00:32:28 Joris de Gruyter
And you can see the monologue.
00:32:31 Mark Smith
Yeah. And I'm like, this is so cool. And once again, I feel like it really does allow you to become almost like superhuman in a weird way in that you are, you know, you talked about burnout and stuff and that's why it's really important to, and I know you've been in the space for, you know, two or three years now and monitoring your mental health and because it could become addictive like anything else could become addictive. It's just, it's creative addiction in a way, but addiction's addiction, right? And you don't want it to be so all-consuming that it affects other parts of your life.
00:33:09 Joris de Gruyter
Yes, and I think we'll have to figure that out. I mean, like I said, I've picked up projects that I'm a huge fan of OneNote, which by the way, I'm trying to figure out like an MD alternative to Do you know what my favorite one is? No, go ahead, please.
00:33:25 Mark Smith
My favorite markdown free is notebook plus plus.
00:33:29 Joris de Gruyter
Notebook plus, okay, I'll have to look into that.
00:33:31 Mark Smith
Oh man, like I've had it as my default notepad for years, but the beauty is it color codes like, you know, so you can see your entire MD files and it's just a note version that says, hey, this is an MD file extension and therefore
00:33:48 Mark Smith
What it doesn't do is you've got to decide where they get stored. They're one individual flat file, but it just lights up everything beautifully. So I've made it the default. Any MD file gets opened by that app. And it's also, you know, you can read Python. You can do anything in it from a coding perspective. It can read the content and goes, this is this language it's been used.
00:34:09 Joris de Gruyter
My main issue, because I've looked at Obsidian, a lot of people love Obsidian. My biggest issue is I need it to work on my phone, which is an Apple phone, and on my.
00:34:23 Mark Smith
PC.
00:34:23 Joris de Gruyter
And the syncing between the two is really, I struggle with, for example, I don't like vendor lock-in, which is funny considering where I work. But I don't like vendor lock-in, right? And so I've had an iPhone for many, many years. All my photos get synced to OneDrive. And there's a good reason for that. One, I have a personal like office subscription.
00:34:47 Mark Smith
Yep.
00:34:48 Joris de Gruyter
I get a TB storage. Like why would I pay Apple for... And then secondly, I mainly work on a PC, although I have a Mac and Linux, I have everything here. But my PC is my main device. And I need my photos on there. And so yeah, there's iCloud sync, but everybody tells me don't use that thing because it's just a mess on PC anyways. So yeah, so I don't like vendor locket. And that's where the problem becomes, right? And so there's some really great note-taking apps that Apple has by default on the phone. There's like free form. I like free form. But there's no way to get that stuff on the PC. There's no sync. There's no, you can't get the data only on the Mac, right? And so that bothers me. That's really my issue. That's why OneNote, to be fair. I mean, that product's showing its age, but it still works.
00:35:36 Mark Smith
So I use the thing called, as in, so I'm on iOS as well, and I use an app called Drafts, D-R-A-F-T-S. And I've used it for years.And it's on my speed bar on the bottom of my phone, right? The only four icons you can have, because I use it that much. And it just allows me to take text, only text anytime, grab it. but it's got a massive rich API exit layers. And what I mean by that is like, you can do the dump and then go send it to wherever you want. And although yes, it works well on a Mac, I don't have a Mac, but it has an integration to that. I have now been able to write the JSON that flicks it into my Azure environment.
00:36:17 Joris de Gruyter
Yeah, but see, that's the thing. I feel like there's options, but all of them require you to set up sort of your own duct taped sync mechanism. And again, we have Claude or GitHub Copilot that vibe code that for us, but that's not what I want. I just want it to work. And there's no reason why I could use Google Drive or OneDrive to sync my MD files between my phone, my iOS phone and my PC, even though it's Google and Apple and Windows, it should, and that really bothers me.
00:36:45 Mark Smith
Yeah, you're giving me an idea now. I'm actually going to make it sync it to my, because you know, obviously I've got OneDrive operating on my phone, on my iPhone, I reckon I can fire that into there and then use OneDrive as the connection point to it all.
00:37:00 Joris de Gruyter
I want it all in MD now, because then I can have my AI take care of it. Because what I have on my phone, like there's these links, right? During the day, you get sent some link, like, oh, it's a great YouTube or whatever, research paper, I want to read that later. But then now it's on my work stuff. And I just want to read it on my iPad when I'm in the couch, which is not managed by Microsoft, so I can't even get to my teams or whatever. Or, that sort of thing. And sure, you can get a Firefox account or whatever, or an Edge or a Chrome account, and then you can sync between browsers. But I don't want that, right? Because I don't want like a browser sync with Chrome and a to-do list with the to-do app. No, I just want to be able on my phone to say, share this to Claude or whatever, and then it just shows up. somewhere where I needed to show up and sort of, that's on my list of things to sort of figure out as well.
00:37:53 Mark Smith
I like it. Well, over time, it's good talking to you and I feel like we could talk for hours.
00:37:57 Joris de Gruyter
Yeah.
00:37:58 Mark Smith
So we'll wrap this one up and then we'll have our private conversation. That's right.
00:38:03 Joris de Gruyter
AI Brain Fry. Tell people about it.
00:38:04 Mark Smith
AI Brain Fry.
00:38:05 Joris de Gruyter
We'll find the link and you can put it in your notes.
00:38:07 Mark Smith
We'll get into the show notes. Thank you, sir. It's been great talking.
00:38:10 Joris de Gruyter
Always.
00:38:12 Mark Smith
Hey, thanks for listening. I'm your host, Mark Smith, otherwise known as the NZ365 Guy. Is there a guest you would like to see on the show from Microsoft? Please message me on LinkedIn and I'll see what I can do. Final question for you. How will you create with Copilot today? Ka kite.




