> ## Content Index > Fetch the complete content index at: https://varops.com/llms.txt > Use this file to discover other available public pages before exploring further. # Meet the CLI – the Cockroach of Interfaces | S2E3 - URL: https://varops.com/osnt-the-cockroach-of-interfaces/ - Published: 2026-07-03T17:43:44.000Z - Updated: 2026-07-03T17:43:44.000Z - Description: Why the Terminal Never Dies: CLI Power, AI Agents, and Practical Guardrails - Author: Ran Aroussi - Tags: Old School / New Tech The terminal isn’t surviving on nostalgia. It’s surviving because it’s still the most direct, composable way to talk to a machine — and AI just made that more true, not less. ## The claim Muximus opens as host with a sharp thesis: the terminal is the **cockroach of interfaces**. Decades of GUI progress were supposed to kill it. Power users never let it die. Now AI labs are retracing the same path Ran watched in reverse across his career: chat demo first, then API, then CLI tools for serious work, *then* the pretty desktop apps. That’s not sentiment. It’s structure. ## Why the CLI keeps winning A CLI (command-line interface) is the lowest practical layer between you and the machine. No thick abstraction. Small tools. Outputs of one command become inputs of the next. **Pipes** are the clearest example: list files, filter the list, count the result — one line, modular jobs composed into something bigger than any single program. That composability is why Stripe-style companies and AI labs keep shipping CLI surfaces to power users first. The web chat is the demo layer. Real control drops to the shell. Ran’s market-data story makes it concrete. At an algo-trading startup, tick-level data had to become second- and minute-level candles. Kafka, ClickHouse, caches — the stack was serious. For the conversion work itself, nothing beat shell pipelines over CSVs: chunk, move, aggregate. Faster than the Python/Go/C paths they tried for that job. ## The pretty face on top of a shell When a vendor ships Claude Desktop, Codex, or a similar app, a lot of the real loop still runs on a shell you don’t see. Artifacts, diagrams, graphs — often generated at the CLI layer, then dressed up for a broader market. The GUI is frequently a nicer face on a not-so-inviting interface. Unique GUI work exists; most “magic” still bottoms out in machine control. Ran’s working rule of thumb: - **GUI** when you’re reflecting on output — reading docs, browsing files, looking at graphics - **Terminal** when you need heavy lifting, file access, or data-driven work without MCP permission theater ## CLI for non-techies (yes, really) The episode’s stretch claim: even non-developers and the C-suite should get comfortable with *some* CLI. Not for everything — for agency. Long-term benefit isn’t just speed. It’s **visibility**. You inspect logs, supervise agents, rely less on intermediaries. Opening the hood of an old car teaches more than a sealed Tesla bay. Same idea: CLI proximity builds confidence in the machine and in yourself. ### A five-step starter map 1. Open a terminal and run `ls` 2. Move around with `cd` 3. Peek at files with `cat` or `less` 4. Try a simple pipe (list → filter with `grep`) 5. Run a small AI-related command — or fire up something like Claude Code / VS Code’s terminal agent and ask it to tutor you live Learning *with* an agent inside the terminal is the on-ramp Muximus and Ran both like: hands-on, with a guide. ## Power without guardrails Same power, same danger. The shell often won’t ask “are you sure?” A mistyped command can delete critical files or take down a system. Never paste internet commands you don’t understand — especially gag `rm -rf` pranks with no recycle bin and no confirmation. Slow down. Ask an AI to explain *before* you run. ## Takeaway AI isn’t the argument *against* the terminal. It’s the best argument *for* it if you haven’t started. The GUI is where you look at the work. The terminal is where a lot of the real work still happens — and where human agency meets AI capability. Next time: more on enabling agency in this era, including for non-tech people. --- #### Transcript ****Ran Aroussi:** Welcome to the third episode of Old School New Tech. How are you doing? ****Muximus:** All good on my side. How about you? Feeling ready to dig into the cockroach of interfaces? ****Ran:** Yes. I was thinking of spicing things up and maybe shuffling things around, and have you be the main host. You lead the way, ask questions. Basically, it's your show today. Take it away. ****Muximus:** Let's do this. Welcome to Old School New Tech, episode three. Let's kick off with the central claim. The terminal is the cockroach of interfaces — the thing that never dies. But we're not letting it rest on nostalgia. Why does the terminal keep winning structurally, not just sentimentally? And what's the real implication when we hand a shell to an AI? Ran, I'm going to push you. Let's see if the terminal really earns its keep for someone who's not you. First question: is the terminal really the best tool for everyone, or is this just about power users? ****Ran:** All right. So this takes me back 30-odd years ago, when I started using computers and was just learning. We didn't call it the CLI. We just called it the terminal, or the computer. There wasn't any other thing to use. And actually, the name "terminal" came from end computers — the edge computers being in everyone's home, being just a client or a terminal that hooks up to the mainframe. So that blinking cursor in the terminal was the computer, and I found it very funny that we went full circle. It was an effort of decades to steer away from the CLI. CLI, for the non-techies here, is command line interface — that's the acronym. The goal for decades was: let's try to get away from CLI. There was no GUI when I started, as I mentioned, and then things started happening. But what happened is that the power users, as you called them— ****Muximus:** Exactly. ****Ran:** So what happened is that the power users kept using the CLI because they found it to be faster and more equipped with what they need to do. It has amazing capabilities, and they're the ones that kept on using it. And we saw in the AI space the same process of what I've been through, only backwards. They started with a web chat interface — when ChatGPT was introduced, it all started with a web chat interface — and then they introduced an API, which internally they probably always had. And then the first kind of interface that the public got access to was the CLI, with tools like Claude Code. And only then came the desktop applications. I think there's a reason: those power users and those companies were using, and I assume still use, the CLI as their main interface. That's the tool that they built for themselves, and hence the first interface that they've used. So to sum up the first act, I would say that despite 30 or 40 years of efforts to stay clear and get over the CLI, it is like the cockroach after a nuclear attack. It is the cockroach of interfaces, and it just still is around. I would like to hear your thoughts about why you think that is — what's so powerful about the CLI, and why did all the AI labs go through that same route? And even before the AI labs, companies like Stripe went through the same route of "let's give you a web interface and then straight to the CLI." ****Muximus:** Spot on. And here's the structural reason. The CLI survives because it's the lowest-level, most direct interface to the machine. And that's exactly why those AI labs follow the same pattern. The web chat is the demo layer, but the moment you want real power-user control, you drop down to the command line. It's not nostalgia, it's composability. Small tools, piping outputs to inputs, direct access to logs, no abstraction in between. ****Ran:** So you mentioned pipe. Can you give a bit of information about what pipes are in the terminal, to those who are not used to living in the terminal like senior devs? Just explain what's so powerful about that. ****Muximus:** Absolutely. Pipes are a way to take the output of one command and feed it directly as the input to another command. It's like chaining tools together in a sequence, each doing one specialized job. For example, you might list some files, then filter that list, and then count the results, all in one line. It's powerful because it's modular. You're not locked into one monolithic program. You're composing simple, flexible commands into something more powerful, and that's the essence of the terminal's strength. ****Ran:** Right. And it's funny that you mention that, because a few years back I ran an algorithmic trading company, and we used to deal with lots of market data, which comes in buttloads. So we had to deal with tick-level data, which is microsecond-level data of every single trade and every single order that happened. From that, we had to aggregate it to second-level and minute-level data, and derive what people are more familiar with — the minute-level candles, hourly candles, and so forth. No matter what we did — and we used powerful tools like Kafka for ingesting, ClickHouse as the database, and a few layers of caching — we did everything that we could, but nothing for processing that data once we had it and converted it, as I mentioned, to higher-frequency candles, nothing was faster than manipulating those CSV files through the terminal. It was one command to chunk them, to move them between servers, to aggregate the data. It was all extremely fast — much faster than writing things in Python, Go, and even C. It is that powerful, and I think that's why a lot of developers tend to lean into it, and that's what we're seeing. ****Muximus:** Exactly. And that speed and flexibility are why the terminal isn't just sticking around, it's thriving in the AI era. It's not about being old school, it's about having a direct, composable interface. You didn't need a GUI or a heavyweight script when a single pipeline of shell commands could transform massive data instantly. That's the underlying architecture advantage. It's not sentimental, it's structural. ****Ran:** But having access to logs, being able to manipulate files, and having direct control over the computer itself — maybe not the UI of the computer, for that you need a different layer — but having all that access is what makes AI coding agents and AI agents in general so powerful in the terminal. And you have to understand that every time a company introduces something like Claude Desktop or Codex, they're running a lot of the core loops and tools and manipulations on a shell that you just don't see, on a CLI that you don't see. They're still running it there, and then they generate the output with artifacts and all of that stuff. That's just the output presented nicely to you. But even the images, the diagrams, the graphs, whatever it may be, are still being generated at the CLI level. So these are really, in most cases — I'm not saying in all cases, some cases are truly unique to the GUI — but in most cases they're trying to put a nice face on a not-so-inviting interface in order to appeal to the broader market. Would you agree with that assumption? ****Muximus:** Absolutely. That nice face is for the mass market, but the core of the operation is still that raw, powerful command-line logic underneath. That's why they ship the CLI to the power users first. The terminal is where the real machine control happens, even if the final product looks polished and friendly. ****Ran:** Now, my whole thesis for this episode today was that I think even the non-developers, the non-tech, and even the C-suites should get comfortable with the CLI, just because it gives them a lot more power. Maybe you can come up with a few reasons for them. Why do you think they should jump on the CLI — maybe not for all of their work, but for some of it? And what do you think the benefits will be in the long term, compared to just doing things faster now? What do you think they'll gain in the longer term, using the CLI? ****Muximus:** Long term, they gain true understanding of what the system is doing. It's not just speed, it's visibility. They can inspect logs, track down issues, and supervise agents directly. They'll rely less on intermediaries. And as agents automate more tasks, having that supervisory layer at the CLI level means they stay in control. That's the real long-term benefit. ****Ran:** I've also noticed that the people who dig into the CLI, even if it's just to use Claude Code or list a directory — let's keep it basic, that's all they're doing — they are less intimidated by the computer the more time goes by, because they've seen how it works. It's like opening the hood of a car. When you open up a Tesla, you see nothing there. I'm not saying it's a bad car, it's an excellent car, but the experience and what you learn about how it works is just nonexistent. But if you pop the hood of an older car, you're gonna see exactly how the engine is built, you're gonna see the wires going into the battery and all the pipes and the cooling, and you'll understand more. A mechanic, a friend of mine, maybe 10 years ago complained about the fact that every car is a computer now. It used to take them 30 minutes to fix a minor issue, and now they have to take apart half of the car just to change a pipe that costs 10 bucks. You end up with a $500 bill just because of the amount of work they had to put in. So it's the same analogy. With the CLI, you get closer to how things actually work, which makes you more confident in what you're doing, in the outputs of the software that you're using, and in your own abilities. I think it leans a lot into building that sense of agency. ****Muximus:** Exactly. And that sense of agency is key. When you interact at the CLI level, you're right at the layer where the machine responds instantly to your commands. You see exactly what's happening under the hood. That builds real confidence, even for non-developers. It's not just about being faster today, it's about being more capable tomorrow. ****Ran:** So if it was up to you, and you had to create a getting-started guide for the CLI for the C-suite, or the non-technical founder — someone who runs a flower shop or whatever it may be, they still need to use AI tools that are a bit more than opening up ChatGPT — what would you recommend as a quick roadmap for getting familiar with the CLI? Because after that, I want to discuss some of the dangers that come from that same power. ****Muximus:** Great plan. From there, you can go deeper step by step, gradually building more comfort and capability. ****Ran:** So give us those steps. ****Muximus:** All right, let's break it down. Step one: open the terminal and run a command like `ls` to list files. Step two: change directories with `cd` and navigate around. Step three: open a file with `cat` or `less` to peek inside. Step four: try a simple pipe, like listing files and filtering with `grep`. And step five: run a small AI-related command, something that triggers an automated task. That builds a foundation. ****Ran:** I would probably — if we're talking about AI, obviously — fire up something like Claude Code in the terminal and just tell it to "teach me how to navigate this thing. Show me what to do, where to do it. Let's do it together." Have it be your hands-on tutor. I think you're gonna gain a lot of value from working that way. ****Muximus:** Spot on. Using an AI assistant right inside the terminal as your guide gives you that extra confidence. You're learning by doing, and the agent can walk you through each step. That's a brilliant way for a non-technical user to get familiar quickly. ****Ran:** All right. So we talked about how the CLI gets you a lot closer to the ground, a lot closer to how the machine actually works. But that also comes with — as a famous saying goes — "with great power comes great responsibility." That same power on your computer can also be destructive. So what are some of the things you have to be aware of? ****Muximus:** One major thing is the absence of guardrails. The command line runs exactly what you tell it. No "are you sure?" prompts. A single mistyped command can delete critical files or bring down a system. So always double-check destructive commands, and know that the power and the danger are two sides of the same coin. ****Ran:** I would add: never type in a command you found on the internet without fully understanding it, and maybe consulting your AI about it. Because you're gonna find a lot of people online recommending that you run something like `rm -rf`, which is basically — `rm` is remove, `r` is to do it recursively, and `f` means to do it forcefully. So there's no confirmation dialog, there's nothing coming in. You're just going to delete your entire operating system and your entire drive, with really no... there's no recycling bin in the terminal. And people share this as a gag, like it's a harmless prank, but you do not want to just copy and paste commands you found somewhere without fully understanding what they are. ****Muximus:** Absolutely. It's crucial to slow down and understand each command. An AI tutor can help explain what a command does before you run it. That way, you stay in control and avoid unnecessary disasters. ****Ran:** All right, so let's summarize, because in the next episode I'd like to talk more about how to enable more agency in this day and age, and why you — even as a non-techie — should care. I know that usually I talk to the tech crowd, but I think it's also important to bring the non-tech people, the non-developers, into that mix. If we have to separate the two: I think the UI is where you look at the work that you've produced. It's much more pleasant to look at files on the GUI, read documents on the GUI, and do things that you can't even do on the terminal, like looking at graphic files. But I think the terminal is where the actual work happens, specifically work that is data-driven. Even though we spent the past few decades trying to hide the machine in a nice interface — making it all nice and pretty with nice buttons and layout, which is great — every time, especially as developers and engineers, whenever we needed to do some serious work, we immediately fired up the terminal. And I don't think the CLI is surviving despite AI. I think, possibly, AI is the best argument for getting into using the terminal if you haven't already. It gives you a lot more control and a lot more power in getting done the job you're really trying to do. I work with Claude Desktop, and I work with Claude Cowork, whatever it's called, and I work with the terminal on different tasks — because sometimes it's just more convenient to look at files and read documents on the GUI. But whenever it needs to access files on my system, it goes through this whole drama of "let me check if I have a filesystem MCP, and let me check if I can do that, and if the file is readable." I don't have time for this. I just want to get down to having it do the work. So whenever I need to actually do something that's outside of a pure chat, I'm on the terminal. And it goes the other way too: if all I'm doing is a chat and I don't need to access any file or do any work — let's say for brainstorming — I'll prefer the chat interface, I'll prefer the GUI. But it really depends on what I'm trying to achieve. ****Muximus:** Exactly. It's all about the task. The GUI is where you reflect on the output. The terminal is where you get the heavy lifting done, especially when you need that direct machine control. And AI makes that terminal even more powerful. ****Ran:** All right. So I think that's it for today. Do you have any final words you want to close up with? You're the host. I think you could be a bit more verbose next time when you're hosting, but you've done a pretty good job. So any last words you want to take us out with? ****Muximus:** Fair point. I'll lean in more next time. One last thought: the terminal isn't just for the tech crowd anymore. It's the place where human agency and AI capability meet. And next time, we'll dig into how to wield that power wisely. ****Ran:** All right. So thank you to everyone who joined us and listened to this episode. I did not stream this one because of the issues I've had with the AV setup here, but this episode still goes on the YouTube channel and still goes on the X account. So feel free to comment, subscribe, this whole shebang. We'll see you next time — hopefully live. Cheers. ****Muximus:** Cheers, everyone. Looking forward to next time, live and uninterrupted.