> ## Content Index > Fetch the complete content index at: https://friendlyai.dev/blog/llms.txt > Use this file to discover other available public pages before exploring further. # Fly without knowing how the plane works - URL: https://friendlyai.dev/blog/freeing-your-ai-from-your-computer/ - Published: 2026-11-11T16:52:53.000Z - Updated: 2026-11-11T23:05:29.000Z - Author: Mike Nakhimovich I fell in love with software engineering as a teenager, when I was lucky to go to a high school with a comp sci department. What interested me about programming more than the coding itself was making something new out of nothing. I loved seeing things and wondering how they could be arranged, put together, taken apart and generally made more useful. Unfortunately, as I first learned all those years ago, building software also means dealing with the real world, and the real world is full of constraints. I remember my first CS teacher, Mr. (Brian) Miller, trying to find a C++ toolchain small enough to run on the candy colored iMacs that had been donated to our class. It felt impossible at first. Then over a few weeks, I watched that my teacher wrangle that hardware into running this: ```c++ cout << "What's your name? "; cin >> name; cout << "Hello, " << name << "!" << endl; ``` Amazing! What a fascinating new world of variables, loops, functions and objects I absorbed that year! What astounded me most was that a computer was running things I never thought it would, the way a tree never expects to become a chair. In comp sci, that transformation seemed to happen out of nothing. I could build things without resources, or so I thought. I went on to work with a strange language called True BASIC and then followed in Mr. Miller's footsteps to Penn State for comp sci. All through those years I kept noticing two things: first, the flurry of butterflies when I had a new idea, and then the disappointment when I didn't have enough CPU, memory, security or other resources to pull it off. I would make animations and spend weeks getting them to run on constrained mobile devices. I would write applications that got bogged down as they grew. That pattern held for 20 years as I went from web to mobile to infra to AI engineer. Dream big, then find how to pay for it. That's how I thought about things. ### Everyone else got the easy version In those same 20 years, the world went from installed applications to web apps to phones, and now to everything being recreated with AI. Each step took something hard and made it easy and instantly available. Gone were the days of installing Corel WordPerfect (yes, I'm old) and getting a book on how it works from a bookstore. I was standing at the school bus stop with my son this morning when I had the idea for this post. It took me less than 30 seconds to go from idea to an open Google Doc on my phone. I jotted the title and closed the app. Back in the house I picked up my laptop, reopened the same doc and kept going. I didn't need to install, upgrade, maintain or host anything. It all just worked. It’s the same for my shopping, my banking, and countless other apps. Consumer services have been decoupled from the hardware in your hand a long time ago. It's become a solved problem. But it's not solved for software engineers. I'm currently surrounded by 2 Macs, a Windows and a Linux machine, and I need all of them to write software. I have to be an expert at setting up terminals, debuggers, toolchains and massive amounts of dependencies before I even get started writing code. If I switch from a blog to DB development, I need to use new tools, a new environment, maybe even a new OS. The biggest change is where we work. A lot of the job has moved off the laptop to remote boxes, cloud dev environments and CI. But the place I sit down to build is still a machine that keeps me chained to my desk. ### AI threw away the playbook Then AI made me throw away the old playbook for my job. Building with AI gives me the same delight I had all those years ago when I first started programming. With AI support, I can synthesize output or find issues I didn't think I had the capacity to solve. I can go through my backlogs, make rubber ducks for thinking things through, code, design, and do all of it a bit better than when I was doing it by hand. As a dreamer, AI is incredible. But like most things, AI comes with a cost. It takes time to stay updated. There are new tools, models to upgrade, skills to adopt, sandboxes to try, factories to implement. I still need most of the pieces I used to write software, and now I need to maintain AI tools, providers, skills and instructions on top. The juice seems to be worth the squeeze, because it's rare that I switch to a new AI process and end up going back. But staying current is a lot of work. Over the last couple of years the process of using AI for development has gone through five iterations: copy paste; autocomplete; using the terminal; web experiences, and finally orchestration and factories. Each change meant taking all your tribal knowledge, integrations and learnings and adapting them to a new way of working. Things that felt right, like RAG or lots of humans in the loop, feel antiquated now. Agents in remote sandboxes are going from bleeding edge to table stakes. Hacking on all this is my favorite part of being an engineer, but I don't think that's true for everyone. Not everyone wants to be an expert on how to use AI, they’d rather spend their energy and time on their own projects. ### What I worked on at Airbnb I spent the last few years at Airbnb working on this transition. How do we move our practices to something new without making that our whole job? My team and I spent our days making abstractions, tools, policies and lessons so the people around us could fly without needing to know how the plane works. Our goal was to enable everyone to use bleeding edge AI instantly, without needing to manage it themselves, and without the churn of switching every time something better showed up. In this project, our biggest win was moving AI agents off our company laptops and onto a server farm where 10s or even 100s of them can work together. That changed the scope and speed of the problems that could be solved. What I can do with many machines is one thing, the bigger deal is what an agent can do with them. ### Agents off your hardware Fast forward to today, and I have left Airbnb to co-found Friendly AI Labs. We began with a simple question: what if we could give an agent the tools to create, observe, steer and validate any number of other agents? With that one change (and a hidden iceberg of infra), I'm doing things I never thought were possible. I can run massive migrations over dozens of machines, watch agents communicate with each other, pick up work on my phone that I started on a laptop, and review intent as well as output. All of it runs off my laptop, somewhere safe, and it keeps going when the lid is closed. It’s been amazing to take a peek at what’s possible, but working this way requires a mountain of infrastructure. It requires secure, sandboxed compute resources that have your toolchain preinstalled and can run 1 agent or 100\. It requires credentials that are accessible but not exposed. It requires a way to talk to the agents and a way for them to talk to each other. Then all of this gets wrapped by web, Slack, voice and whatever interface comes next (maybe holographs?). It's a considerable stack to piece together before you get to access the magic. We've spent the last few years refining how to correctly do this for thousands of engineers in big tech, and now we’re ready to make this experience accessible for everyone through Hive. With Hive, you can put your agents to work while you’re out for a run, or stuck in traffic on the bus. No more figuring out whether things are secure or whether you have enough disk space for another worktree. Just pick up your device of choice and go. We are opening up a small beta to show folks our vision and we'd love for you to try it out [https://hive.codes/signup](https://hive.codes/signup?ref=friendlyai.dev)