I Analyzed Dozens of AI Projects. Here Are the 3 Problems They All Solve.

The revolution isn’t what LLMs can do, but the dramatic collapse of the barrier to entry

Image Credit: Event Horizon Telescope Collaboration (2019) The first image of a black hole, showing the supermassive black hole at the center of galaxy M87. Retrieved from https://eventhorizontelescope.org/

Times are changing. Just as the internet revolutionized our world in ways we couldn’t fully grasp at the time, Large Language Models (LLMs) are creating a similar seismic shift. They represent truly life-changing technology, but exactly what those changes are is hard to identify.

This past week gave me a glimmer of insight, as I had the opportunity to review dozens of LLM-powered projects during our company “vibeathon” (formerly known as a hackathon). What made this exercise interesting is that there were no constraints on the projects people could create — only that they would be assisted heavily by LLMs. What I saw was a wide variety of projects from a diverse group of professionals; in essence, a microcosm of the problems that industry will solve using LLMs. By studying these projects, clear patterns emerged as to exactly what problems LLMs can currently solve.

To start, consider the following three examples. These are more or less real projects that were submitted.

Project 1 — Credit Compass*

A credit score improvement tool with a clean, modern interface. Users upload their credit report (or in a production environment, it’s automatically pulled during signup), and an LLM analyzes the data to provide personalized advice for improving their score. The system highlights specific actions and their potential impact on the user’s creditworthiness, all presented in an intuitive, user-friendly interface.

• side-note, the word compass was a dead-giveaway that the app is AI-generated — that was a separate pattern.

Project 2 — TalentSifter

A automated resume screening workflow built using an AI automation tool called n8n. When a resume arrives by email, n8n was used to automatically upload it to Google Drive and then Claude Sonnet analyzes its compatibility by cross-refencing the job description (in a separate google drive location). The LLM highlights candidate strengths and weaknesses, then populates a Google Sheet with all applicants ranked by compatibility score. The hiring manager simply reviews the Google Sheet, with hyperlinks to the original resumes if needed.

Project 3 — Sports Betting Arbitrage

A sports betting arbitrage app that monitors odds across multiple sportsbooks in real-time. The system identifies profitable opportunities created by asynchronous updates between platforms. For instance, if BetMGM offers odds at +10 while competitors show +30 for the same event, the app immediately alerts users via email or push notification. In advanced implementations, it could even execute bets automatically to capitalize on these temporary market inefficiencies.

Takeaways

Now, you may have realized that all of these apps already exist, albeit in slightly different forms. You were perhaps rolling your eyes at the idea of a sports betting arbitrage app, thinking “this has already been done.” And, indeed, it has. Likewise, at a previous job we had a “Credit Scorecard” with tailored information for customers about how to improve credit scores, which is very similar to Credit Compass. And, there are plenty of AI-powered resume-scanning tools that do more or less exactly what is described by TalentSifter. In short, there was nothing fundamentally novel submitted, just LLM-generated variations of tools that already exist.

The lesson here is that the barrier to entry has dropped to almost nothing. These projects were submitted by anyone from marketers to software engineers. Granted, the software engineers submitted more sophisticated projects, but all of these projects were generated in a matter of hours. What this tells me, is that any problem that is even remotely close to one of the three examples listed above will be solved by LLMs in the near future.

So, let’s formalize exactly what types of problems LLMs can solve.

Category 1: A clean UI with intelligent assistance

All apps, websites, etc. that are designed to help us with something (think banking app) will be fully integrated with AI agents. I’m not talking about the AI chatbots which, as of Sep 2025, are the standard AI tool that all big companies are marketing, but instead, I’m talking about genuinely helpful AI intelligence layered deep in the fabric of the architecture.

This full integration of AI in mobile/web design means all information is available to an intelligent hub that can pull and push information as needed. The problem with today’s chatbots is their limited ability to take meaningful action. They primarily answer questions you could Google, occasionally enhanced with retrieval-augmented generation from proprietary databases (essentially, predetermined protocols for answering specific questions).

The future of web/mobile development will feature AI with comprehensive system access, guiding customers through processes while accepting natural, unstructured input. We’re moving from structured forms with bolted-on chatbots to flexible interfaces with deeply integrated AI capabilities that can actually get things done.

Category 2: Extract, Transfer, Load

This category is about full stack pipelines. These are tasks where humans typically extract information, often through a phone call or an intake form, and then transcribe it into a new place, usually in a different format. At its core, this is data transformation: moving information from one place to another. AI can automate these pipelines. It can hydrate databases with scraped web data, modernize intake forms with cleaner interfaces, or guide users through complex processes like disability insurance or taxes. Think of patient intake forms automatically feeding into doctor summaries and notes. It is information flowing from point A to point B, with transformations along the way.

Category 3: Centralizing Scattered Knowledge and Instant Insights

This category is all about pulling in data from disparate sources and transforming it into cohesive, bite-sized chunks of intelligence. In our example, it was sportsbooks, but the concept extends far beyond gambling. Wherever fragmented information exists across multiple platforms, AI can consolidate and synthesize it into actionable insights.

Consider commodities trading: an AI application can simultaneously track oil tanker movements, refinery operations, geopolitical developments, weather patterns, and market sentiment to deliver comprehensive analysis that would take a human analyst weeks to compile. The value is in automatically gathering data, extracting the useful information, and highlighting what matters most.

Naturally, there is some overlap between these categories. An extract-transfer-load pipeline might pull from multiple data sources and synthesize information, blending elements of Categories 2 and 3. Nevertheless, most applications I reviewed aligned predominantly with just one of these patterns.

Summary

As we move forward, I expect these functions to be increasingly automated. Jobs that revolve around designing basic interfaces, transferring information between systems, or manually consolidating data from different sources will soon be obsolete. These tasks are not just being augmented by AI, they are being replaced entirely, with no need for a human in the loop.

I am not the type to jump on the “AI is taking your job” bandwagon. What interests me more is what AI, and specifically LLMs, are actually capable of in the near future. The categories above highlight tasks that AI is uniquely well suited for, and the barrier to entry is now so low that it is inevitable AI will absorb them all. It is the same dynamic that made thesauruses and phone books irrelevant once search engines arrived. There was simply no reason to flip through paper listings when everything was instantly available online.

What shocked me most about the vibeathon was how easy it has become to automate these solutions. With a chatbot like ChatGPT or Gemini, you can design and deploy a polished HTML page in a single day with no real software development experience, as long as you have some basic programming knowledge. If you know a little more, tools like Cursor AI let you go further, creating not only front-end pages but full-stack data pipelines with backend databases and built-in automations.

This article has focused on the near-term inevitable impact of AI on industry as we know it. However, there’s also the industry as we don’t yet know it. Predicting how AI and LLMs will transform domains that don’t yet exist is impossible with any certainty. Consider how the internet enabled social networks, which in turn fundamentally altered human psychology and social dynamics in ways nobody anticipated. Or, how the industrial revolution wasn’t just about steam engines but ultimately transformed global climate, contributing to wars, refugee crises, and political polarization that we are still dealing with.

LLMs will undoubtedly open a Pandora’s box of good and bad novelty. Perhaps the attention mechanism that powers these models will yield profound insights into human cognition. Or, perhaps embodying LLMs in hardware that allows them to interact with the physical world will lead to the next round of technological revolution.

There is simply no way for us to see beyond our current event horizon.