Building software has become dramatically easier. Today, a founder with a laptop and AI tools can build a working prototype in a weekend. You can set up backend logic, design a user interface, create a landing page, assemble marketing materials, and automate early outreach. All without a large team, enormous resources, or extended timelines.

This shift introduces a new, often-overlooked challenge: while building software is faster, achieving product-market fit remains as difficult as before.

In fact, it may even be more difficult now. With rapid product launches, more tools compete for attention, resulting in many impressive products that fail to address real user needs.

The primary challenge has shifted from building products to deeply understanding problems in order to develop effective solutions. Achieving this requires engaging with customers, identifying their frustrations, building trust, and observing details that reveal genuine opportunities.

AI can assist with coding, information analysis, and increasing efficiency. However, in my opinion, it cannot replace the genuine conversations, observations, and decisions necessary for achieving product-market fit.

This is ultimately positive. When powerful tools are widely available, founders who invest in understanding people will distinguish themselves.

Activity isn’t traction (and speed is a poor measure of progress)

Earlier in my career, I fell into a trap that gets most founders: I confused shipping fast with making actual progress.

When you’re building at lightspeed, it feels incredible. You’re committing code daily, pushing new features every few days, and watching the product transform right in front of you. You feel like you’re winning because you’re busy. But I’ve learned the hard way that speed without clear directional accuracy is just high-velocity noise. Moving fast doesn’t matter if you’re driving in the wrong direction.

It’s the exact same trap happening in enterprise AI right now. When you look at MIT NANDA’s The GenAI Divide study—which analysed real-world AI implementations—95% of AI pilots failed to move the needle on the bottom line (MIT NANDA, 2025). The problem wasn’t that the AI models were dumb; it was that the teams behind them rushed to deploy tech without understanding the messy, unpredictable human workflows and incentives they were stepping into.

(State of AI, 2025) and BCG (2025) have published almost identical numbers: tons of companies are experimenting with AI, but only a tiny fraction are seeing any real, scalable business impact.

The lesson I had to learn the hard way is simple: output is not progress.

You can make your app look sleeker every single week and still be solving a problem nobody cares about. When your core assumptions are wrong, no amount of AI-generated code or quick patches will fix the foundation. All AI really does here is act as an accelerator—and if you don’t deeply understand your customer’s pain, accelerating development just means you’re making the wrong choices faster.

You can’t prompt-engineer customer discovery (or decode real human intentions)

Whenever I talk to other founders struggling with customer discovery, I observe the desire to bypass the difficult and uncomfortable part: studying human needs. Reaching out to strangers is not only hard and awkward, but also time-consuming and costly. So they ask an LLM to “simulate customer interviews” or generate fake buyer personas.

I strongly advise against this. It creates a dangerous echo chamber where you’re just interviewing an algorithm’s guess about reality. Because AI models operate on statistical probabilities trained on past web data, a synthetic persona will always give you the most generic, average answer imaginable. But in my experience, product-market fit is never found in the statistical average. It lives in the weird edge cases, the strange manual workarounds, and the non-intuitive things real humans do when they think nobody is watching.

More importantly, AI cannot decode a user’s true intentions. An algorithm can read what someone types, but it has no idea why they said it. Figuring out real human intent requires a deliberate UX process run by a human who knows how to listen—not an automated script.

This is why AI completely falls apart on Rob Fitzpatrick’s The Mom Test (Fitzpatrick, 2013):

• It misses hidden motives: People lie to founders out of politeness. They’ll smile, say your idea “sounds super useful,” and then never log in again. An AI logs “I’d love a feature like that!” as glowing validation. Anyone who has run real user interviews knows that’s just a polite way to say no.
• It misses micro-behaviours: An AI can’t catch body language—the long, uncomfortable pause when you mention pricing, or the tiny wince when someone struggles to find a button on your screen.
• It misses the real root problem: When a user tells you, “We just use a Google Sheet for that,” an AI writes down “Google Sheets” as a competitor. As a human researcher, I know to dig into the actual emotion behind that: “How late do you have to stay at the office on Friday nights just to update that spreadsheet?”

AI is great for helping me speed up side tasks—like organising competitor feature matrices or cleaning up call transcripts—but it will never run the UX process needed to uncover genuine human intent and the problems actually worth solving.

Trust is still a human-to-human sport

The McKinsey Global Institute (2024) recently found that as AI automates routine tasks such as coding, data gathering, and content drafting, the most valuable skills are empathy, sound judgment, and relationship-building.

Consider your first 10, 50, or 100 customers. They are unlikely to purchase your product because it is flawless or has seamless onboarding. Early-stage software typically has imperfections.

Customers buy because they trust you. They believe you understand their challenges, will respond promptly to issues, and are committed to developing solutions that meet their needs. Automated emails or AI avatars cannot build this level of trust; it requires genuine human interaction.

Excessive reliance on AI for early customer interactions can create issues that are difficult to resolve later.

A 2026 study by Belanche, Casaló, and Flavián, published in SAGE, looked at how people respond when automated systems fail compared with when humans make similar mistakes. Their findings highlight an important difference: people are usually more forgiving when a person makes an error because they assume there is intent, effort, and accountability behind it.

AI does not receive the same benefit of the doubt.

When an AI system provides an incorrect answer, fabricates information, or misunderstands a customer’s situation, people respond differently. The primary issue is not the mistake itself, but the resulting loss of trust.

For early-stage companies, trust is often the primary reason customers try an unfamiliar product. A single negative experience with an automated system can undermine a relationship that took months to establish.

PMF lives in the noise, not the trendlines

AI models are designed to eliminate noise and focus on average trends. However, product-market fit is often concealed within the noise.

Slack did not originate from a market survey on enterprise communication tools. It began as an internal chat application developed by a team attempting to salvage a struggling online video game.

Instagram (originally Burbn) began as a feature-heavy check-in application. Human intuition revealed that users engaged primarily with the photo posting and filter features.

An AI analysing Burbn’s usage metrics might have recommended enhancing the location check-in feature to align with market trends. However, it required human insight to recognise users’ preference for photo sharing, leading to a significant pivot.

The wrong lesson about AI

I’ve noticed an interesting trend in Poland recently. For the first time in years, universities reported a massive drop in applications for traditional computer science (IT studies) degrees—down by nearly 10,000 candidates in a single recruitment cycle. One of the main reasons being tossed around in media discussions is the belief that AI will soon replace software engineers, making tech careers obsolete.

I think this misses the bigger picture entirely.

AI is fundamentally changing software development, but it isn’t eliminating the need for people who can understand complex problems, navigate trade-offs, and build products that customers actually care about. If anything, as the technical barrier to writing code drops to zero, the value of product thinking, domain expertise, and deep customer discovery goes through the roof.

Learning to code is still valuable. But learning to understand human beings, ask better questions, and make sound product decisions? That is becoming the ultimate competitive advantage.

How I use AI when searching for product-market fit

AI is an excellent research assistant, but a poor substitute for customer discovery.

Personally, I use it to organise information, summarise interviews, compare competitors, and challenge my thinking. But I don’t expect it to tell me what users really need or which problems are worth solving. Because those insights still come from conversations, asking the right questions, observing, and being curious.

If AI gives me back a few hours every week, I don’t use them to ship more features. I use them to spend more time with customers. That’s where product-market fit is found.

Because every meaningful product decision I’ve been involved in has started with a conversation, not a prompt.

Final thoughts

AI has simplified code generation and content creation, making them less of a long-term competitive advantage. When features can be replicated quickly, software alone does not define your business’s value.

Your true advantage lies in understanding your users’ genuine intentions. Allow AI to manage repetitive, high-volume tasks, but prioritise engaging directly with real people to gain meaningful insights.

References

• Belanche, D., Casaló, L. V., & Flavián, C., 2026, When AI Fails: The Impacts of Hallucinations and Errors on Customer Trust and Brand Advocacy. SAGE Journals.
• Boston Consulting Group (BCG), 2025, Where’s the Value in AI? Global Enterprise AI Adoption Benchmark. BCG Report.
• Fitzpatrick, R., 2013, The Mom Test: How to talk to customers & learn if your business is a good idea when everyone is lying to you. Robfitz Ltd.
• McKinsey & Company, 2025, The State of AI in 2025: Generation AI Comes of Age. McKinsey Global Survey.
• McKinsey Global Institute, 2024, Human skills will matter more than ever in the age of AI. McKinsey & Company Insights.
• MIT NANDA, 2025, The GenAI Divide: State of AI in Business 2025. MIT Initiative on the Digital Economy / NANDA Research Report.