This article is a part of Poland Unpacked. Weekly intelligence for decision-makers
Wordware was on track to make serious money, but Filip Kozera decided to change course. The result is Sauna, an AI assistant designed to handle professional tasks autonomously. The company plans to spend USD 1 million on its Polish launch, establish an office in Warsaw and raise another major funding round.
Two years ago, Wordware raised around PLN 120 million (USD 30 million) in a seed funding round, at a very early stage of development. The startup attracted prominent investors with a platform that allowed users to build AI agents using natural language. Today, the company is heading in an entirely different direction.
Wordware has radically changed its strategy, focusing instead on a new product called Sauna. The AI agent is already operating in the United States and is now launching in Poland.
"It was a difficult decision because things were going well. The team was against it. They kept telling me that our business was working. I could see that too, but I had promised our investors, myself and everyone on the team that we were building a company that would eventually be worth USD 100 billion or even USD 1 trillion. I couldn't see that happening with our previous product," Filip Kozera, Wordware's founder and CEO, admits in an interview with XYZ.
He is now convinced that achieving this ambition is possible. He has chosen Poland as the first international market for Sauna's expansion.
"Maybe I should have chosen another country. Brazil, Germany... But I love Poland and the Polish people. I feel we have to do something here," the entrepreneur explains.
They changed strategy despite rapid growth. The team opposed the decision
Kozera admits that changing the company's direction was far from easy.
"We were well on our way to USD 10 million in annual recurring revenue (ARR). But then something happened that nobody expected: vibe coding, or building applications by giving instructions to AI. It advanced so quickly that I no longer believed we could reach USD 100 million in revenue over the long term, let alone USD 1 billion," says Kozera.
Changes in startup financing were equally important. Until recently, quickly reaching USD 10 million in ARR could justify a very high company valuation. Today, investors expect even greater growth potential.
Kozera therefore decided that Wordware would focus on the project now known as Sauna.
Investors reacted differently to the decision. According to the entrepreneur, Spark Capital, which led the startup's previous funding round, supported the strategic shift. One Polish angel investor, however, reportedly demanded the return of their investment.
"Spark Capital gave us incredible support. In San Francisco, investors don't approach founders with instructions about what they should do. At this stage, they can advise you on how to do something once you've made the decision," Kozera explains.
USD 1 million for Poland. Sauna's first major test
Poland is the first market outside the United States where Wordware intends to develop Sauna on a larger scale. For now, the company has no plans for an equally intensive expansion elsewhere.
According to Kozera, the promotional campaign in Poland will cost around USD 1 million. The company plans to spend as much as USD 10 million on another major launch in the United States. The results of the Polish experiment could therefore influence its future marketing strategy.
The plan is to give 10,000 Polish users PLN 400 each in credits to spend on Sauna.
"For now, it's an experiment. But this isn't a small amount of money for us," says Kozera.
The choice of Poland also has a personal dimension.
Kozera moved to the United Kingdom at the age of 16 and later studied at the University of Cambridge. He now wants to become more actively involved in developing Poland's technology sector, which is also behind the idea of opening an office in Warsaw.
"I'd like to open a small office in Warsaw and gradually expand it. We have excellent engineers and product development specialists in Poland. We're still somewhat weaker when it comes to marketing, sales and business development," Kozera argues.
For now, however, he has not presented any concrete plans for the Warsaw office.
Sauna wants to get to work before users even ask
What is supposed to distinguish Sauna from popular chatbots?
Above all, the way it initiates work.
With a traditional AI assistant, users must specify the task they want completed. Sauna, by contrast, is designed to use information available through connected services to identify tasks requiring attention, prepare solutions independently and present them to users for approval.
"Even I, as the head of an AI company, often don't know where I could use AI. When you're staring at an empty box where you're supposed to type something, it's very difficult to see how AI could help you.
"That's why Sauna makes the first move. During the first few days, it learns how you work. Then, every day, it suggests specific tasks and tries to get the work done for you before you've even thought about it," says Kozera.
Wordware has integrated Sauna with thousands of tools. It is also developing mechanisms that allow AI agents to collaborate with one another.
A supervising agent can coordinate the work of other agents, while users can communicate with the system through services including WhatsApp and iMessage.
The basic idea is simple: reduce the need to constantly switch between applications, identify tasks manually and remember every obligation.
Expert's perspective
AI agents are already taking work off employees' hands
Whether agents can take over most responsibilities, however, depends on the job. At scale, they need access to company knowledge, tools and permissions.
In our implementations, we treat agent-based systems as a company's operating system, connecting AI with organizational knowledge and processes. That becomes a decision for the entire organization, not just an individual employee.
I also wouldn't equate caution with reluctance. Users are caught between predictions that entire professions will disappear and warnings about potential risks.
Instead of sensationalism, we need specifics: which tasks can safely be delegated, and how much work does this actually eliminate?
AI agents in practice
Kozera demonstrates Sauna's capabilities through his own experience.
Every morning, he receives a summary of current matters and plans from the agent. The assistant independently suggested adding a motivational voice message to his morning briefing. It also detected a discrepancy between a product's advertised price on a website and its actual price.
In everyday work, the agent can draft replies to messages and suggest several possible responses. It also learns users' preferences from the decisions they make.
It can identify potential job candidates, prepare emails and LinkedIn invitations, and then present them for approval.
Kozera uses a dedicated dashboard to review and approve the proposed actions. He claims that Sauna built the dashboard for him too.
"A large part of my job now consists of approving what Sauna has prepared on its own. I'm no longer doing the main work. I just check in and see what's happening.
"I'm often out cycling, and I'll leave a voice note about an idea. Two hours later, I have a finished video for my social media," Kozera explains.
He describes the sales model as business-to-professional, meaning an offering aimed directly at knowledge workers.
He wants to reach users who need relief from tedious, repetitive responsibilities.
Expert's perspective
Fewer tasks, greater responsibility
Does it relieve me of most of my responsibilities? When it comes to performing tasks, yes. When it comes to accountability, no.
The work shifts from carrying out tasks to checking their results.
Over 12 days, the agent treated a preview of several messages as if it were an entire email thread on four occasions. In one case, it saw just five of 84 messages. Based on that incomplete information, it presented an outdated account of the situation.
I added a rule and an automatic reminder, yet the mistake happened twice more.
The most dangerous AI agent errors don't look like errors. They look like finished work.
I think many people sense this. They expected to save time, only to find themselves with a new responsibility: supervising the work of artificial intelligence.
The agent makes mistakes. Humans remain responsible
The autonomy of AI agents comes at a price.
Kozera admits that his personal assistant has already made several serious mistakes. On one occasion, it shared information with a colleague that it should not have disclosed.
On another, Sauna sent duplicate messages to around 400 recipients within just a few minutes.
"That's a serious problem. Since then, we've developed safeguards based on APIs. For some things, we don't just tell Sauna in words not to do them. We simply block those actions at the software level," Kozera explains.
The entrepreneur emphasizes that users must remain in control of the agent and take responsibility for the actions they approve.
They should also correct the system's mistakes and establish clear limits on its autonomy.
Expert's perspective
AI assistants are becoming increasingly popular
They work well in operational processes, searching for and analyzing information, content creation, customer service and programming.
Unlike the chat-based AI assistants familiar to most users, well-designed agents can plan, analyze and independently execute multistep tasks. This can deliver tangible benefits to companies.
In my view, however, they will not replace humans in areas requiring interpersonal relationships, an understanding of emotional context or significant responsibility for decisions.
According to an October report by HRK, nearly 77% of professionals and managers in Poland use AI.
They primarily rely on assistants, such as chatbots, for simple everyday tasks. They often do so without adequate training or appropriate licenses, and sometimes without their employer's knowledge. This can put corporate data security at risk.
AI agents, by contrast, are used by a much smaller group.
Using them requires more than asking questions in a chat window. Users must be able to set objectives, monitor outcomes and delegate tasks, something we often struggle with even when working with other people.
The barrier is therefore more organizational than technological.
There is considerable work ahead, from education and carefully planned implementations to ensuring security and building a culture of responsible AI use within organizations.
Tens of thousands of users and a billion-dollar valuation target
Kozera says Sauna already has tens of thousands of active users.
He does not disclose exact figures or revenue, citing competitive pressure.
Wordware currently employs 22 people in the United States. According to its founder, 20 of them are involved in programming to varying degrees, while he personally handles much of the marketing, advertising and content creation.
"We have engineers who spend USD 12,000 a month on AI credits. If they know how to use those credits effectively, I have absolutely no problem with that.
"A team of 20 people can now operate like a company employing several hundred, but you have to accept that you'll spend money on these tools," says Kozera.
Expert's perpspective
The problem is not a lack of enthusiasm, but a lack of skills
They perform best with tasks based on repeatable procedures that leave little room for guesswork or mistakes.
There is an easy way to test this: count how many times you need to call or message someone to complete a particular task. Each interaction is a point at which an agent may start guessing.
That is why independent professionals who control their own work currently stand to gain the most.
Problems arise when the outcome of a task depends on several people. The obstacle for the agent is no longer performing the work itself, but the difficulties of human cooperation: unclear decisions and conflicting expectations among stakeholders.
No agent will resolve those issues for us.
I also disagree with the claim that people don't want to use AI agents. They do, but they don't know how to get started.
For the first time in history, technology this advanced has reached people without technical backgrounds on a mass scale. They have not had the opportunity to develop the basic skills needed to work with it.
In the AI skills development programmes I'm preparing for companies in 2026, the word "agent" comes up more often than any other.
There is no shortage of enthusiasm. What is missing is expertise and systems thinking, both of which can be developed through education and experimentation with one's own tasks.
"We need the money because this is a battle"
Wordware is competing with the world's largest developers of artificial intelligence models.
Kozera, however, does not intend to compete with them on model size. Instead, he wants to win through product quality.
He believes the company's competitive advantages lie in integrations with other tools, the way it guides users through successive tasks and the ability of its agents to work with external systems.
To implement these plans, the company will need another funding round.
"Within 12 months, we want to raise funding at a billion-dollar valuation. That will be an important milestone for us.
"We need the money because this is a battle," says Kozera.
Despite Wordware's ambitious financial targets, its founder insists that what matters most to him is AI's potential to transform everyday work.
"I think that often 90% of people's work is reactive and just 10% is creative. I'd love to see what humanity would look like if we reversed that," says Kozera.
The name of the new product is no coincidence.
Sauna is intended to evoke a place where people can escape notifications, their phones and constant distractions.
Artificial intelligence is supposed to provide that peace of mind by taking over some of their everyday responsibilities.
Key Takeaways
- Wordware has radically changed its strategy despite the rapid growth of its previous product. Two years ago, the startup raised USD 30 million in seed funding and was on track to reach USD 10 million in annual recurring revenue. However, Filip Kozera concluded that advances in AI-powered application development tools were limiting the original platform's long-term potential. He therefore shifted the company's focus to Sauna, a personal AI agent. The decision met with opposition from the team and mixed reactions from investors, although Spark Capital supported the change. Within 12 months, Kozera intends to raise additional funding at a company valuation of USD 1 billion.
- Poland is Sauna's first major expansion market outside the United States. Wordware plans to spend approximately USD 1 million promoting the product. As part of an experiment, 10,000 Polish users will each receive PLN 400 in credits to use in the application. The campaign's results could influence preparations for another major US launch, on which the company plans to spend as much as USD 10 million. Kozera's decision to enter Poland is also driven by personal motivations, as he wants to play a more active role in developing the country's technology sector. He is also considering opening a small office in Warsaw, although no concrete plans have been announced.
- Sauna is designed to identify tasks and propose completing them before users even issue instructions. The agent draws on information from connected services and integrates with thousands of tools. It can prepare correspondence, organize responsibilities, support recruitment and create social media content. The technology, however, still makes mistakes. Kozera admits that his agent disclosed information it should not have shared and sent duplicate messages to hundreds of recipients. The company has since introduced additional technical safeguards. According to its founder, Sauna already has tens of thousands of active users, but Wordware does not disclose precise figures on its operations or revenue.