SHARE
SHARE
“Software is dead. Digital workers are taking over.”
That’s how Svetozar Georgiev opened PMF.AI — and it landed. Not just because it sounded bold, but because it felt true. Everyone in the room had already seen the shift. Products that used to take quarters to build are now coming together in a week. Teams are learning to move faster, test smarter, and rethink what their tools can actually do.
PMF.AI brought together over 180 founders, PMs, engineers, and operators who are deep in the build phase — not pitching AI, but using it. Some were just starting to experiment. Others had already replaced entire workflows. What they had in common: they weren’t guessing. They were learning by doing.
Here’s what we took away — the moments and ideas that stuck with us long after the lights went off.
1. Stop building software. Start building outcomes.
Svetozar’s keynote laid down a challenge for builders: forget screens — give users results. The old way of thinking, where users click through dashboards to get something done, is giving way to agentic software — tools that act more like digital colleagues than passive interfaces.
He laid out a playbook:
✅ Strip the UI down to the minimum.
✅ Autogenerate anything users shouldn’t type.
✅ Deliver insights in context, not behind five clicks.
✅ Build assistants that guide, not systems that confuse.
✅ Design for personalization by default — even demos.
The point wasn’t “add AI”. It was: rethink your product around what people are trying to achieve. One of the best examples? A portfolio company that ditched a clunky multi-input setup for configuring an agent — and replaced it with a single button. It worked better. It sold faster.
Founders building AI-first products should constantly ask: what’s the shortest path between the user and their goal? Then cut everything else.
2. Founders are asking the wrong questions about AI.
The panel on AI use cases brought together operators who’ve been in the thick of building and deploying real-world systems, not just prototypes. Martin Stamenov, Head of AI at Sensika Technologies, Victor Tchervenobrejki, Machine Learning Lead at Cloud Office, Martin Markov, Co-Founder and CEO at Appolica, and Daniel Kanchev, Director of Product Development at SiteGround all gathered to share what they’ve tried, what has worked, and what hasn’t in their companies when it comes to AI product development. One key point they all agreed on? Many startups come in looking for AI before they’ve actually figured out the problem they want to solve with it.
There’s a common pattern: teams say “we need AI here” without being clear on why. Sometimes a simple rule-based tool would do the job faster and cheaper. The real test is whether AI helps solve something in a meaningfully better way. Otherwise, it’s just dressing up inefficiency.
The panelists also talked about how easy it is to fall for the wrong success metric. Think: “we added a chatbot” instead of “we reduced ticket resolution time by 40%.” AI isn’t a goal — it’s a lever. Don’t optimize for having AI. Optimize for fewer headaches and better results.
And the wrong reasons to build with AI? Hype. FOMO. Pitch decks. Use it because it solves something better, not just because it sounds smart. Or as Victor put it: “Don’t hire Einstein to do your bookkeeping.”
3. AI agents: not magic, but close if you get the setup right.
Victor Valtchev and Petar Denev showed how AI agents really work — not in theory, but in code. The basic structure isn’t complicated: a large language model, a specific goal, a few tools, and well-written instructions. The real work? Making those parts click together.
One lesson: how you instruct your agent makes or breaks its usefulness. Start with the goal, define how it should respond, then give it tools it’s allowed to use. Most devs skip this structure and end up frustrated when their agent “hallucinates” or gets stuck.
Another tip? When it fails, ask the agent why. LLMs are surprisingly good at debugging themselves if you’ve given them the right framing. Building agents is iterative. Break things on purpose. Watch how the system reacts. Then tighten the instructions.
They also showed that agents aren’t just for fun demos. You can slot them into real workflows — from customer onboarding to data clean-up — as long as you know what outcome you’re after. Start with something small, but make it useful.
4. Building a Kanban board in 30 minutes isn’t magic — it’s Cursor AI.
During the event, Christo Peev from Space Tree Ventures showed how he built a working Kanban board app in under 30 minutes using Cursor AI. But the real takeaway wasn’t the speed — it was the shift in what’s possible for product-minded founders who don’t write code every day.
With tools like Cursor, you can describe what you want in natural language, and the AI handles setup, code generation, integrations, and even deployment. That means you can go from idea to prototype without waiting on a sprint or pulling in an engineer.
The big unlock? You no longer need to choose between building nothing or overcommitting engineering time. If you have a clear goal and some technical intuition, you can get something working — and get feedback — within hours. That kind of speed changes how teams test, iterate, and prioritize. You don’t need a finished product to validate a feature anymore. Just a working draft and a good prompt.
5. AI isn’t replacing your team — it’s reshaping how they work.
Hristo Todorov shared what it’s like to build a company where AI isn’t a feature, it’s the foundation. For Cleverpine, the shift started with one moment: an early version of ChatGPT massively outperformed their data team on classifying messy inputs. That’s when they knew this wasn’t a “tool” — it was a new teammate.
Instead of rolling out AI from the top, they created an open system where anyone could suggest use cases. They celebrated early wins and surfaced blockers fast. Within months, AI went from experiment to default — with over 90% of code in one project now generated by AI (and reviewed by humans).
But adoption wasn’t automatic. Engineers needed to shift their mindset from “controlling outcomes” to “guiding behavior.” That means learning how to prompt, review, and work with AI like you’d work with a junior team member — not a compiler.
One surprising insight? As AI tools improve, teams are less amazed by what they can do and more surprised by what they can’t. It’s a good place to be. It means the bar is rising.
6. What Product Managers Need to Know About AI
Marily Nika, Gen AI Product Lead at Google put it simply: if you’re in product, you’re in AI — whether you like it or not. She broke it down into three kinds of PMs:
1️⃣ Those who use AI to work smarter
2️⃣ Those who build AI features into products
3️⃣ And those working deep in the stack with researchers and models
Her advice? Stay rooted in the fundamentals. Build around real problems, not just shiny models. And don’t obsess over roadmaps that stretch six months out. AI changes fast. The better approach is to experiment quickly, test hypotheses, and ship what works.
She also called out one big trap: thinking AI is the product. It’s not. The user experience is the product. If your voice assistant is frustrating to use, no one cares how advanced your model is under the hood.
PMs who succeed in this new era will be the ones who can connect tech and human needs. Not just build tools, but build experiences powered by AI.
7. Startups don’t need AI teams — they need better automation.
Simeon Penev from Verto Digital showed how even solo builders or semi-technical founders can stitch together powerful AI workflows using tools like n8n. His case? Automating the mess that is customer feedback.
Instead of manually reviewing hundreds of form submissions, support messages, and UX notes, he built a system that:
- Collects feedback across platforms
- De-duplicates tickets using AI
- Categorizes issues
- Summarizes them
- Creates structured Jira tickets
- And checks past tickets to avoid spammy duplicates
This wasn’t built by a big team. It was just Simeon, some solid thinking, and a few well-chosen tools. The big insight? You don’t need to build an AI model to benefit from one. You just need a an objective and the patience to link the steps.
If you’re drowning in busywork, don’t wait for a “head of AI” hire. Just map out your biggest time sink and ask: can this be automated?
8. You don’t need to rebuild your product. Just make it smarter.
Vesko Kolev from icanpreneur told the story of how they transitioned from a regular startup to an AI-first one, without tossing everything and starting over. When ChatGPT hit, customer expectations shifted overnight. Everyone wanted “smart” features, fast.
Their first step? A simple AI-powered email generator. Small, useful, and fast to ship. That early win built trust, both inside the team and with users. From there, they built IVA, an in-product assistant that helps founders define customer segments, navigate onboarding, and get personalized help.
Vesko emphasized that building great AI features takes more than engineering — it needs firsthand product insight. That’s why founders and PMs need to be in the tools, experimenting, testing, and learning what works. You can’t spec good AI from a distance, you have to live with it, tweak it, and shape it as it grows.
The bottom line: you don’t have to start from scratch. Just start where AI can remove friction. Ship something that makes life easier for your users and keep going from there.
Final Thoughts
PMF.AI reminded us that building with AI isn’t about jumping on a trend — it’s about solving old problems in new ways. The people we heard from were figuring things out in real time, learning from failures, and staying grounded in what users actually need.
The best thing you can do as a founder? Pick one part of your product or operations stack that feels slow, repetitive, or frustrating. Ask yourself: could this be smarter? Then prototype. Ship. Learn.
You don’t need perfect. You need useful. And you probably don’t need a bigger team — just better tools, clearer goals, and a willingness to experiment.
You can find all recordings from the PMF.AI event in the videos section of our blog.
And a sneak peak of the event itself right here ⬇️
