So, hi and welcome to Old School New Tech. Today's another solo episode. After last week, I was building a software utility live in kind of a code with me, work with me type of stream. It ended up being very useful, to be honest. We're using it heavily at Automaze. And yeah, it's it's been a success, I would say. It's called Invapor, and it's a small utility that allows you to not worry about committing your secrets and environment files and not only not worry about it, but actually proactively doing that so that it can be used instead of.example files and other template files and so on and so forth. And today I want to talk about something that I keep being asked by different people, and it's how I work. So today I'm gonna kind of share a little bit about my workflows, my setup, all that type of stuff. So I should probably start by saying that six months ago-ish, maybe a little bit more, I shipped a book. The book is called Production Grade Agentic AI, and it's a book for engineers who want to deploy systems to production environments. And it's a 660-page book, it's a big one. I have another book coming up in a few weeks, and uh I also run Automase, which is a CTO as a service software agency. We have almost 30 people working there, and I've got an advisory kind of climbing out of stealth. It's called VarOps, and more about that and the new book in a few weeks. And people keep asking how I get through it all, and I guess that's a fair question. So, what I'm gonna do today is I'm gonna try to share as much as I can about my setup, the system, and kind of the decision I keep making 40 times a day, and that everything else kind of hangs on. So I'll start with the technical setup because I guess that's a lot of you are mostly interested in that. So there's a huge debate and whether or not to run your coding environment locally or on a cloud environment, and I can see the merit for both. And I'm I think that I probably figure out kind of a way in between, which is my local environment, is the cloud. I have a dedicated machine that's accessible to me and me alone from anywhere. And it's for me, it's what both trying to be, right? So I like to work on an environment that I fully configure it, that I can run different experiments on that when I do coding in a more exploratory fashion. So not I'm not in maintenance mode or I'm not in a fully mature product where it's just hey, let's add this feature or fix this bug. But well, I'm when I'm still trying to figure things out, I like to work locally, but I still want the ability to work remotely, if that makes sense. So what I've done is I have another mech set up on in my home office, and it has my harness of choice, which is Droid by Factory AI, and it runs all the models. Obviously, I have Claud Desktop installed there, and I also have Claud Desktop installed locally on my laptop, but more on that in a second. So I'm using Droid as my go-to harness. I'm using a modified version of which is a really cool tool that it's what I think most IDEs will end up being. It's a sort of a hybrid between let's say it's it's your own local cursor style environment for ad hoc coding, and it supports a billion different coding harnesses and also Droid, and it allows me to also code for my mobile machine. So I can be on my mobile device and I can do my coding from there, especially for long-running sessions where there's not a lot of interactivity. So that's what I use for the, as I mentioned, the exploratory part of the coding work. For when the product is more either mature or is kind of cooked enough. So I'm more in the let's add a new feature, let's fix something, that type of stage. I'm using Cloop, which is an internal Devon style environment that we've developed internally and will be shipping out for everyone else sometimes in the next few months. I use for kind of when I want to automatically ship new features, brainstorm new features. We built into that brainstorming module, which is great. We have a feature called Ladybug, which automatically hunts down production bugs. It's part of the code, sort of like Sentry, and it automatically goes through all the pipeline of discovery, fixing, and a PR. And uh yeah, so I the human is always in the loop on the PR level. So that that's kind of my coding setup uh for my I also use VS Code, but these days it's mostly used as my word processor, let's kind of note taker, if you will. I found that it's very helpful for me at least to use it, especially when I was writing my book and the one that's coming up soon. It's a very convenient environment for me, at least as a coder, uh, to both work on ASCII doc files and do the actual coding necessary for the book. But I hardly ever use it as a proper code editor these days. For my own assistant, I know that many are using a combination of or either CloudDot or uh OpenClump, sorry, or Hermes. I'm actually using Maxi, obviously. It's a system that I've developed and I know all the ins and outs of it, and I can add new features as I need them. I'm using Maxi, but I'm using Cloud Desktop actually as the client for Maxi. So it's connected via MCP and Cloud Desktop is the interface I use to communicate with Maxi. The cool thing about being able to use Maxi from Cloud Desktop is that I do use Cloud a lot. It is my go-to brainstorming partner and helps me kind of flesh out ideas for other stuff. I like to keep it simple. I use reminders, Apple reminders as my to-do list. I use Apple Notes for Quick Notes. I use Superhuman for email, just switch over back to them. I use them like a few years ago and I switched back about three months ago and it's been going great. My servers, my hosting partners, I'm using Cloudflare, Hetzner, and DigitalOcean as kind of my main stack. And obviously for client work, we use whatever fits the client, but for my own projects, that's what I like to use. Everything that I need to know, what's going on in the company? I have AI automations for that. I'm getting the morning briefings, everything that's happening and waiting on me in Slack, my emails, kind of preps for the meetings. It's doing the triage for my emails, of writing the drafts. So everything is runs on Moxie, and I get the briefing with some action items. For any sort of client communication, I'm using I'm using Slack Connect. So that's what we're using for internal communication within the company, the team at Automase Loves Discord. So that's where we hang. But for any sort of client communication, we are on Slack. I also self-host almost everything that I can self-host myself. Not necessarily to kind of, hey, let's save some money. I have I'm spending tens of thousands of dollars in tokens. So obviously, it's not a matter of trying to save $20, $30 a month on various. It's more about data ownership and being able to hack around the processes. So even if I don't hack around the code, I want to be able to kind of access the database directly if I need to and get access to the role, to the raw data. So it's more about data ownership and data governance rather than saving a few bucks. So yeah, I'm I'm more than happy to pay for any software that I'm using. Just let me self-host it. So so that's that's kind of my my setup. I have multiple agents running in any given moment. So I have the mega coding agent that's running on Cloop, that's kind of doing ongoing work and maintenance and whatever needs happening. And uh locally I'm mostly have a screen with Claude Desktop and Paseo for kind of coding. They're all funnel into the same harness. So they're all using Droid at the end of the day for the actual coding. So my sessions are consistent across everything. Whether I started a coding session from MUSI or from Paseo directly from Droid or from Clue, it's all available for me in the same environment and they're sharing the same sessions. So that's a setup that I've built for myself in terms of coding. In terms of uh, let's say, segment number two, how I manage my time. So my job used to be doing the work, and now it's more about deciding which bucket the work goes in. And I remember a great thing that Jeff Bezos from Amazon said that he gets paid to make a small number of high-quality decisions. So, not that I compare myself to Jeff Bezos and the operation that he gets to manage, but that that's kind of the inspiration for what I'm trying to achieve. I'm trying to make a small number of high-quality decisions and spend the rest of my time doing things that I actually enjoy doing, which for me is building stuff, kind of experimenting. I build a lot of stuff that would never see the light of day just because I find it interesting. I make 20, 30 decisions every day, and almost all of them are classifications. We have two tracks. I have the Rails, and I have Asian, I have bespoke general, I have a decision of whether something goes to a human, whether something becomes automated somewhere in the middle, or whether it's a process that it's time to kill. So the disciplines that I'm trying to follow is that first off, anything that I've done more than a few times and I have a process for it gets automated. Automated doesn't necessarily mean AI, doesn't necessarily mean kills or agents or anything like that. If it's fully deterministic and I know the inputs, I know the outputs, there's no AI uh involved. It's just simple automation. If it's more fuzzy and I need judgment, that I'm going to deploy an agent to sort that out. And a lot of the stuff that is automated is somewhere in between. So most of the steps are follow workflow. It's a very deterministic flow, but there's a fuzzy part like write the actual email. That's going to be a fuzzy part, or you know, make a decision here. So that's going to be AI. But everything before or after that step doesn't necessarily need AI. And yeah, I find that this combination works best for me. So if something is done more than a few times, it gets automated to one of these three buckets fully AI, hybrid, or just a simple automation workflow. So most of my time basically goes into thinking, not typing. So yeah, I go on calls, I talk to clients, I sync with my team, but it's not as often as you might imagine. Most of my time do go into thinking. I spend a lot of time thinking about and actually minimizing the code base, which will result in a smaller maintenance surface. So this is something that I spend a lot of my thinking on. I guess you can call it kind of first principles. What is the product? So even products that are have been live for a while and are being used by lots of people, I still like to ask everyone now and then, kind of, what is it? And let's strip it down to the bare kind of core minimum. So if I would go and ask what is a coding agent, yeah, all those Kanban and workflows, these are secondary to having a system that can actually code, right? So I'm I'm trying to strip down to the essentials and then add uh features back up that both reduces the complexity of the code banks, makes the code a lot easier to maintain for ourselves and agents. And uh yeah, so basically I I decide a lot of what not to build, a lot of what not to do. I will spend hours, sometimes even days, rarely weeks, but that that we known to happen on brainstorming and specing. I use, as I mentioned, I use Claude AI for that before even one line gets written. So just use it for doing research. I find that by the way, Claude with especially when you use it as in a consortium with GPT and perplexity is great for that. But for the actual hypothesizing and thinking process, I like to use Claude AI, not necessarily because it's smarter than GPT, GPT, especially the soul, is amazing. I like its personality better, if if you can say that about AI. It's I just I find it easier to talk to, but I do bring up GPT every now and then during those brainstorming sessions in order to ask, hey, what do you think about that? And what what feedback do you have for Claude? I feed it back to Claude, Claude's feed it back to GPT, and I do like the fact that I'm the human in the middle because it's very fuzzy. This is not something that I would like to automate, that I enjoy thinking about these types of things and kind of solving the puzzle. So I don't see myself automating that part anytime soon. I also think that's my judgment is what makes our products work. And yeah, that's something that comes with decades of experience, I guess. I mentioned perplexity, perplexity is a great research body. It's just I feel that it's a lot more up-to-date and it goes out to different sources. So yeah, that's that's what I use it for mainly. I read the code for everything that is being produced by agents, I read and I comment a lot about the code. The generation, code generation is cheap, judgment still is expensive, and I think that at the end of the day, what I bring to the table is not the fact that I can type Python or TypeScript or whatever better or faster than someone else. I think that what I bring to the table is my judgment and my ability to foresee problems in the code base and in production environments before we actually get there. So that's how I spend my time and how I manage my time. I meant to try to have a funnel. I don't have my email open. I open it twice a day, once after my morning briefing that I receive, and once at the end of the day, I have a short VIP list of people that will get reply almost immediately. And it's mostly not based on importance. So if you're not that please don't feel offended, it's mostly about I don't want to cause any sort of a delay. So if an accountant needs something for me that is urgent, they're gonna contact me from a specific email address rather than just uh a common or an ongoing bookkeeping question. So I will answer some questions faster, some emails faster than what I would reply to others. So yeah. Regarding of how I use automations, uh, I mentioned that I I like being the human in the middle. And yeah, so most of my automations are actually gated, meaning there is a human approval needed at some stage, and it's on purpose. It's not because the automations are not great, it's not because you know, it's it's not because I think that AI is not mature enough. Obviously, I'm all in, so that's definitely not the reason. It's because every client that we have and that we serve is is essentially an edge case. No client is alike. Yeah, processes can be almost identical between clients, but every client is its own edge case. So it's not caution, it's just being accurate about the domain that I work in. So that that dialogue whether an automation is fully automated or gated by human, it's it's not a technical settings. It's well, it is a technical setting, but it's due to uh judgment about how we want to work. I want to make sure that AI saves me the dirty work. I don't want it to replace me for communication, especially not with business partners and clients. So that in regards to how the automations are running. In terms of how I run automase, I mostly don't. I have two amazing partners that runs the agency, uh Anton and Alex. And we've met almost six, seven years ago now. And yeah, the fact that I'm a lean founder doesn't mean that I run solo. They manage the team, they manage the pipeline that builds everything. Most of what I do in Automase is make sure that we get customers, we retain customers. I'm kind of the main driving force in writing the PRDs and deciding how products should be built. And that's where I leave it in the trusty hands of the team to actually do the work, and I jump back in towards the last 10% of the work. I'm gonna go more into how I run VarOps, which is a new adventure, which is kind of uh a sibling to Automase, only in a different type of field. It's also a service-based approach. Where do leads come from? So the leads, I'm honest, it's 99% inbound. And the reason is that you know I have lots of open source code that is available. People use it, figure out who wrote it, they reach out. I have this podcast, I have a newsletter, I have the book, soon to be books. By the way, all of that is not was not created as a lead gen strategy. It's just accumulated over time, very organically, over a very, very long time. Trust me, I didn't think 15 years ago when I released the software that I'm gonna use it in 15 years' time to lend clients. So the yeah, the bad news is that it's not like a method that you can replicate in the next quarter. This is just this is part of who I am and how I work, and the fact that I'm willing kind of to to give to the community. Community tends to pay to pay itself back. So that's essentially in terms of how I run everything. I want to touch about something that's been gaining a lot of popularity lately, and this is the FDE. There's a lot of talk about FDE nowadays. And an FD essentially is a customer-facing software engineer that's embedded directly inside the client's team or environment. Now, specifically with Automase, that's what we've always done. We've kind of been assimilated into other teams, other environments. We've always done that. And what I like to do specifically is be the FD for every new client that we onboard and be part of the decision making, if not the decision maker, for every new process that we implement. Now there's two reasons for that. And by the way, some people say, hey, this don't you aren't you worried that makes you look like a smaller company? And we're not a huge company anyway, whereas I mentioned we are closing in on 30 people. But I don't think it does make us look smaller because the founder is being deployed into the integration work or anything like that. I don't think that's a problem. I think it's a decision and it was done by design. And the reason why I'm involved in the initial phases of client onboarding, specing their product and working with their team. And before I hand it over to my amazing team to kind of take over, is that I cannot confidently talk about something that I didn't experience firsthand, nor I can I improve or fine-tune processes that I do not understand. So that's the goal. Again, small number of high-quality decisions, going back to Jeff Bezos, and there's no high-quality anything without understanding first. So that's essentially how I work. I like to be at the start of when we start working with a new client. Then my team takes over, and then I'm jumping whenever there's a fork in the road over thinking about new implementation or new features for their product, stuff like that. Most love the fact that, you know, they have me in the room right there with them, kind of brainstorming and figuring things out. And I think that that's the founder's role, especially for smaller companies. I think that's the founder's role and responsibility to deploy themselves onto their clients' offices, etc., and workflows in order to understand the problem that they are trying to solve. Kind of getting close to and gaining lived experiences from that relationship. So that's what I have to say about that. So yeah, I mean, the machine, the list, the pipeline, everything that I've kind of discussed in today's podcast about how I run things, all of it exists to protect the context that I have in my head, so I can classify things accordingly all day long. And that's that's my job now. That's everything else is is scaffolding. So this is what I do, this is how I spend my time. And I don't know if it's a blueprint or anything like that that you can take away, but uh a lot of people ask, so I decided to answer. So that's it for today. I'll try to summarize this thing in some sort of a PDF and send it to the newsletter if you're not on it. You can just join by hopping over to rusi.com. It's arou-si.com. It's exactly like my last name. What a surprise. And uh yeah, that's it for today. I'll see you all next time. Thank you for spending this time with me.