During the summer of 2026, three meteorology students spent nearly three months building weather applications with AI. One developed a storm chasing dashboard packed with professional forecasting tools. Another created a fishing forecast application designed around the weather conditions anglers care about most. A third built a turbulence dashboard to help pilots visualize conditions before they fly.
At first glance, these projects might seem like examples of AI-generated code replacing traditional software development. In reality, they demonstrated something much more valuable.
Throughout the summer, our interns worked with AI to generate code, troubleshoot errors, explain unfamiliar technologies, and accelerate development. But every successful feature, every corrected mistake, and every meaningful improvement depended on something AI couldn’t provide: human expertise.
Their experience revealed an important lesson that extends far beyond meteorology. AI is making it easier than ever to build software, but it hasn’t eliminated the need for people who understand the problems they’re trying to solve.
AI-Generated Code Isn’t the Same as Expertise
AI has become remarkably good at writing code. It can generate interfaces, explain programming concepts, suggest solutions, and even build working features in minutes. It’s easy to see why so many people wonder if software development is changing forever.
In many ways, it is.
The barrier to writing code is lower than it has ever been. The barrier to understanding whether that code is correct hasn’t changed at all.
Throughout the internship, AI regularly suggested solutions that looked convincing but weren’t the right answer. Sometimes it misunderstood a request. Sometimes it made incorrect assumptions about weather data or hallucinated entire datasets. Other times it confidently recommended approaches that simply weren’t the best way to solve the problem.
The interns quickly learned that AI isn’t an authority. It’s a collaborator. Like any collaborator, its work has to be reviewed, questioned, and improved.
That became one of the defining lessons of the summer. The quality of the final application depended less on how much code AI produced and more on how well the person behind the keyboard could evaluate it.
Domain Knowledge Makes Better Software
Every intern approached AI a little differently, but one theme appeared over and over again in their AI journals: expertise guided every important decision.
Storm Pointer:
When Alicia began expanding Storm Pointer from a simple idea into a comprehensive storm chasing dashboard, AI became an invaluable development partner. It helped troubleshoot APIs, explain unfamiliar data formats, brainstorm solutions, and write code. But it also produced answers that sounded correct while missing important meteorological details.
One of the clearest examples came while building forecast soundings. Rather than relying on existing graphics, Alicia wanted the flexibility to generate her own HRRR-based soundings, hodographs, and severe weather parameters for any location. AI helped her explore the technical implementation, but she still had to compare the results against trusted meteorological tools to verify they were scientifically accurate. Without that validation, attractive graphics could have communicated misleading information.
FishCast:
Lex discovered something similar while creating FishCast, an application that combines weather data with fishing conditions. AI helped accelerate development, but it couldn’t determine which weather variables actually mattered to anglers or why. Those decisions came from research, personal experience, and understanding how weather influences fishing success.
As her application grew, AI occasionally misunderstood prompts or struggled to implement new features correctly. Rather than accepting the first answer, Lex refined her approach, experimented with different solutions, and continued improving the project. The technology helped build the application, but it was her curiosity and subject knowledge that shaped what the application ultimately became.
RuffRide:
Antonio’s experience building RuffRide, a dashboard that helps pilots visualize turbulence before a flight, highlighted another important aspect of working with AI: context.
Early in the internship, he realized that AI performed dramatically better when it understood a project’s history. Instead of starting fresh each session, he maintained detailed Markdown files that documented important decisions, project structure, and implementation details. Those running notes helped both him and the AI stay organized, reduced repetitive setup work, and made it much easier to transition between projects.
In every case, AI became more effective because someone with expertise knew how to guide it.
Why AI-Generated Code Still Needs Human Review
One of the biggest misconceptions about AI is that if code compiles successfully, the problem is solved.
Software development has never worked that way.
Code can be syntactically correct while producing misleading results. A dashboard can display beautiful graphics while visualizing the wrong dataset. An algorithm can function exactly as written while answering the wrong question.
These are problems AI often cannot recognize because they aren’t programming problems. They’re domain problems.
Meteorology provided the interns with an important advantage. Years of coursework taught them how weather systems behave, how forecast products should look, and which datasets were appropriate for different applications. When AI produced questionable results, they had the knowledge to recognize something wasn’t right.
That process mirrors how experienced professionals use every other tool available to them.
Meteorologists compare multiple weather models instead of trusting a single forecast. Engineers validate calculations before approving designs. Scientists verify experimental results before publishing conclusions.
AI-generated code deserves the same level of scrutiny.
The faster AI becomes at writing software, the more valuable careful review becomes.
Better Software Comes From Better Collaboration
Perhaps the most surprising outcome of the internship wasn’t that three meteorology students built sophisticated weather applications.
It was how quickly they shifted from asking AI for answers to working alongside it.
Instead of treating AI as an all-knowing expert, they learned when to ask follow-up questions, when to provide additional context, when to reject a proposed solution, and when to investigate the problem themselves. AI accelerated repetitive tasks, but critical thinking remained entirely human.
That’s an important distinction because it changes how we think about the future of software development.
Rather than replacing expertise, AI amplifies it.
Someone who understands weather can build weather applications more efficiently. Someone who understands aviation can build better aviation tools. Someone who understands emergency management can create more useful emergency management software.
AI helps close the programming gap, but it doesn’t replace years of education, experience, or professional judgment.
Expertise Is Still the Competitive Advantage
The summer of 2026 wasn’t an experiment to see whether AI could replace meteorologists or if vibe-coding works.
It became an experiment in how far knowledgeable people could go when they worked with AI instead of against it.
By the end of the internship, every student had built an application that would have been significantly more difficult to create just a few years earlier. AI accelerated coding, explained unfamiliar technologies, and removed many of the technical barriers that often slow development.
But the success of those projects wasn’t defined by AI-generated code. It was defined by the people who knew when that code was correct, when it needed improvement, and how to transform it into software people could actually trust.
As AI continues to evolve, that lesson will become even more important. The professionals who benefit most won’t be just those using AI. They’ll be the ones who combine AI with real expertise, ask better questions, challenge questionable answers, and understand the problems they’re solving.
In the end, that’s what our interns demonstrated all summer long.
AI can write code.
Expertise builds great software.