AI House took place just outside Bucharest, over an unassuming summer weekend at the end of June. The initiative was curated by Andrei Ilie, Head of AI at Instantly.ai, and Alex Gavril, CEO of ▲promocrat and founder of Neomixers.
With support from ElevenLabs and Bolt, its premise was straightforward: bring together 50 founders, engineers, operators, researchers, and AI practitioners for two days of conversations about how AI is changing the way they build products, run companies, and make decisions.
Before the first session had even begun, attendees were asked to do something rather unusual. Around the venue, a series of boards asked some of the biggest questions surrounding AI today.
They ranged from whether LLM wrappers can become defensible businesses to whether AI will continue to reduce headcount across organizations. Each one touched on a different aspect of building with AI: products, workflows, customers, data, and organizational change. Rather than debating them immediately, participants simply cast their vote and moved on.
The boards remained in place throughout the weekend as the talks unfolded around them (quite literally). Spoiler alert, by the time participants returned to vote again on Sunday evening, many of the answers had changed.
Those questions resurfaced during technical panels, workshops, meals, and late-night chats, accumulating perspectives from people working across very different industries. While the specifics changed from one conversation to the next, the tensions underneath them were actually very similar.
The bigger picture, before the implementation
Before turning to products, workflows, and implementation, the opening sessions zoomed out to examine the direction AI itself is taking.
Emil Muthu, CEO of Neurony Solutions, opened with a presentation on the realities of moving AI from demonstrations into production, arguing that many of the hardest problems emerge only after companies begin deploying systems at scale. The technology itself is only one part of the equation; operational complexity, organizational readiness, and reliability quickly become equally important.
That broader perspective continued during the panel featuring Sergiu Neguț, President of Romanian Business Leaders and co-founder of FintechOS, Matei Psatta, Chief AI & Growth Officer at Vola and CMO at Blindspot, Silviu Gresoi, PhD researcher in Data Science and AI at the University Politehnica of Bucharest, and Aleodor Tabarcea, Engineering Manager at Stripe. The talk revolved around AI’s second-order effects: how companies organize themselves, where competitive advantage begins to shift, and which assumptions about work are likely to change over the coming years.
The LLM question: if the models are becoming commodities, what exactly are customers paying for?
What began as a discussion about wrappers gradually became about where value comes from once the underlying intelligence is available to everyone.
For Mario Popescu, founder and CEO of Tailent, it begins with the customer rather than the technology. “You need to understand what the customer is willing to pay for. It would be better to start from the money inside the customer's wallet and work your way back through the technology,” he noted.
He later illustrated the point with an analogy.
"If you're wrapping an LLM right now, you're mostly creating revenue for the LLM provider. It's like buying a chair from IKEA and putting your own sticker on it."
Alexandru Constantin, co-founder and CTO of Stock Estate, took a more nuanced view. Wrappers, he argued, aren't inherently weak businesses. Whether they become defensible depends on where the value sits.
"Many people use foundation models to build something quickly, but that doesn't mean wrappers can't have a moat. Building software has become easier. That simply means you have to think much more carefully about where your differentiation comes from," he argued.
Others approached the same question from different angles. Bogdan Enescu, VP of Data & AI at FintechOS, pointed to the strategic risk of building a company whose roadmap ultimately depends on someone else's model. In his words:
"If the underlying model changes the way it works, you're out of your business immediately."
Lucian Popovici, founder at Bridging Gaps, took the topic back toward implementation, arguing that generic AI assistants often struggle to gain adoption, while systems deeply integrated into existing workflows tend to become part of everyday work.
"I've seen companies deploy internal ChatGPT wrappers that nobody uses because they're simply not useful. The value starts to appear once AI is integrated with existing systems and the information people already work with." Popovici said.
By the end of the panel, the topics had largely left wrappers behind. The speakers disagreed on where defensibility begins, but they kept returning to the same building blocks: proprietary data, product design, deep integrations, and a clear understanding of the problem customers are actually paying to solve.
▲promocrat, Instantly.ai, ElevenLabs, and the new operating questions
If the panel discussions explored the broader questions, the company sessions focused on the work behind the scenes, exposing the experiments, frameworks, and decisions that shape how AI is adopted inside organizations.
Alex Gavril, CEO of ▲promocrat, kicked off day 2 with a deep perspective on how AI is not replacing, but repositioning people in the workplace, with references from how the AI has led to significant changes in consulting models, ranging from Big 4 examples to modern agency models.
”There are 4 stages in the AI-driven transformation flywheel: fear, adapt, productivity, domination. And the cycle resets every time new technology gets introduced. Even so, if this flywheel demonstrates anything, it’s that human expertise and trust have an essential role in AI-driven transformation. We asked the audience <<if you could automate everything your trusted partner is currently doing, would you fire them?>> Unsurprisingly, no one said they would. That is the power of human connection in business relationships.”
Alex also approached the topic from the perspective of a growth agency, sharing how AI has become part of its own internal operating model. New ideas are first tested manually, refined through repeated client work, automated where possible, and only then turned into repeatable systems and productized. AI is becoming horizontal infrastructure rather than a specialized capability, making organizational learning itself a competitive advantage, and the companies moving first are those that are quickest to rethink how work gets done.
That same emphasis on operating models surfaced in Instantly.ai's GTM session. The team’s Head of AI and event co-organizer, Andrei Leonard Ilie, broke down the mechanics behind modern outbound, showing how buying signals, ICP definition, messaging, and lead qualification change once AI becomes part of the workflow.
Outbound was framed as one component of a broader growth engine, illustrating how AI is beginning to reorganize entire go-to-market processes. In his own words: "Outbound is one vertical, not the whole engine."
ElevenLabs offered an extensive demo on the frontier of generative voice, showcasing the company's latest models and conversational capabilities. But the session spent just as much time on the responsibilities that come with increasingly realistic synthetic speech. Technical progress, deployment, and ethical safeguards were treated as an integral part of the product.
Why the format worked
The informal parts of the agenda turned out to be just as valuable as the formal one. The audience was capped and curated, which meant participants arrived with enough context to contribute meaningfully. Most attendees were already building something, implementing something, or trying to solve a specific problem inside their company.
The casual format helped as well. The barbeque, coffee breaks, workshops, and poolside mixer offered plenty of spaces for casual business chit chat. The diversity of the audience was a good mix of perspectives, too. Because AI now touches so many functions and industries, the most useful comparisons can come from people working in different verticals.
This cross-pollination created a room yard where differences between industries were compared to become a source of insight.
Member discussion