“AI is obviously a very interesting technology, there’s a lot of hype around it, but don’t get lost building something in search of a problem” – Ivan Zhang, co-founder of Cohere. 

If I had a dollar for every startup at Web Summit that started their pitch with “We’re revolutionising [industry] with AI,” I could have started my own VC fund. The widespread accessibility of integrating AI has created a cohort of early-stage startups that are using AI as a default feature, not because it resolves major pain points, but because it’s expected in today’s world.

With APIs like GPT-4.1 nano costing as little as $0.10 per million input tokens, experimenting with and incorporating LLM outputs into a product has never been easier. Founders are creating the illusion of novelty and intelligence by building wrappers or chatbots that provide little value-add when compared to directly using generative AI tools. But this isn’t real innovation. In recent years, the market has been oversaturated with AI automation tools that complete menial tasks such as answering FAQs, extracting data, and sorting documents. This surge of low-impact applications conceals the fundamentally transformative potential that AI holds.

The term “AI” is often mistaken for stand-alone Natural Language Processing applications like ChatGPT, Gemini, or Claude. In reality, it’s the architecture underneath it, such as artificial neural networks that ingest data, recognise patterns, and learn relationships in a way that mimics human reasoning. For example, GPT-3.5 was trained on 570GB of data, creating a general-purpose platform that makes it versatile but not precise. At Web Summit, many of the booths I stopped by were simply plugging into this general model. The next frontier won’t be about building on top of universal algorithms, but collecting vertical-specific data and using it to train specialised systems or applying these algorithms in niche industries. This data-driven approach acts as a point of differentiation and a moat. Innovation will come from people who understand this architecture and use it as a starting point. 

Two of the finalists at the Web Summit Pitch Competition exemplified this approach. VodaSafe’s AquaEye uses a combination of sonar and its proprietary AI model to locate drowning victims underwater. It’s a life-saving solution trained on niche data that solves a critical problem. Glüxkind’s smart stroller, built on the NVIDIA Jetson Platform, uses embedded AI that incorporates domain-specific, real-time data from sensors to detect obstacles and assist parents caring for their children. These companies show that what piques investor interest is data collection, embedded intelligence, and solving real problems. 

While AquaEye and Glüxkind perfectly embody the phrase “AI is the new electricity," the messaging is often misapplied by the greater tech community. When capital overwhelmingly favours AI solutions (57.9% of global VC dollars in Q1 went to AI and machine learning startups), founders start to reverse-engineer. Instead of starting with a validated problem and asking whether AI is the right solution, they’re looking at the capabilities of "plug and play" APIs and retrofitting a market around it. This causes products to exist because AI exists, not because there’s a real pain point demanding a fix.

The rush to treat AI as a default feature is diluting real innovation. AI is undoubtedly a powerful tool, and the mindshare is deserved. But it should remain a tool, not a stand-alone solution.

Not every problem should be solved by AI, and when it does, the impact comes from applying it in a specialised manner. Until founders build products beyond superficial AI offerings and start treating it as more than just a buzzword, we risk the market being flooded by applications that distract from its potential. 

Mayako Kruger is a cognitive systems student at UBC.