How Michael Batko gets AI to do the busywork

The outsized returns you can get with AI are usually not actually on the cost cutting side of things, it's actually on the revenue building side of things.

You and I are in this massive bubble and what I mean by bubble is like we're in startups, we're in VC, we're like talking to tech people, everybody's using AI around us and it just like feels like no brainer. Everybody else should be using it. How will you get my dinosaurs on board with this change?

Thanks so much for having me.

Yeah, let's start. I left Startmate and now I'm building Hourglass AI. Hourglass AI, we go into companies, we ask about all the problems and we build the AI solutions for them really practically. And it's literally around the outcomes that people want. Like we don't shy away from custom work, we just want the whole team to be using AI and just solve their problems. So there's like no products that we're necessarily selling, which is probably a little bit counterintuitive compared to what I was always preaching at Startmate around scalability and VC. But honestly that's what I enjoy the most. Essentially we go into businesses, we solve the AI or like the IT challenges. So often it's also just around building deterministic software as well. It's not always fully AI, kind of like Applied Wizardry, but the types of companies that we work with are really fun because we now work with venture capital firms, financial services, we work with an accounting company, a forestry and meat processing business. So like SMB versus kind of like VC and startups and financial services, it's very different, especially SMB.

After eight years at Startmate, it was just time for me to do something different. And the only reason I say that is because I was like, I'm 35, the world's a big place, there's so many opportunities I need to do and try to, to do something myself, otherwise I'll always regret it. So that was actually completely separate to me than starting it because when I left Startmate, I had no idea what I'll do. I was like, I've got the best job ever. I'm like investing in founders, helping them out, raising money and just like seeing people progress. I loved it. So it's literally like, it was very scary because like, will I ever find a job this fun again? But then I left and took a couple months out and I literally gave myself permission for three to six months to just have fun, experiment and, and then Hourglass, the reason for it or the reason I started it is because most people know me for my productivity and I like, literally like help people set up their inboxes and to do lists. And how do you actually, like get more done and have more time with your family and loved ones?

I mean, this is because most people knew me for. Because I exactly had the side hustle, which I didn't sold. I trained like 500 people, had to become more productive and so on. But now with Hourglass AI, I literally get to do that on steroids because now I get to like make whole companies super productive, which is like, I love that

Oh my God, it's crazy. From a high level, what it changes is essentially two things, like, number one, speed and number two, customization. So like the speed of building has never been faster, Right? We all know it. Like you can actually ship something and especially if you're non technical, you can actually get something out there which you previously just never even could dream of. And the other one, though, which people underestimate is actually the customization. Previously you always had to run a waterfall to build one custom product for like those 15 enterprise clients and you could never like swerve away from that. Whereas now nothing stops you to like build the most crazy custom stuff for every single person. And obviously this is what Hourglass does. I even had conversations with SaaS founders now who are like, previously our advice was like, never ever that can do custom work for any of your clients. But now they're just like, why not? It's actually quite fast and as long as you keep the core kind of consistent, it can potentially work. So some of them actually started experimenting with that as well. Now

Totally. Oh my God, such a great question. So true. Number one, you and I are in this massive bubble. And what I mean by bubble is like we are in startups, we're in VC, we're like talking to tech people. Everybody's using AI around us and it just like feels like no brainer. Everybody else should be using it. I now pitch SMBs and I can tell you not everybody's ready to do it. And even the SMBs who are actually approaching us, I'm like, you are literally ahead of the wave because you are the top 1% who even is thinking about adopting it. So I'm getting super disillusioned with 99% of the businesses, haven't even started exploring it. And then to your question of what's actually blocking it, it's really two things. Number one, and it's the hardest thing is behavior change and two, systematic thinking. So behavior change hardest. When I do an AI transformation for a company, there's the tech side and there's behavior side. The tech side actually is quite easy because you can launch things much faster, much more custom, specifically to an outcome now. Whereas the behavior change, so much harder. Every single client asks us. Actually I just got off a client call just then and the first question they asked is like, how do you get my team on board to actually using AI? The team of 35 people, so not even that large, but they already concerned that the team might not use it because they're just scared of change. Like it is a big, big shift in the way you work. Like first thing I do when I open my laptop, I've got eight Claude Code windows that open at any given time and that's what I do. Like manage my inbox, manage my calendar, whatever it. And I literally don't even use the interfaces anymore that we previously used. But it is scary to people who haven't done it before. They actually, the clients who I just got off call with, they called it, how will you get my dinosaurs on board with this change? And that's the behavior change. It's so hard.

Oh, 100%. Okay, two answers to this question. Number one is doing the discovery well. And number two is human in the loop, which I can tell you about. Number one is actually just doing the discovery well. Like it's so easy to build stuff. So probably you and I have built like 100 agents and 24 hours later, some of them don't work anymore. And then you discard all of them and then you just revert back to like the best three you've actually built. And that is because the discovery hasn't been done well. Where like, what is actually truly important to you and what will save you the most time and or make you the most money. That is what we now do with businesses really well where we go in and we actually do as step number one, we actually audit them. We go in, interview your team and you actually tell us like, what do you wish existed? But also like, what do you spend most of your time on? The best way to discover what people truly want is not to ask them of like, how often are you going to go in the gym next next week? Because everybody's going to be like, oh yeah, five times a week. Especially to ask them, well, how often did you go to the gym last week? And suddenly the answer is very different. So someone's like, not what are you going to do next week with AI? But it's actually like, how much time have you spent on X and Y and Z in the last week

Is that

Absolutely, yeah. Yeah. So it's like you want the actual practical part, not just the wishful thinking. Yeah,

I mean, totally. Like previously there was a job of product managers, right, to actually understand your customer well. But previously you couldn't even uncover all those problems that well because it was a waterfall process anyway. You kind of had to go at the core. Then you did a bit more customer discovery, then you went deeper and deeper. Whereas now with AI, you actually need all of that context up front to do that really well. Because now you can almost one shot builds.

Yes, exactly.

Yes, I love that. And you can make it really practical. Like this is kind of our ICP now, which is essentially companies of 20 to 100 FTE with no engineering team. Why? Because they literally can't build it themselves previously, whereas now we can actually go in and build it specifically for the company. And the other ICP, which is really interesting. Exactly to the point you just mentioned, there is actually SMBs. So let's actually take the meat processing business as an example, which is amazing because there is no meat processing software. Nobody's ever done that mass build that you mentioned for me, the meat processing industry. Why? Because it's probably like really hard to do marketing and sales for those guys. Whereas now with AI, I'm like, what outcome do you want? Let me build literally this n of 1 software for you specifically to solve your problem.

I mean this is kind of like the beauty of the business that we're building where we always tell our customers we're going to be here for the long term with you. Like we essentially care about the outcome. Not just now, in the next three months, but we're actually going to be a partner for you for the next few years. Because when we build it, and we exactly acknowledge what you just said there, like software requirements will keep changing, you're probably going to get more people on board, you're probably going to want to recut it, you're probably going to want to have changes and so on. So the way we actually do it is and we build that MVP, get it, live with the client and then actually stay with them with like a maintenance fee to keep making small changes and iterations along the way. Because we recognize exactly that of like the software will keep changing. And with AI you can actually keep changing it to fit exactly your need at the right time as well.

That's an awesome point because I probably actually see it slightly different. I agree with the premise of like, the critical business software is technically like the biggest lift you could do to actually drive the most value for business. But from an AI perspective back to what does the client actually care about? It's usually just two things. One is I want to save time or save expenses. The other one is make more money. And then the critical software part is actually that's just an assumption of, for example, vibe coding your CRM. I actually personally never go there because what I actually care about is solving those outcomes. So what I care about is where do you spend most time? To be honest, the answer usually, especially with financial services VC, is in my inbox. Sure it's critical infrastructure, but like the AI solution here is for me to pre draft all your emails based on perfect context from the entire company on the company AI brain. And that kind of thing like, is actually very simple to do and or doesn't necessarily incur that much critical infrastructure tech debt. The other side is again like making more money, which is again like a way more simple kind of automation of finding your custom segments, emailing them. But a point is almost like with the stuff that we're building, we always try to keep it as simple as possible. Like every single one of our builds only does one thing and does that really well rather than just bloating IT into this massive system that controls everything

Yeah, honestly my pitch to go back to clients is almost like the critical infrastructure is actually your unique IP. That's the part which you actually should be doing incredibly well and you almost don't want somebody external to touch it and to do it for you. For example, if you took a company which is prides themselves in branding and marketing, like I should not be doing your branding and marketing with AI, otherwise like you're just going to AI slop it

Our engagements usually start with context and collaboration. Context is a company AI brain and collaboration is essentially sharing skills and workflows with each other. And those two things already unlock all of the businesses that we work with. Context itself. Company AI brain is essentially sucking in all of your information from inbox, CRM and wherever else. And from a technical perspective, honestly, it's actually very simple. It is a perfectly structured MD file tree that sits on the company servers as well from a cybersecurity perspective. And they control it, they can take it wherever they want to. They can plug any AI into it as well. So

Yeah, so like Obsidian kind of like from like a front end. Yeah, so that's kind of like the visualization.

I mean I'll give you kind of like the high level here because I'm definitely a non technical part of our team. But the first one is the ingestion, which is the big bulk because for example we've had to download like 14 years worth of data of companies and like hundreds of thousands of files. And that's actually the tricky challenge for like building ingestors essentially across all the different software that companies have. And then there we actually do pretty custom stuff. For example, some of the limitations were like massive PDF files or even just like Notion attachments and stuff. And then from like a keeping ingesting perspective. What we do then is overnight, every single day on a 24 hourly kind of basis. We just suck in all the context from companies over and over again. And the way the MD file is structured and works is it specifies with each one of them who actually should have access to it in the company as well. Because you want to ring fence information across people in a company as well. And then you kind of keep restructuring that MD file tree as well based on the kind of usage. For example, if a company has multiple rebrands, you obviously never want to use like an old brand. You just always put like the newest brand on top and stuff. So there's like a whole kind of like continuous restructuring of the MD file tree as well. Which is why that maintenance is important because we stay with our clients over the long term.

Yeah, that's right.

So usually the second step is once you've got the brain. Honestly, companies actually love it already because suddenly the whole team, when they actually chat to Claude or ChatGPT, they suddenly have like such awesome context. Because now every presentation you do is based on company colors. You've got your tone of voice, you even have your personal tone of voice as well if you want to draft your emails, etc. So like the context itself is already a massive unlock for lots of our clients. The second one is then collaboration. As you can think about companies, that's called 40 to 50 FTE. Like everybody's getting excited about AI. Everybody builds themselves the same tools. Venture capital industry, we've got three, four VC funds, we work with like investment team. Every single one of them builds themselves a due diligence AI. The problem is they're all inconsistent, they all do different things. So now with collaboration, really the simplest thing that we can do is is have one person build the skill, put it in, and then everybody in the company has instant access to that skill and you actually just maintain one skill at the time.

So that would be like the second kind of build that we usually do for companies. Yeah, so then you for example, even have an admin who can approve whether a skill gets updated or not and kind of maintain the skill repository across the company.

Yeah, I love that. That's actually a great analogy. And honestly, even with those two things, people already are flying because like, you've got perfect context. You can actually share things across the team. Things don't keep breaking down. You can actually have consistency across the team as well. What comes next completely depends on each company. For example, we're working with a forestry company. They really wanted us to increase the revenue. So essentially their idea was we want to scrape all those government websites and the DA approvals and then whenever somebody's got a DA approval, put our super premium wood in front of essentially the client. So that's a great way to just do outreach at the right time at the right place.

Honestly, 95%. It's exactly what you said. The general kind of harnesses. We actually love working with Claude itself and you can pretty much do most of the things there. One specific tool I could actually maybe mention is one of our clients has lots of sponsors and he himself does lots of filming. He needs to report all of those numbers in terms of social media back to the sponsors every single month. Takes him hours and hours every single month for each client. So instead we've built him that report of like just literally press a button and sucks in all of your social media data across all of his clients and puts this perfect report together. And that I believe we've actually built it for an app called Apify, which then actually makes our lives a lot easier to not have to connect with Instagram, connect with X and Y and Z, but actually just scrapes it all for us. So then it is sometimes actually worth paying whatever it is, 30 bucks per month to not have to custom build every single thing.

I mean it's all the standard ones where like to be fair, if you actually do SEO well these days you actually even get ranked on AI to get recommended. The classic one here is Supabase for the database itself. I think everybody is using it now because it's like the first thing that Claude recommends. Second one is actually for example, for lots of emails to clients and sending out emails en masse for like your direct consumer product, for example. There's actually an Aussie company called Nitrosend now as well, which you can actually use. But those types of things, like, sure, you can vibe code it, or you can kind of like build yourself the connectors, etc. But like, is it worth it? If it's 30 bucks per month, you're probably better just buying it.

You know what, I'll make it super practical. I'll tell you about my favorite workflow which I think every single founder should actually have. The one which I use religiously every single day and I think I use it literally like six or seven times a day is just the simplest one possible which is post-meeting follow up emails. It's like number one, you have a meeting, you 100% have to record it. Like I love using Granola. Granola just sits there, records the whole meeting. As soon as the Granola finishes, a Claude Code window opens up, summarizes the meeting itself. It goes more through my Batgo AI brain, finds my tone of language, finds all of the previous kind of connections that I've had with that particular person and drafts that perfect kind of reply back to the person. The reason why Claude Code is great because I can then whisper flow back into it if I want to make any changes, otherwise you shoot it off. Second part of the workflow is actually even more magical though, because it actually finds who introduced me to that person, if anyone introduced me, and also drafts an email saying thank you, this is what we chatted about. I really appreciated the introduction. So that's always beautiful because like nobody ever loops back with you when you do an introductions and I do a lot of introductions and people do them for me too. And the final part of the workflow is actually one where it then takes the whole conversation and even suggests a blog post I should write about the topic because it could be interesting in something I haven't written about before and even then itself is actually a great idea database of content.

Love that.

I'll actually add to that one because that is a great use case as well. Is my wife and I collaborating on our budget and is now on a personal kind of basis. But instead of asking a question of like how do interest rates affect us, now we need to remortgage, refinance, etc. Like throwing all that information into a Claude Project including your whole budget and even just like monthly expenses and stuff. And suddenly Claude does have perfect context. And you can literally ask any question of like, hey, what would happen if this kind of like interest rate up and down and stuff like that. Making it super practical like that for financial modelling is awesome.

Yeah, I love that. Honestly, like even the prompt that I used there was an interesting one of throwing all of my tax statements into it as well with essentially that financial planner and asking the question of like what could I have done better in the last few years and what should I do better in the future? And for example, just learning around like unused super caps in the last three to five years and it's actually a thing and you can actually top up, it's like I never would have known that. And like you said like getting accountants times and stuff is also like painful. Plus they might not even tell you because there's so many other priorities.

It's literally the depletion of willpower because you're making so many decisions, you have so many concurrent processes open with Claude, and if you clog that kind of brain space, then you just feel so dead by the end of the day. And to be honest, my biggest tip there is actually one which is completely non AI related. So now actually going back to my productivity teachings at Patton Ford, but it's, it's the classic now because you can have 10 different processes running at any given time and then still feel unproductive at the end of the day, which is so frustrating because at the end of the day it's like, what did I actually do and what did I actually achieve? So my biggest tip here is really just the one of a weekly and daily practice. The weekly practice is a very simple one. Monday morning, top three priorities for the week and block that time out in your calendar to get those things done. I personally like sending those as an email to my team. So my team holds me accountable and they know what I'm working on. And the other practice is a daily. What's the one thing that's going to move the needle on my goals today? And it's just one thing, as soon as you get that done, you can actually call a day as well and just be comfortably going into the evening being like, I don't need to work more because I've actually done the big thing already. And just that in itself is just a massive unloader of brain space.

The only thing I'd add to the workflow is we've got it hooked up to our goals as well. So that the AI assistant and also just like holding you accountable, like, is this actually driving your goals?

Especially because if you use, for example, just Claude, it's got perfect context on your actual calendar, but also perfect context on what you were asking Claude about. What work were you actually doing? It's a fascinating summary. At the end of the week,

Yeah, more than happy to talk about it and go deep as well. It's quite unique maybe, and I don't know if I would recommend it to everybody, but my Claude is actually set up hooked up to my terminal. I literally open up my terminal and my flaw is that I'm non technical. So I was also like, oh my God, like how do I now learn about, like which project to open up, where does it actually live, which folder? And I didn't want to like have the hassle of working it out. So I literally asked Claude like, hey, can you help me out here? And essentially the shortcut that I've got is I press C on my terminal, hit Enter and it opens up every single file project I have and the last few conversations in that project that I've got going as well, and that in itself, I just hit a number or a letter and it just opens up exactly the conversation and it's like continues the conversation as well. And it just shuts those conversations down within 48 hours of not being used. So that actually is an awesome shortcut because I can continue conversations but also have like essentially go straight into a folder where I want to use the Hourglass AI Brain or Batgo AI Brain or work on a specific client and so on. So like that already helps me so much. So that's kind of like my setup in terms of terminal and Claude. I didn't know that. The Claude app actually has gotten a lot better though, so I probably should transition back into that as well at some stage.

That's a Big Mac. That's what we found as well. And so many of our clients actually ask us, oh, what AI should I sign up for? And so on my answer to clients there is usually just a one of like go whatever, it makes it easy for your company. And all those models are actually of similar power anyway. And for you as like somebody getting started, it actually doesn't matter whether it's an OpenAI model, whether it's an Anthropic model and stuff, because they're actually all equally good for like basic use cases.

I mean this is maybe the other one which is why you're not churning into other models, right? Because you just never cap out of things because you probably pay enough money to just be able to get your whole week done. As soon as you cap out, you probably pretty quickly going to move on to another model and or pay up to small. But no, I 100% agree because like for me as well, like I just kind of fall in love with one model and I don't need to keep experimenting with the latest Astra models, et cetera. Like I honestly Opus 4.5 does the job so well is token efficient. Honestly, even if AI model advancements stopped at Opus 4.5, I think I'm pretty happy. Whereas some of the previous models, they were actually very token heavy and the quality wasn't quite there. But I do feel like we've reached the stage now where you can stay with one thing rather than keep shopping around. Even though I agree with you, people are just so quickly to switch between things. And the lock in really is the interface itself, because you're just comfortable with something. And this kind of comes back to our previous conversation on behavior change. Even you and I now, right, we're just like, we like our setup. It's painful. Even if it takes me 30 or 60 minutes to go across somewhere else. Like, oh, man, it's another like hour just like faffing around.

Then what wish would work is a classic one where, like, if you explain it to somebody once, they just like, get it set up and get it working and just like a happily ever after. But it just, like, never is the case, especially with people who are reluctant. What actually works in organizations, like, to be honest, like 10 to 20 FTE companies, as long as the leadership is on board, and that's maybe the condition the leadership team needs to be on board, that usually works because if the leadership team exemplifies something, then the rest of the team will as well. And as soon as the companies grow roughly above the size of 20 plus, you will always have a couple people who are reluctant. Even if it's just 1 or 2% of your staff. It literally means one or two people probably really hate it at that stage. What we've seen working really well is AI championship in companies. Find the people who actually are genuinely, really bullish about it. Don't start with the whole company, but actually start with them to really get them unlocked. And then they actually just act as incredible inspiration for everybody else. Because once you see them working and what they can do, suddenly everybody else in the organization asks themselves, wait a second, I could do that too. And that's where the behavior change happens. Like one of our clients literally called them Claude Lords and they just even gave it a fun title. And they meet once a week. And now other people want to be part of the Claude Lords. And that is actually kind of like really interesting behavior change where you inspire

100%. Yeah, there's two schools of thought there, right? There's the romantic school of thought, which to be fair, I'm a subscriber to, which is like with AI, your team is now going to

With AI, you can now get your team to do the higher value work and actually grow your business and do higher leverage things. So like you almost like essentially 10x your team without having to cut the team. There's of course the other side of things, which is the more and more ruthless side of the world, which is maybe like a more PE type thinking of if I get AI into the business, I can now cut my workforce by 10, 20, 30%, whatever it is, which is very much like a cost efficiency type of question. And to be honest, like I like to think about it as the outsized returns you can get with AI are usually not actually on the cost cutting side of things, it's actually on the revenue building side of things, right? So it's like if you can actually get your team to unlock, to actually get more clients, have better personal relationships with customers and clients and actually win more business, you're going to get so much further than just being able to cut 10 or 20% of costs.

The one example which comes to mind is for one of our clients, what we've done is we've automated all of their inboxes with this perfect company AI brain, where suddenly they all have pre drafted emails based on company context and they can just shoot off those emails or edit them. And the AI kind of keeps learning what you've done. And the specific use case here is actually executive assistants. They manage massive inboxes of C level companies of 100 FTE. They get inundated with emails, but suddenly with AI, that work of five hours a day is now cut back all the way to like 30 minutes of triaging and shoot off, shoot off, shoot off. One of them actually went on holidays for two weeks and she left it on and just auto shot all those emails for her for two weeks and it was fine. So that's where the question kind of comes exactly, like, is the job still necessary? And then it kind of comes back to exactly like you said, the person themselves of like, do they have the skill sets to go anywhere else? And the company as well? Do you want to actually retain somebody like that as well? And how good is the person? What we've seen in clients is actually always been the case of, hey, I've hired somebody who's amazing, even though that part of the job is gone awesome. Now we can actually repurpose them onto something else, which is higher value, which is usually the kind of interpersonal, relationship building kind of space.

Oh man, so many things have surprised me. But what has surprised me the most, what has actually surprised me the most, and this is pretty, pretty basic, so I can maybe give you another one, is just like the speed of change with AI. Like it is crazy how good it has become since January, February this year in only the last seven, eight months. If you took it literally like two years ago, like AI was nowhere there yet. So maybe the surprise has been how people also get stuck with old mindsets. Like two years ago, the conversation with AI was like all around hallucination. You asked AI a question and the hallucinator gave you the wrong answer. And then people churn. And then two years later you still hear them talk about one time that AI hallucinated. Whereas now two years later, like, honestly, with great context and a good prompt, like, AI is actually very, very spot on. Like, you don't really even hear hallucinations anymore, but yet people get that stuck in their heads and don't want to change. So that's actually what surprises me the most, is like you have one bad experience and then you just don't revisit that mindset anymore.

And this may be the other part, which is really surprising, which is how few businesses actually have started exploring AI and going deep into it. Because I do still think most people's experience of AI, and that's 99% of Australians at least, is literally just a chat box of asking a question, getting an answer. And that's the experience extent of the AI use. Whereas the, the magic moment really comes with actually being able to give it a prompt and it builds you something real. And usually the easiest way is just to build yourself a website and suddenly you just like have a completely different mindset.

Yeah, amazing. To be fair, like, that's an interesting one because, like, I actually wonder how quickly that change is going to come around with giving agents access to your finances and like to your wallets. Because I feel like that's like a whole other leap of like, faith in AI, which my prediction is actually it will take way longer than we think.

Have you given yours actually access to your wallet to pay for things?

Yeah, interesting. Okay. I mean, I like that the argument you just made, but I think the fact is like, you and I still haven't given it our credit cards and I feel like we're already so far ahead of the curve.

No, no, but that's exactly what I mean. Like, I've also given it to pretty much all my financial information, yet I'm stopping at the point of actually being able to spend my money.

Yeah, I reckon that's actually going to be the next kind of thing which is going to arise. For example, plug it into whatever is like a Stripe account, but just specifically for your agent.

I think so too. Yeah. I do think it's going to take a little bit longer than we think, but I do agree with that.

Let's do it. Yeah. I feel like the other interesting conversation is actually personal agents as well, because I would actually say that I'm a bear on personal agents because I think this is similar to what you just mentioned there. Like lots of people say they want one but they actually just will not use it unless it's crazy proactive.

Company itself. If you do want us to come deep with AI and build stuff for you, but also upgrade your team, go onto the Hourglass AI and otherwise. Best to follow me on LinkedIn. I'm very responsive as well, so feel free to DM me anytime.

Thanks so much.