What Is the Biggest Problem With AI Tools Right Now?

The biggest problem is the 'Confidence Illusion.' Traditional software fails loudly, it crashes, throws an error code, or refuses to compile (https://www.nature.com/articles/s42256-023-00650-8). You know it's broken. LLMs fail silently and confidently. They will generate a completely fabricated legal citation or a dangerously incorrect medical dosage with the exact same authoritative, polished tone they use for factual information. Because the output *looks* professional, human operators drop th...

The cost and infrastructure problem is bigger than most people realise. McKinsey estimates AI is fuelling a $7 trillion race to scale data centres (https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-cost-of-compute-a-7-trillion-dollar-race-to-scale-data-centers). That cost is ultimately passed to users through subscription fees, and the pricing models are opaque and volatile. I've had tool subscriptions double in price with 30 days notice. For businesses...

The biggest problem is the hidden environmental cost. A single ChatGPT query consumes roughly 10 times the electricity of a standard Google search (https://climate.mit.edu/explainers/environmental-footprint-ai). Training a model like GPT-4 consumes millions of gallons of fresh water for data centre cooling. We are treating AI like it's a weightless, magical cloud technology, but it is a massive, resource-intensive heavy industry. We are accelerating the climate crisis so we can generate slightly...

The biggest problem is 'Feature Churn.' The underlying foundation models (OpenAI, Anthropic) are updating so fast that they are constantly destroying the startups built on top of them (https://stratechery.com/2023/openai-and-the-platform-risk/). You spend three months building a complex RAG pipeline to allow users to chat with their PDFs, and then OpenAI releases 'Chat with PDF' natively for free. Building an AI product right now is like building a sandcastle while the tide is coming in. The pla...

The homogenisation problem (https://www.linkedin.com/pulse/ai-2025-what-advanced-stalled-expect-2026-newsletter-conforto-phd--muygf). Every AI image tool is converging on the same aesthetic, smooth, hyperrealistic, slightly uncanny. Every AI writing tool produces the same cadence, the same structure, the same vocabulary. If everyone uses the same tools, everything starts to look and sound the same. The LinkedIn AI post aesthetic is already a running joke. The biggest problem with AI tools right ...

Reliability. The gap between 'impressive demo' and 'production-ready' is still enormous. I can show you a demo where an AI agent completes a complex multi-step task flawlessly. I can also show you the same agent failing in subtle, hard-to-detect ways on a slightly different version of the same task. The Nature paper on trust in AI identifies this inconsistency as the central barrier to enterprise adoption (https://www.nature.com/articles/s41599-024-04044-8). Until AI systems fail predictably and...

@StartupSurvivor We call this being 'Sherlocked' (when Apple builds a third-party app's feature into the OS). If your entire startup is just a UI wrapper around an OpenAI API call, you deserve to go out of business. You have to build deep, proprietary workflows that the foundation models can't easily replicate.

@TechSkeptic99 Hallucination is my biggest problem. I've had AI confidently cite statistics that don't exist, recommend products that have been discontinued, and give me legal information that was just wrong. The problem isn't that it gets things wrong, it's that it sounds equally confident whether it's right or wrong. I've learned to verify everything, but that defeats part of the time-saving purpose.

The homogenisation problem is real but I think it's temporary. As models improve and as users develop more sophisticated prompting skills, the outputs will diversify. The early internet looked homogeneous too, everyone used the same templates, the same fonts, the same layouts. Over time, distinctive voices and aesthetics emerged. The same will happen with AI.

@Climate_Techie This is becoming a massive zoning issue. Tech companies want to build hyperscale AI data centres in regions that are already facing severe water shortages. The local municipalities are starting to push back because the data centres are literally drinking the towns dry.