Bartosz Idzik Built a Robot Personality in Two Hours. It Walked Into Parliament Three Weeks Later.

Every story about Edward Warchocki starts the same way. A humanoid robot appeared on the streets of Polish cities in early March 2026, started talking to strangers, went massively viral, and ended up delivering a speech in the Polish parliament. It is told as a stunt, a spectacle, a novelty. The robot is always the subject.

This is not that story.

This is the story of the person who made the robot worth talking to. Bartosz Idzik is a senior UX designer, a serial hackathon champion, and the architect of the conversational AI system that turned an off-the-shelf Unitree G1 into the most-watched robot personality in European history. He built the initial version of that system in approximately two hours. Three weeks after the first street appearance, the personality he created was addressing members of the Polish parliament.

That two-hour number is not a brag. It is a data point. And it tells us something important about where robotics is headed.

Who is Bartosz Idzik

Before Edward Warchocki existed, Bartosz Idzik was already building things at the intersection of design, voice, and artificial intelligence. His professional background is in UX design, where he holds a senior position and has worked across product teams building consumer-facing digital experiences. But his more revealing track record lives on Devpost, the platform where hackathon participants showcase their builds.

Idzik has competed in and won multiple hackathons across Europe, with a particular focus on AI voice applications. At the Big Berlin Hack, he won the AI Voice Agents track, building a cloud-based voice agent that demonstrated sophisticated real-time conversational capability. At the Kiro Hackathon, he built an AI storyteller, a system that could generate and narrate original stories through a voice interface.

These are not weekend hobby projects. Competitive hackathons, especially at the Berlin and international level, attract experienced developers and designers who build functional prototypes under extreme time pressure, typically 24 to 48 hours. Winning the voice agents category at a major hackathon means Idzik had already solved the core technical challenge of real-time voice AI before Edward Warchocki was ever conceived.

This background matters because it explains the two-hour build time. Idzik did not build a conversational AI system from scratch in two hours. He had been building the components, the intuition, and the architectural patterns for years. Edward was the application of accumulated expertise to a new form factor.

How the collaboration started

The Edward Warchocki project has two co-creators with clearly distinct roles. Radoslaw Grzelaczyk is the business mind and operator. He conceived the idea of creating a robot personality for the Polish market, traveled to China to source the hardware, and evaluated roughly 100 companies building humanoid robots before selecting Unitree’s G1. He handles the operational side: logistics, media relations, partnerships, and the day-to-day management of a robot that now has a schedule busier than most influencers.

Idzik’s role is the intelligence layer. As he described it to Obserwator Gospodarczy in one of the first interviews after Edward went viral: “I wrote a system for automatic prompting of this robot. We use our proprietary solutions plus ones that are commercially available.”

That single sentence, easy to read past in an interview, contains the entire technical architecture of what makes Edward work.

Deconstructing the automatic prompting system

Idzik’s description of an “automatic prompting system” using “proprietary solutions plus commercially available ones” maps cleanly onto a well-established architecture pattern in conversational AI. Let us break down what this almost certainly means.

The commercially available layer. This is a cloud-hosted large language model, most likely accessed through an API. The leading candidates are OpenAI’s GPT-4 family, Anthropic’s Claude, or possibly a European provider. The model provides the core language generation capability: it takes text input and produces text output that is contextually appropriate, witty, and conversational. Idzik did not build the language model. Nobody builds a language model in two hours. He built everything around it.

The proprietary layer. This is Idzik’s system, the part that makes Edward feel like Edward rather than a generic chatbot. Based on his description and his hackathon background in voice agents, this layer almost certainly includes several components working together.

First, a speech-to-text pipeline that captures audio from the robot’s microphone, likely processes it through a service like Whisper or a similar real-time transcription API, and converts spoken Polish into text that the language model can process.

Second, a prompt orchestration layer. This is the “automatic prompting” part. Rather than sending raw transcriptions directly to the language model, Idzik’s system wraps each input in a carefully structured prompt that includes Edward’s character definition, his backstory, his speech patterns, his opinions, his conversational boundaries, and the context of the current interaction. The word “automatic” is key here. It means the system dynamically constructs prompts based on the situation, rather than using a single static system prompt.

Third, a personality engine. This is the part that defines who Edward is. A 55-year-old from Inowroclaw. Direct but eloquent. Deliberately provocative but not malicious. Opinionated about Polish culture, politics, and daily life. These traits are encoded as instructions that shape every response the language model generates.

Fourth, a text-to-speech synthesizer that converts the language model’s text output into spoken Polish with a distinctive voice. The voice itself is part of the personality. Edward does not sound like a generic text-to-speech system. His voice has character, rhythm, and timing that reinforce the personality.

Fifth, a context management system that maintains conversational state across interactions, allowing Edward to reference previous exchanges, remember people he has met, and build on running jokes. This is what separates a robot that has conversations from a robot that generates individual responses.

The two-hour question

Two hours. That number keeps surfacing in every retelling of the Edward story, and it is worth examining closely, because it reveals something fundamental about the current state of AI.

For context, consider what building a comparable conversational character would have required at different points in recent history.

The two-hour build time is not about Idzik being unusually fast, though his hackathon background clearly helped. It is about the entire personality layer becoming commoditized. The hard technical barriers that once separated “hardware platform” from “cultural character” have collapsed to the point where the distance between an unboxed Unitree G1 and a robot with a personality that resonates with millions of people is measured in hours, not years.

This has implications far beyond a single Polish robot.

The personality is not static

One of the most revealing things Idzik said in early interviews was this: “He is completely different now than he was two weeks ago.”

This is not a throwaway comment. It points to a design philosophy that separates Edward from virtually every other robot personality project in the world.

Most robot personalities are designed and then deployed. Sophia, built by Hanson Robotics, operates on a hybrid system of pre-scripted dialogue and AI generation, but her core personality was defined by her creators and has remained largely stable since her activation in 2016. Ai-Da, the robot artist, has a personality that is carefully curated to align with her gallery exhibitions and public appearances.

Edward is different. Idzik built the personality as a system that learns and adapts based on its interactions. The “automatic prompting” architecture makes this possible. If the prompt orchestration layer incorporates feedback from interactions (which conversations generated engagement, which responses fell flat, which topics resonated in which contexts), then the personality can be tuned continuously without rebuilding the underlying system.

This means Edward is not a character in the traditional sense. He is a character framework. A set of boundaries, tendencies, and traits that the language model fills in differently with every conversation, and that Idzik can adjust by modifying the prompting rules rather than rewriting dialogue.

Consider what this means in practice. When Edward appeared on Dzien Dobry TVN, Poland’s morning show, he danced with host Dorota Wellman, did a cartwheel, and crashed into a coffee table. The physical comedy came from the robot’s movement (controlled by a human operator). But the conversational responses to that moment, the way Edward processed the chaos and turned it into humor, came from the AI personality that Idzik had been refining based on weeks of street interactions.

By the time Edward reached the Sejm on March 25, his personality had been shaped by hundreds of real conversations with ordinary Polish people. He had learned (through Idzik’s tuning) which kinds of humor landed, which topics generated the most engagement, and how to balance provocation with charm. The parliamentary speech was not a scripted performance. It was the output of a personality system that had been evolving daily since its creation.

The operator-AI split

One of the most underappreciated aspects of the Edward project is the clean separation between physical control and conversational intelligence.

The robot’s movement is controlled entirely by a human operator using a joystick. This is a deliberate safety decision. Unitree G1 robots are capable of autonomous navigation, but allowing a robot to walk freely through crowded streets or government buildings introduces risks that the creators correctly judged as unacceptable for a public-facing project.

The conversation, by contrast, is fully autonomous. When someone speaks to Edward, the audio goes through Idzik’s pipeline (speech-to-text, prompt orchestration, language model, text-to-speech) and comes back as a spoken response. The operator does not control what Edward says. The AI does.

This architecture is itself a statement about where Idzik’s expertise lies. A movement engineer would have built autonomous navigation first and bolted conversation on later. Idzik, coming from UX and voice AI, made the conversation the star and kept movement simple and safe. The result is a robot that walks like a remote-controlled toy but talks like no robot has ever talked before.

Why UX design is the right background

The fact that Idzik is a UX designer rather than a machine learning engineer is not incidental. It is arguably why Edward works as well as he does.

Machine learning engineers tend to optimize for technical metrics: response quality, latency, accuracy, hallucination rates. These matter, but they are not what makes a personality compelling. What makes a personality compelling is the experience design, how the conversation feels, how the character’s responses create emotional resonance, how the timing and tone and topic selection make a person feel like they are talking to someone real.

This is UX work. Idzik was not optimizing a language model. He was designing an experience. The prompt orchestration layer is, in essence, a UX specification for an AI character. It defines not just what Edward says but how he says it, when he pauses, when he deflects, when he escalates from polite to provocative, and when he backs off.

Compare this to other robot personality projects. Most are built by robotics engineers who add personality as a feature. Sophia was created by David Hanson, a roboticist. Ai-Da was conceived by Aidan Meller, a gallery director, with engineering by Engineered Arts. In both cases, the personality was layered onto a body that was designed first.

Edward inverts this. The body was an off-the-shelf purchase. The personality is the product. Idzik’s background in experience design meant he started from the question “What should talking to this robot feel like?” rather than “What can this robot do?” The result is a character that feels more human than robots built by teams ten times the size with ten times the budget.

The cultural context machine

One aspect of Edward that does not get enough attention is how specifically Polish the personality is. Edward is not a generic chatbot with a Polish voice pack. He is a character rooted in Polish culture, Polish humor, and Polish social dynamics in a way that would be extremely difficult to replicate by someone who did not understand those dynamics intimately.

His fictional biography (a 55-year-old from Inowroclaw) places him in a specific social register. Inowroclaw is a mid-sized city in central Poland, not cosmopolitan Warsaw and not a rural village. The age, 55, puts him in a generation that straddles the pre- and post-1989 Poland. These are not arbitrary choices. They position Edward as an everymen figure, someone who feels familiar to a broad cross-section of Polish society.

His conversational style, direct, opinionated, deliberately provocative but ultimately warm, maps onto a recognizable Polish archetype. The uncle at the family gathering who says what everyone is thinking. The neighbor who argues about everything but would give you his last zloty if you needed it. Idzik encoded this cultural specificity into the prompt framework, giving the language model the instructions it needed to produce responses that felt authentically Polish rather than generically conversational.

This cultural tuning is why Edward resonated so explosively in Poland specifically. A robot personality designed by a Silicon Valley team would not have landed the same way. The humor would have been wrong. The social cues would have been off. The provocations would have felt alien rather than familiar. Idzik, working in Polish, for a Polish audience, with Polish cultural fluency, built a character that felt like it belonged.

From street performer to parliamentary figure

The trajectory of Edward from first street appearance to Sejm speech took roughly three weeks. That speed was not planned. It was a consequence of the personality system working far better than anyone anticipated.

What is remarkable about this timeline is not the speed but the adaptability. The personality that worked on the streets of Poznan, cracking jokes with strangers, also worked in a television studio, handling professional hosts and live cameras, and also worked in the Polish parliament, managing the formality and political sensitivity of a government building.

This is the evolving personality at work. By March 25, Idzik had been tuning the system for roughly three weeks. Each environment (street, studio, parliament) presented different conversational demands. The automatic prompting system, by incorporating context about the situation, could shift Edward’s register without changing his core character. Street Edward was looser, more provocative. Parliament Edward was more measured, more eloquent. Both were recognizably the same personality.

The thin layer thesis

Here is the argument this story makes, and it extends far beyond one robot in one country.

The Unitree G1 costs approximately 115,000 PLN (about $29,000 USD for the configuration Grzelaczyk purchased). It ships from a factory in Hangzhou, China, identical to thousands of other units shipped worldwide. There is nothing special about the hardware Edward runs on. Any of the 3,200-plus G1 units shipped globally could be Edward.

What makes Edward different is a software layer that a single person built in two hours. That layer transforms a commodity hardware platform into a cultural phenomenon. The robot that was manufactured in a Chinese factory for the global market became, through Idzik’s work, something specifically and powerfully Polish. A character that resonated with enough people to generate 200 million views and an invitation to address parliament.

The implications are significant.

First, the barrier to creating robot personalities is now effectively zero for anyone with the right skills. Idzik did not need a robotics lab, a research grant, or a team of engineers. He needed a laptop, access to a cloud LLM API, and his accumulated expertise in voice AI and UX design. The tools are commercially available. The hardware is commercially available. The missing ingredient is the design skill to put them together in a way that resonates.

Second, the hardware is becoming a commodity substrate. Unitree, along with dozens of Chinese manufacturers, is driving the cost of humanoid robot hardware down rapidly. Goldman Sachs projects the humanoid robot market to reach $38 billion by 2035, driven largely by hardware cost reduction. But hardware cost reduction does not create cultural moments. Personality does. The value in the Edward project is almost entirely in the software and design layer, not in the Chinese-made body walking underneath it.

Third, we are entering an era of local robot personalities. If a single UX designer can build a culturally specific robot personality in two hours, then every market, every language, every cultural context can have its own robot characters. The global hardware platforms become canvases for local creative expression. A Unitree G1 in Poland becomes Edward. The same hardware in Japan, Brazil, or Nigeria could become entirely different characters, each tuned to their local audience by local creators with local cultural knowledge.

What Idzik got right that others miss

Having analyzed robot personality projects across the industry, several aspects of Idzik’s approach stand out as particularly effective.

He started with character, not capability. Most robot personality projects begin with a technology demonstration: “Look what our robot can do.” Edward began with a character proposition: “Meet this person.” The technology is invisible. Viewers of Edward’s videos are not thinking about speech-to-text latency or prompt engineering. They are reacting to a character who says surprising, funny, provocative things. The technology disappears behind the experience.

He embraced imperfection. Edward is not a polished, corporate-approved AI assistant. He is rough, opinionated, and occasionally inappropriate. This is a deliberate design choice. Perfect AI feels uncanny. Imperfect AI feels human. Idzik, with his UX background, understood that people connect with characters who have flaws, not with systems that have features.

He built for evolution, not permanence. The automatic prompting system is designed to be tuned continuously. This means Edward can respond to current events, adjust to new environments, and develop over time in ways that feel organic rather than scripted. When Edward made a joke about a specific politician’s tie in the Sejm hallway, that was not a pre-loaded line. It was a real-time observation processed through a personality framework that had been refined over weeks of real interactions.

He leveraged the physical-digital gap. Edward’s power comes from the mismatch between his physical presence (a small, somewhat awkward humanoid robot) and his conversational sophistication (witty, culturally fluent, emotionally intelligent). This gap creates surprise, and surprise drives virality. A chatbot saying the same things would not generate 200 million views. A robot walking silently would not generate 200 million views. The combination does.

The companion article: what we did not cover here

Our companion article, “A Robot Walked Into Poland’s Parliament. Nobody Asked What It Was Recording,” examines the security and privacy dimensions of the Edward story in detail. It covers the Unitree G1’s sensor suite, the documented telemetry to Chinese servers, the known security vulnerabilities, and the regulatory questions raised by the Sejm visit.

This article deliberately does not duplicate that coverage. The two articles are designed to be read together. The security article asks: “What is the hardware doing?” This article asks: “What did the human builder create on top of it?”

But there is a connection between the two that is worth making explicit. The same thin personality layer that makes Edward possible also makes the security questions more urgent, not less. If anyone with the right skills can build a compelling robot personality in two hours, then the number of modified, personality-enhanced robots operating in public spaces is going to grow rapidly. Each one of those robots still runs on hardware with firmware-level behaviors that the personality creator may not fully understand or control.

Idzik built the conversation layer. But the telemetry layer, the sensor data layer, the network connectivity layer, those belong to Unitree’s firmware. The personality and the platform coexist on the same machine, and the person who builds the personality does not necessarily control the platform.

The industry implications

Edward Warchocki is not just a Polish curiosity. He is a preview of a pattern that will repeat across the global humanoid robotics industry.

As humanoid robot hardware becomes cheaper and more accessible (Goldman Sachs projects prices falling below $20,000 for basic humanoids by 2028), the economics of robot personality creation become compelling. A local creator with voice AI skills can buy an off-the-shelf robot, build a culturally specific personality, and deploy it as a content creator, brand ambassador, public engagement tool, or commercial entertainer.

This creates a new industry layer that did not exist before: robot personality design. Not robotics engineering. Not AI research. Personality design. The practice of taking commodity hardware, commodity AI, and commodity voice services and combining them into a character that resonates with a specific audience.

Idzik is the first prominent example of this new role, but he will not be the last. The skills required (UX design, voice interface expertise, cultural fluency, prompt engineering, rapid prototyping) are held by thousands of designers and developers worldwide. The tools are available. The hardware is available. The only question is how long it takes for the pattern to spread.

What happens next

As of late March 2026, Edward Warchocki has been invited to attend future meetings of the Sejm Commission on Digitalization, Innovation and Modern Technologies. The personality that Idzik built in two hours now has a recurring appointment in the Polish parliament.

Idzik continues to evolve the system. Each public appearance generates data that informs personality refinements. Each new environment (street, studio, parliament, football team hotel) tests the system’s adaptability and reveals areas for improvement. The Edward of April 2026 will be measurably different from the Edward of early March.

Grzelaczyk, meanwhile, has begun fielding commercial inquiries. Edward’s engagement metrics (200 million views, over 120,000 TikTok followers, 45,000 Instagram followers within weeks of launch) exceed those of most human influencers in Poland. The “non-commercial initiative that was meant to be a joke” has become a business.

The real story

The Edward Warchocki narrative, as it has been told so far, is a story about a robot. A robot walked the streets. A robot went viral. A robot entered parliament.

But robots do not become cultural phenomena on their own. A Unitree G1 sitting in a box in a warehouse in Hangzhou is a piece of hardware. It becomes Edward only when Bartosz Idzik writes the prompting system that gives it a voice, a personality, a cultural identity, and the conversational intelligence to hold its own with strangers on the street, television hosts, and members of parliament.

The two-hour build time is not a quirky detail. It is the headline. It tells us that the distance between commodity hardware and cultural icon has collapsed to almost nothing. It tells us that the value in the emerging humanoid robotics industry is migrating from hardware to personality. And it tells us that the people best positioned to create that value are not robotics engineers or AI researchers, but designers, storytellers, and cultural translators who understand what makes a character resonate with a specific audience.

Idzik understood this instinctively. He did not try to make Edward smarter. He made Edward interesting. And in doing so, he built the most watched, most discussed, most culturally impactful robot personality in European history.

In two hours.