Ferenc @ferenc

AI is making it easier to build an app. It's not making it easier to build a business. I sat in on a call this week with a founder who showed me a working product built over a weekend. Clean interface, real features, no users in the three weeks since it launched. The build was never the bottleneck. Finding someone who actually pays, figuring out the price, and checking whether the number is enough to run a company on - that's where most of this falls apart.

It took me a few years in SaaS consulting to learn this: "Build for outcomes, not tools" is the right destination. But: Even obsessing over business outcomes gets you nowhere if you can't name the one metric your product is uniquely positioned to move - that nobody else is measuring yet. There's something specific in that gap. It's where positioning becomes a moat. Simply because most founders never stop long enough to find it. Building for outcomes without that specificity is consulting with worse margins.

The longer I work with AI-powered SaaS founders, the more I realize: Your pricing model isn't fully yours. It's downstream of infrastructure decisions made by companies you've never sat across from. For example: OpenAI's "Jalapeño" chip - built with Broadcom, deploying before the end of 2026, reportedly 50% cheaper than GPU inference - just rewrote the cost assumptions underneath every AI product priced in the last 18 months. Most founders modeled their margins on today's inference costs. Those costs are now someone else's competitive lever.

Ideas are officially worthless. Anyone can ship an app in a weekend now. The market is flooding with AI-built apps that look like products but aren't businesses. Building was never the moat. It was just expensive enough to look like one. The real filter: Not "do you like it.

Survivorship bias threatens your SaaS strategy. Saw two viral Reddit threads - one celebrating vibe-coded SaaS to $10k MRR, another breaking down 7 months of the exact same approach with "what nobody tells you." Both experiences are real. Neither represents the full distribution - only the founders still posting. The winners kept going and kept sharing. Everyone else quietly stopped. Founders aren't benchmarking against everyone who tried. They're benchmarking against everyone who survived.

AI made writing free. Now nobody can afford to read. We built an internal knowledge base with AI. Standards, decisions, best practices. Impressive volume in weeks. Then it stalled. "Review 10 of these by next week" killed it. Nobody came prepared. The meeting collapsed. The content wasn't the problem. The process was. AI removed the old bottleneck: creation. It created a new one: human review. If you don't redesign your workflow around review capacity, AI output isn't a knowledge base. It's a landfill with a search bar.

The niche that sounds too specific to matter is where the best exits happen. A company just sold for nearly a billion dollars disrupting 911 call centers. Nobody was building for that in 2021. Horizontal AI tools are racing for the same $5B TAMs. Vertical builders are quietly owning $200M markets with no competition and enterprise contracts that never churn. The moat isn't technology. It's the fact that nobody else bothered.

"We're splitting the budget and moving half to next year." Every agency founder gets this message. Most misread it as good news. It's not. A split budget is a downgrade dressed as a plan. Budgets don't restructure themselves. Someone made a decision, and the decision was: this project matters less than it did. 3 things I do when this signal arrives: 1. Ask what changed in their business, not in the budget. The budget is the symptom. Priority moves happen first, budget moves follow. If you don't know the reason, you can't fix it. 2. Treat the confirmed half as the whole project. Scope something complete within it. A half-finished deliverable makes the second half easy to cancel. A finished one makes it easy to fund. 3. Update your own forecast immediately. Money that moves once moves twice. The vendors who get burned are the ones who book deferred budget as future revenue. Deferred budget is not revenue. It's an option the client holds. Your job is to make exercising it obvious.

Crazy how often founders tell me their AI agent is live and assume that means it's done. I find this costly. First, live and reliable are not the same milestone. LangChain's State of Agent Engineering 2026: 57% of organizations have agents in production. 1 in 3 say quality - not cost - is their biggest blocker. Second, it misses what enterprise clients actually audit: not whether the agent runs, but whether there are approval steps, an audit trail, and a rollback for when it acts on bad data. Shipping is the easy part. Trust is the product.

If a client's exit surprises you 2 weeks out, your contract is broken. Not your relationship. Our rule for retainers: 60 days out: you should sense if it continues. 30 days out: it must be in writing. Don't rely on people asking brave questions. Hope is a comfortable default. Put the question in the contract. Then nobody has to be brave.

The first enterprise client who sends a security questionnaire will ask about audit trails. Founders usually find this out in the meeting, not before it. An audit trail in a B2B SaaS is a specific answer to a specific question: When an action happened in this product, who triggered it, what data was touched, what changed, and can it be reversed? For products where AI takes actions (filing a document, issuing a refund, routing a lead, updating a record), this matters more. The agent acts faster than a human would. That speed is the feature. But speed without traceability is what makes procurement teams pause. What a functional audit trail covers, at minimum: → User or agent that triggered the action → Timestamp and session context → What data was read and what was written → Whether the action is reversible and by whom → What the end user sees vs. what the admin retains I've seen well-built products lose enterprise deals at the security review stage because the audit trail was an afterthought. The product worked. The record of what it did didn't exist. Build the trail before you need to show it.

Our first profitable month this year had nothing to do with sales. We cut fixed costs. Revenue fixes take quarters. Cost fixes take weeks. Most founders attack the slow lever first because it feels like growth. Survive on the fast lever. Grow on the slow one.

The standard SaaS idea advice: find an underserved market. The approach that helped one founder hit $9k/month faster than expected: find where money is already flowing toward a broken solution. There's a real difference between the two. Underserved markets are often underserved because the problem isn't acute enough or the buyer can't pay. Bad solutions are different. The customer is already spending. The problem is already validated. You're not convincing anyone the problem exists. The signal worth looking for: complaints that repeat across multiple platforms from people who are already paying for something that doesn't quite work. I've watched founders spend a year doing customer discovery trying to confirm a problem exists. The ones who moved faster started by looking for money already moving in the wrong direction. That's your replacement opportunity.

I keep hearing the same question from early-stage founders right now. "If someone can clone my product in a weekend, what's protecting me?" It's the right question. The answer just isn't about the code. A weekend clone can match your feature list. What it can't replicate: • six months of customer feedback that shaped why those features work the way they do • the specific edge cases you've already burned through and fixed The moat moved. Before AI coding tools, the cost of building was the barrier. That's gone. What replaced it: depth of understanding. How well do you know the workflow your customers actually use this product for, and can you prove it? In my experience, the founders who lose ground to fast clones were building broad with no real insight advantage. The ones who hold position have a specific vertical and real workflow knowledge no weekend project can replicate. The fast clone is a threat to the generic solution. It's not a threat to the founder who knows the customer better.

Ben Cera raised $30M at a $250M valuation for a company with no employees. Polsia runs on 9 AI agents, serves 7,600 business customers, and the fundraise happened live on Twitter with a public dashboard. The obvious take: AI just replaced the entire team. The less obvious take: Polsia works because it sells software that runs business workflows, not software that builds reliable products for other businesses. That distinction matters because a workflow automation tool can scale without headcount when edge cases are bounded and failure modes are recoverable. The agent handles the routine. Rare exceptions escalate or fail gracefully. A SaaS product serving enterprise clients is a different problem. The edge cases aren't bounded. A failure at the wrong moment isn't recoverable by restarting the session. Someone has to know what they're looking at. I've worked with founders who ran too lean too early on the wrong type of product. The gap between "it runs" and "it's reliable under real conditions" doesn't close automatically. Polsia is a real business. The question is whether your product has the same shape.

GitHub moved every Copilot plan off per-seat pricing on June 1. You now pay for what the AI burns, not the seat it occupies. Microsoft, Amazon, and Google all killed flat-rate pricing for their main AI tools this year. Three hyperscalers, one year. That's not a coincidence. It's a pricing model dying in real time. The question this poses: what are you actually pricing right now? One user runs your AI feature twice a month. Another runs it 400 times a week. The seat fee treats them the same. The heavy user looks unprofitable on your P&L. The light user churns because they never got enough value to justify the cost. If your SaaS has an AI layer, that same math is running in your product right now.

VivaTech ran through June 20 in Paris, and the headline this year was AWS making European sovereign cloud its centrepiece. When the world's largest cloud provider puts data residency front and centre at Europe's biggest tech event, the compliance conversation that's been killing B2B SaaS deals in finance, health, and public sector starts to change shape. What follows for founders building in Europe right now: → Enterprise buyers in regulated industries have an infrastructure narrative they can take to legal → "Where does our data live?" becomes a shorter, less load-bearing question in procurement → Deals that stalled at the compliance stage in 2024 and 2025 become viable conversations again I've watched European founders lose deals with no product problem. The block was always the residency question. This week, that question got a louder answer.

You wouldn't sign a lease where the landlord also bought your tools, your compiler, and your testing framework last quarter. Yet that's what building on OpenAI is in 2026. I mapped the acquisitions out: 8 in the first half of the year. Tooling, security testing, deployment. So, the entire stack. You're not building on top of the platform. You're building inside it. And the difference only becomes clear the day you need to leave.

Lovable hit $200M ARR with 146 employees. $1.37M revenue per person. At a typical SaaS dev agency, $150–200K per head, that same ARR needs around 1,000 people. A client asked me last month whether to go with Lovable or hire a dev team. The honest answer starts with understanding what Lovable can't do. Dev agencies that can't answer that question have already lost the conversation.

Spoke to a founder last month who spent four months building and got a wrong product at the end of it. Expected to hear they'd skipped the research. They hadn't. Customer calls, feedback loops, X monitoring, weekly shipping. "We had more data than we knew what to do with." That's it. What founders think they heard and what customers meant are two different conversations.