One of the perks of being in the thoughtbot universe is that we meet a ton of people building cool things, and many of them are startup founders, mentors or advisors. Lately, we’ve chatted with them about how they’re using AI tools for startups, both where AI adds value and where they’re seeing risks.

AI tools have made it possible for non-technical founders to ship a working prototype in a weekend, but after talking with people who are building with these tools every day, a more nuanced picture emerges. AI is changing what’s hard about building a startup, but it’s not eliminating the hard parts all together. Here’s what we’ve been hearing lately.

Most founders already have access to AI coding tools like Claude, Lovable, Bubble.io or similar platforms, and they’re using them. One early-stage leader created a working app in a week with a technical collaborator and Claude Code; then brought the output into Figma to refine fit and finish. Others are assembling their own AI-native stacks using combinations of Claude, GitHub, Vercel, APIs, Reddit workflows, and custom tooling.

But nearly everyone we’ve spoken with separates “AI helped me build” from “AI helped me decide what to build.” The entrepreneurs doing the best work still do much of the unglamorous parts by hand: defining product logic, writing tickets and user stories, and conducting user research and competitive mapping. AI accelerates execution once the big-picture strategy is done, but it doesn’t create the roadmap for most startup execs.

Know when to vibe code and when to call in developers

As one business owner told us, “I don’t trust myself to know if the AI built it correctly.” He’s impressed by the capability and speed of AI tools, but he’s not technical enough to comb through the code produced to validate overall quality and spot any potential risks.

His solution? He turns to AI for prototyping low-stakes tools and experimenting with them reactively. But he brings in experienced engineers for apps involving scale, infrastructure complexity, sensitive data, and business-critical reliability.

Other non-technical founders told us about hitting ceilings when using no-code AI platforms and not knowing how to move forward without technical help. They’ve faced broken logins, challenges ensuring data security, limited database options with some no-code app builders, and sometimes overwhelming complexity.

It often boils down to building too much, too fast, on the wrong platform. Not everyone has the knowledge and experience to scope down their vision for a first version, pick the right tools, and arrive at a shippable app.

AI won’t tell you what’s wrong with your pitch or your product

We’ve all experienced AI glazing, and several people told us that AI being agreeable posed a limitation for them. One founder tried using an AI chatbot to prepare for an investor pitch and found it disappointing as the AI viewed everything through “rose-tinted glasses.” Often, the need isn’t validation as much as push back to surface counterarguments and pressure-test the underlying thinking before walking into a pitch meeting.

That agreeability shows up elsewhere, too. AI tools are good at generating options but don’t naturally walk you through tradeoffs or hold a stable plan across multiple sessions. One entrepreneur told us that real-time AI suggestions kept “diverting and changing the flow of everything.”

Human taste and strategy still drive successful startups

Where AI did consistently help was as an input to human judgment rather than a replacement for it. Founders fed user interviews into AI tools to help surface notes and trends. One used AI as a brainstorming partner while iterating on a fundraising deck, going back and forth on framing. But in every case, folks described AI as a sounding board versus a decision-maker.

As AI makes the mechanics of building cheaper and faster, the things that don’t get automated, such as prioritization, customer insight, design taste, and strategic judgment, become more valuable. One startup mentor noted that founders who lean too heavily on AI to do their thinking for them can see their own reasoning and execution decline over time. Those who fare best treat AI as leverage for the work they’ve already thought through themselves.

Where does all this leave founders?

AI is a genuine accelerator for execution: code, research, drafting, and exploration all move faster with it. But for now it doesn’t replace the work of defining the right product, validating tradeoffs, or owning the decisions that determine whether a startup succeeds. If anything, those skills matter more than ever.

Many startup leaders expressed the need for AI tools with more guardrails. Something to take you through a proven process for defining the right MVP, understanding customers and making tradeoffs without making the crucial decisions for you.

If you’re navigating that gap between fast prototyping and building a product that’s ready to scale, we’d love to help. Get in touch about one of our Shaping Sprints.