In this week’s episode of Founder Fundamentals, , founder and product coach at , joined the community to share practical insights on when founders should and should not build with AI. Drawing from his experience across software development, product management, startup coaching, and AI-powered product development, Moshe unpacked the difference between traditional systems and AI, the key questions founders should ask before choosing AI, and how entrepreneurs can better assess cost, risk, and real product value before building.
Not Every Problem Needs AI
One of the strongest takeaways from the session was that founders should not default to AI simply because it is trending. Moshe encouraged participants to move away from the pressure of adding AI for the sake of it and instead focus on whether it actually fits the problem being solved.
As he put it, “people going directly into that direction and it’s not always the right decision.” That idea shaped the session from the start. Rather than framing AI as something every startup must adopt, Moshe pushed founders to ask more practical questions about when it is useful, when it is unnecessary, and what trade-offs come with using it.
He explained that traditional software is deterministic, meaning it follows fixed rules and predictable logic. AI, by contrast, is probabilistic, which means it works through patterns, models, and likelihood rather than certainty. That distinction matters because some problems simply do not need AI at all.
“If it’s deterministic, there is no reason to build it with AI,” Moshe said, noting that it is often “way more expensive and it’s less accurate” in those cases. For founders, that means the real challenge is not figuring out how to add AI, but figuring out whether AI belongs there in the first place.
Understanding How AI Learns
To make this clearer, Moshe walked through the difference between traditional engineering and AI systems. Traditional software, he explained, is built on logic that maps inputs to outputs in a predictable way. AI systems behave differently. They rely on models trained on data and return results based on probability.
“The main thing about a traditional one is that it is deterministic,” he explained, while AI is “probabilistic… it’s always about probabilities.”
This distinction helped founders understand why AI can feel powerful but also unpredictable. A model can generate useful output, but it can also hallucinate, misclassify, or behave in ways that are hard to fully control. That uncertainty is built into the system itself, and founders need to understand that before using it in products where accuracy and reliability matter.
Moshe also reflected on his own experience moving from traditional product management into AI products. “It actually drove me crazy,” he admitted. “I didn’t really understand that perspective that it’s only an experiment and I’m not sure what the result is going to be.” It was an honest reminder that working with AI often requires a mindset shift, especially for teams used to building systems with predictable outputs.
AI Is Not One Thing
Another major focus of the session was helping founders understand that AI is not one single tool or method. Moshe broke down the main types of AI learning, including supervised learning, unsupervised learning, semi-supervised learning, reinforcement learning, and deep learning.
This section gave participants a broader understanding of what it actually means to build with AI. Different models require different levels of labelled data, different computational resources, and different forms of oversight. Some approaches are more accurate but expensive to train. Others are more scalable but harder to validate.
Moshe emphasized that each option comes with trade-offs, and founders need to think carefully before assuming AI is a simple plug-in solution. “Each situation will be a different fit,” he said, adding that every founder needs to understand “what does it mean and there is pros and cons and cost associated with each one of those.”
That grounded the conversation in reality. Using AI is not just about capability. It is also about resources, time, infrastructure, and whether a team can realistically support the system they want to build.
How Much Should AI Be Allowed to Decide?
The session also explored a question that becomes increasingly important as products become more automated: how much decision-making should AI actually handle?
Moshe introduced a helpful framework built around prediction, judgment, and action. In some cases, AI may only support a decision by generating predictions. In others, it may also influence judgment by ranking options or narrowing choices. In more advanced cases, it may go all the way and take action on behalf of the user.
He used a recruitment example to show how this plays out in practice. AI could help a job seeker identify the best-fit roles, or it could help a recruiter filter hundreds of applicants down to a top shortlist. At its most automated level, AI could even schedule interviews without human review.
This led into an important conversation around human oversight. When asked whether there should always be a human in the loop, Moshe made it clear that the answer depends on context. “It all depends,” he said. “It depends if you need human in the loop or not.”
That answer became even more useful when participants pointed out that risk matters. A low-stakes action may not require human review, but a high-stakes action probably does. The session reinforced that founders need to think not just about what AI can do, but what it should be trusted to do.
Building With AI Means Thinking About Viability
One of the most practical parts of the workshop focused on what it actually takes to build an AI-powered product. Moshe introduced a canvas designed to help founders evaluate whether an AI use case is viable by thinking through value, risk, inputs, training costs, feedback loops, and outcomes.
The framework moved the conversation beyond hype and into execution. Founders were encouraged to think about what success would look like, what failure would cost, what data or infrastructure would be needed, and whether the organizational value justified the investment.
He stressed that it is not enough to build something impressive. Founders need to understand what is required to support it. “There is a bit of work that needs to be done to assess the cost that it might take you to build that,” he said.
He also reminded participants that many teams forget to define success early enough. “Always remember that you want to define what the success looks like,” Moshe said. “How do you measure it?” That point extended well beyond AI. It was a strong product lesson in general: if a team cannot define success or track it clearly, it becomes very difficult to know whether the product is actually working.
Buy, Build, or Adapt?
Another valuable part of the discussion focused on whether founders should build their own AI systems or use existing models and APIs. Moshe encouraged early-stage founders to avoid overbuilding, especially before product-market fit.
“If you’re early stage, pre-product market fit, probably you want to try to use existing models,” he said. Today’s landscape offers many more options than even a few years ago, including hosted APIs, open-source models, and tools that allow for fine-tuning or retrieval-based systems.
At the same time, Moshe acknowledged that there are cases where building or hosting custom models makes sense. Privacy requirements, latency needs, or high usage costs can all shift the equation. This made the conversation especially useful for founders trying to make practical decisions rather than theoretical ones.The broader point was that AI infrastructure choices should come from product and business needs, not from pressure to look more advanced than the company really is.
Not All AI Companies Are the Same
Toward the end of the session, Moshe introduced a useful way to think about different kinds of AI companies. He broke them into three types: AI-native, AI-embedded, and vertical AI.
AI-native companies are built entirely around AI from the start. AI-embedded companies use AI as a meaningful enhancement within a broader product. Vertical AI companies apply AI to specific industries or use cases where domain expertise and specialized data matter.
This helped founders better understand their own positioning. Not every company needs to be AI-native. In many cases, a better approach is using AI in a focused, meaningful way that strengthens an existing product without making it the whole story.
Moshe cautioned founders against adding AI superficially. In AI-embedded products especially, the real question is whether it creates value or is just “a gimmick.” That distinction is what determines whether users actually care.
About Founder Fundamentals
is a 12-week workshop series hosted by and and powered by designed to equip you with essential entrepreneurial skills. Attend 9+ workshops to earn a Certificate of Completion and take the first step toward entrepreneurial success!
About the Speakers
is a product management leader with a background in enterprise B2B software engineering, he brings a lean, user-centered approach to building impactful products. He is the founder and product coach at and a Product Director at Rootquotient, where he also mentors founders through leading accelerators. He teaches Information Systems and Design Thinking at the Schulich School of Business, co-hosts two product-focused podcasts, and is the author of the novel The Resurrector.