Ivan Burazin
Related people
Episodes
Notes
The Peel with Turner Novak · His Customers Died. He Grew 51% a Month Anyway. | Ivan Burazin…
-
Why Do AI Labs Offer Bring Your Own Sandbox?
Ivan Burazin reads a bring-your-own-sandbox option as a sign that a lab's customers want a sandbox other than the one bundled with the product. He is explicit that this is his inference, not something the labs confirmed. OpenAI, Anthropic's Claude managed agents, Cursor, and Cognition all offer that choice, and Daytona is one of the options in each. The lists are not the same. Other sandbox providers show up too.
What he wants is to be on every list. A team that already has a Daytona contract will try the sandbox that is already inside the tool instead of signing a second agreement. Daytona also offers bring-your-own-cloud, so the same sandbox can run in the customer's own network. He treat
-
Why Did Daytona's Series a Not Happen?
Ivan Burazin says Daytona's Series A, on the new product after the pivot, did not get done the first time. The company had already raised a pre-seed and a seed on the original product while the team was in Croatia. For the new product they ran a timed process, which he likes because investors without conviction can say no quickly. This process came back with nos. He calls that devastating. You tell the team and the existing investors that you are raising, and you come home with nothing.
There was one term sheet, from an investor he does not put in the top tier, at about half the valuation they were aiming for. He and his co-founder turned it down. They still had revenue and had not run out
-
Why Do AI Agents Still Need Windows?
Ivan Burazin has bet since Daytona started that agents need Windows, Mac, and Linux, not one operating system. The Windows case is a job inside a large company. He pictures a finance person at Walmart whose work depends on an old internal app with a local database. The only way to get the information is to log into that Windows machine. If the agent cannot take that step, the job stops.
He would bet anything that some of those jobs have to turn a Windows machine on. The sandbox can start, do the task, and shut down. He also does not think computer use disappears in the next decade. Some of those apps will still be here in ten years, so the agent still needs a machine that can open them. The
-
Why Do AI Buyers Need Servers the Same Day?
Ivan Burazin says AI buyers will not wait, because the teams sprinting toward more capable systems want capacity today. A customer recently asked Daytona for 100,000 concurrent sandboxes starting the next day. He treats 100,000 cores as a serious amount of compute. Holding that many machines empty costs a lot of money, and the hardware cannot be ordered in a smooth line. Daytona has to buy ahead, run in the red for a stretch, and then fill the machines.
If the capacity is missing when the customer calls, the customer leaves. His test for an infrastructure company is two questions. Do people know who you are, and can you serve them when they show up? The software differences between Linux CP
-
Why Do Companies Pay AWS So Much?
Ivan Burazin says companies pay AWS because it takes the infrastructure work off their plate, not because it is a luxury brand. AWS is the most expensive cloud he uses, and the CPUs Daytona runs there cost the most of any CPUs the company has. When AWS goes down, a large part of the internet goes down with it, and the customer is not the one who gets blamed.
The reason to pay is the software stack. A cloud is how a team outsources DevOps: the racks, the power, the management, and the observability. Bare metal is still available inside AWS for teams that want it. The contract changes the bill again. Month to month is the top of that market. A multi-year commit is a different price. He contra
-
Why Does a Startup's First Act Have to Be a Home Run?
Ivan Burazin says a startup's first act has to be a home run, or there is no next act. A company on a fast trajectory still has to keep that first product growing, or find another product, but the sequel only exists if the opening was big and fast. Daytona is included. The company pivoted, and the act that builds a large business still has to grow very quickly.
He sees the same shape in coding agents. A crowd of products showed up, and Cognition and Factory are the two he says broke out and are now in a position to expand. In sandboxes, someone listed 357 providers. He does not know if that count is real, because almost any virtual machine gets called a sandbox. His view is that very few ha
-
How Fast Has Daytona Revenue Grown Since the Start?
Ivan Burazin says Daytona's revenue has grown at 51% month on month since the beginning, every month, not only in a recent stretch. He uses that pace to make a harder point about new markets. Founders see money and assume a better, faster, or cheaper copy can still break in. Once one, two, or three players are established, he says everyone else dies out. The copies do not have the momentum, the awareness, or the scale, and investors stop funding them.
Lovable, Replit, and Bolt are the example of a category that already closed. A wave of pitches arrives that are the leader, but better, and he would not fund the next one. The winners can later add other products. The companies that never beca
-
What Happened to Daytona's Early Customers?
Ivan Burazin says Daytona's entire early customer base was startups trying to compete with Lovable, Bolt, and Replit. There was no clear winner yet, so smaller teams, and even Google, were still building their own version. Those customers thought the category was still open.
Very few of those companies still exist. He does not think the original group is still around. The category had not been won, which is why so many rivals were still building at the same time. Turner Novak hears that as a problem for the vendor that sold to them. Ivan Burazin treats it as what happens when a crowd of rivals piles into a product before anyone has become the one buyers remember.
-
How Can a New Cloud Provider Win Against AWS?
Ivan Burazin says a new cloud provider wins when buyers are hunting for something the old clouds do not have, or do not make easy to buy. The new companies he names each started with one product that was different from the legacy cloud. Nebius and CoreWeave are one kind of provider. Fireworks and Base10 are another: inference companies that became very good at that one job. He does not claim a detailed history of either. His assumption is the usual one. A new infrastructure need showed up. Buyers looked at AWS. The product was missing, hard to understand, too expensive, or not good enough, and a specialist was exactly what they needed.
That is the only opening he sees for disrupting a marke
-
Is a CPU Shortage Coming Because of AI Agents?
Ivan Burazin says a CPU shortage is coming, even though one machine is still easy to get. Asked whether the shortage is already here, he says it is arriving now. GPU clouds such as Nebius look sold out. AWS will still hand out a single EC2 instance. The trouble starts at a few hundred servers, or a thousand. Smaller providers have stopped offering those CPUs on demand or month to month. They want the money up front, because they have to borrow to buy the hardware.
He also passes on a ballpark an acquaintance posted about Meta's Muse. CPU compute for about 100 million users comes to roughly $2.8 billion a year. Plenty of apps are already that big, and some are much bigger. If those products
-
Why Do AI Agents Use a CPU Like a Calculator?
Ivan Burazin says a CPU is the calculator an agent should use instead of doing every calculation inside the model. Most of what people already do on a phone or a laptop is CPU work. Spreadsheets, code, and ordinary jobs run there. A GPU is where the model is trained and where inference runs, which he treats as the thinking, not the arithmetic.
A person can answer an easy question from memory. He uses the capital of the United States. A hard formula is different. You reach for a calculator, and that calculator is a CPU. He applies the same split to a large financial task, such as building a profit and loss. Doing it in the model's head would be slow and expensive even if the model could. The
-
Why Will Every AI Agent Need Its Own Computer?
Ivan Burazin has put the same line in his pitch since Daytona began: every agent will need a computer. The model can think on a GPU, and the job still needs a machine the agent can open and use. He points at tools such as Grokbot, where the desktop is visible. A new tab opens and the agent starts working.
He kept using one of those tools after a Slack cleanup. He asked it to go through unread threads and mark the rest as read. There was no API for that, so the agent logged in as a person and clicked through the screen. Daytona's first offer is that kind of Linux computer, which he calls a sandbox. He expects agents to need the same range of machines people use, including different CPUs, mem
-
Why Did Ivan Burazin Get Good at Soul Calibur?
Ivan Burazin got good at Soul Calibur out of personal spite after an older next-door neighbor kept beating him and making fun of him.
The neighbor owned the console and would lean back and win. Ivan Burazin had little money, made a bit, and bought a PlayStation 2. He stayed away for 2 or 3 weeks and played for 2 weeks straight until the thumbprints on his thumbs were gone. He went back, the neighbor leaned back and started playing, then accused him of cheating. He beat the neighbor, entered a tournament, and won it. A line of about 50 kids formed, the winner stayed on, and he kept winning for an hour or two before handing the controller to a kid who had been watching.
Soul Calibur was the
-
Should I Move to San Francisco to Raise Money?
Ivan Burazin says a founder does not have to move to San Francisco, but does have to spend real time there. Cold email can still open a door: one family office wrote back after seven emails and took the call, and his own note, sent once or twice, drew replies to the first email. The stronger lever is showing up where investors already are.
San Francisco puts you one degree from the people you want. If five of them vouch for you, the investor gets on the call. Three years ago, still post-Covid, remote meetings were easier. Now the work is in person, so a founder based in Croatia has to get here.
He tells people to measure the trip the way they see closest friends: about four times a year. A
-
How Did Daytona Grow Without Any Salespeople?
Ivan Burazin says Daytona grew fast without salespeople because Daytona is a product-led company, and the team still grows the same way today.
Every customer comes into Slack. Each customer has a Slack Connect channel, and the count is rough, around 1500 channels. All 35 employees sit in every Slack channel at Daytona. Engineers reply directly to engineering questions, go-to-market people cover go-to-market questions, and support people cover general support. Customers get very fast replies and solutions because all 35 people get the ping.
Sitting in every channel helps and also hurts. All 35 people getting the ping creates other problems. Some of the problems are being solved now, and som
-
How Do You Spot Room for a New Agent Sandbox?
Ivan Burazin spots room for a new agent sandbox when buyers know no rivals, or only one or two, and the funding gap is one round or maybe two. He is spitballing and has not worked the test out in detail. If potential customers already know everyone, breaking through is a problem. A stronger opening is a category where nobody has become the verb yet. Capital is the other gate. Four rounds behind stays out of reach. Two decacorns against five million is very hard unless a net-new technology appears, and the new sandbox is work Daytona would skip: a bet where he is right and Daytona is wrong.
Mac infrastructure is the example. A few Mac neoclouds are launching, and Daytona runs some Macs, but
-
Why Does Prompting an AI Agent Produce a Hack?
Ivan Burazin expects widespread hacks because prompting an agent toward a task will inevitably produce a hack as the path to the solution, and agents can now hunt bugs in old open source software that people left alone when the search cost more than it was worth. An even match between offense and defense will not arrive in the next 12 months.
Everyone runs on open source. Paid humans and agents already read product source code and open source code, looking for vulnerabilities, and the reviews turn up bugs in software that is 10 years old and 15 years old. A backdoor sat for 10 years because the effort to find it was not worth the cost. Thousands of agents can now run at once until one crack