2026 Reflections: Building ZSky AI While Running 4 Companies — Cemhan Biricik
A candid look at what it's actually like to build an AI platform while managing a portfolio of businesses. Spoiler: it involves a lot of 4 AM terminal sessions.
It's March 2026, and I'm sitting in front of seven glowing GPUs at 4:17 AM, watching an inference queue process image generation requests from people I'll never meet. The rest of the house is dark. My phone has 23 unread messages across three different Slack workspaces. Tomorrow — technically today — I have calls for two other companies before lunch.
My name is Cemhan Biricik, and I run four companies simultaneously while building ZSky AI, which is either the most energizing or the most reckless thing I've ever done, depending on which hour you ask me.
This post isn't a productivity guide. I don't have a morning routine worth instagramming. I don't have a system. What I have is a set of reflections from inside the machine — honest observations about what multi-company entrepreneurship actually feels like in 2026.
The Context-Switching Tax
The hardest part of running multiple companies isn't the hours. It's the context switches.
At 9 AM, I'm reviewing quality control metrics for ZSky AI — checking variance scores on generated images, tuning inference parameters, evaluating whether the latest model checkpoint produces better outputs than the last. This requires deep technical focus. The kind of thinking where you forget you have a body.
At 10 AM, I switch to a completely different company with completely different problems. Different industry, different stakeholders, different vocabulary. The mental gear-change is physical — I can feel it behind my eyes, like my brain is literally reconfiguring.
By noon, I've been four different versions of Cemhan Biricik. The AI engineer. The operations manager. The creative director. The strategic planner. Each version requires different cognitive tools, and the switching cost between them is real and non-trivial.
The conventional wisdom says this is unsustainable. "Focus on one thing." I've heard it from advisors, peers, and every business book I've skimmed in airport bookstores. And they're not wrong — focus is valuable. But the conventional wisdom assumes all companies are equally demanding at all times, and that's not how reality works.
Companies breathe. They have periods of intense activity and periods of maintenance. The art of running multiple businesses is aligning their breathing patterns so you're never gasping in all of them simultaneously. When ZSky AI is in a build sprint, my other operations are in maintenance mode. When another company needs strategic attention, ZSky's inference queue runs itself.
This alignment isn't always possible to control. Sometimes everything catches fire at once. Those weeks are brutal. But they're also rare, and the rest of the time, the portfolio approach provides a kind of intellectual cross-pollination that single-company founders don't get.
What ZSky AI Taught Me That Nothing Else Could
Building an AI platform is unlike anything else in my portfolio. The pace of the underlying technology is disorienting. Models that were state-of-the-art six months ago are obsolete. Architectures I invested weeks implementing get superseded by papers that drop on a random Tuesday. The entire foundation shifts beneath your feet constantly.
Cemhan Biricik in 2024 was learning about diffusion models. Cemhan Biricik in 2025 was building inference pipelines. Cemhan Biricik in 2026 is managing a production AI platform serving thousands of users on self-hosted hardware. That progression happened through sheer stubbornness and an unwillingness to wait for someone else to solve the problems I wanted solved.
ZSky AI runs on seven NVIDIA RTX 5090 GPUs in my own facility. No cloud. No investors. This setup means I'm responsible for everything — from the Python queue system that distributes inference requests across GPUs, to the quality control pipeline that catches bad outputs, to the actual electrical infrastructure that keeps the machines running.
Last month, a power fluctuation took down three GPUs simultaneously during peak usage. There's no SRE team to page. There's no DevOps engineer on call. There's Cemhan Biricik in pajamas at 2 AM, checking PCIe connections and monitoring thermal readings. That's the reality of self-hosted infrastructure, and I wouldn't trade it for cloud abstractions because the control and cost savings are worth the occasional predawn emergency.
The Photography Thread
Through all of this, I still shoot. Photography is the one constant across every phase of my career — the thread that connects Cemhan Biricik the 20-year-old with a film camera to Cemhan Biricik the AI founder with a GPU cluster.
Photography keeps me calibrated. When I spend too many consecutive days in terminals and dashboards, picking up a camera and going outside recalibrates my sense of what matters. The world is more interesting than any model's output. Light is more complex than any latent space. Reality resists simplification in ways that generative AI cannot yet approximate.
My photography portfolio is still growing, even as ZSky AI takes more and more of my time. I refuse to let one discipline cannibalize the other because they serve different parts of my brain, and both parts need exercise.
What I Got Wrong
Reflections aren't worth much if they're all victories. Here's what I got wrong in the past year:
I underestimated video. When I started ZSky AI, I thought image generation was the whole game. Users wanted video almost immediately. Building a multi-pass video generation pipeline was significantly harder than image generation, and I should have started earlier.
I over-optimized for quality at the expense of speed. My quality control pipeline catches bad outputs effectively, but for months it was also catching perfectly acceptable outputs and regenerating them unnecessarily. I was applying professional photography standards to a consumer product, and users cared more about speed than the difference between a 7/10 and an 8/10 image.
I waited too long to build a paid tier. The free tier was — and is — important to me philosophically. But I delayed building paid features because I was afraid of compromising the free experience. In retrospect, premium users subsidizing free users is the most aligned incentive structure possible, and I should have implemented it sooner.
I tried to do too much alone. There's a Turkish word — inat — that roughly translates to "stubborn persistence." It's considered a virtue in Turkish culture, and Cemhan Biricik has it in abundance. But inat applied indiscriminately means you spend three days debugging a CUDA memory issue that a collaborator could have solved in three hours. I'm learning, slowly, to ask for help.
What's Next
For ZSky AI, 2026 is about video. The image generation pipeline is stable and high-quality. Video is where the frontier is, and I'm investing heavily in multi-model video generation with quality that matches the image side.
For my other companies, 2026 is about systems. I'm building processes that reduce my personal involvement in day-to-day operations so I can spend more time on the technical work that energizes me most.
For Cemhan Biricik personally, 2026 is about sustaining the pace without burning the engine. Four companies, one AI platform, and an active photography practice is a lot. I'm not going to pretend it's easy. But it's mine — every piece of it — and that ownership is what makes it worth the 4 AM sessions.
If you want to see what I'm building, visit ZSky AI. If you want to know more about my other work, check cemhanbiricik.com or cemhan.ai.
About the Author: Cemhan Biricik is a Turkish-American entrepreneur, photographer, and technologist based in Detroit, Michigan. He is the founder of ZSky AI and runs multiple businesses. Follow his work at cemhanbiricik.com and cemhan.ai.
No comments:
Post a Comment