Why Building with AI feels less exciting than it should

Many of the things we dreamed of building for years are finally within reach, and a lot of engineers feel flat about it. Here is why that happens, and why the point where magic turns into infrastructure is the part worth working on.

You can now build a chatbot, recommendation engine, or research assistant in a fraction of the time it once took. The technology is objectively more powerful, the barrier to entry is lower, and yet for many core engineers it feels less rewarding. Part of it is that people stop noticing anything once it becomes normal. The other part is that the hard problems moved somewhere else.

A strange thing is happening

A strange thing is happening in AI, ML, and software engineering.

Many of the things we dreamed of building for years are finally within reach. You can create a chatbot, recommendation engine, coding assistant, document search system, voice agent, image generator, or research assistant in a fraction of the time it would have taken just a few years ago. Yet a lot of core engineers feel flat about it.

I remember a time when building intelligent systems felt like climbing a mountain. You spent weeks reading papers, blogs, posts, and books, setting up infrastructure, collecting data, tuning models, and fighting through countless failures. Many projects (especially ML ones) never reached production. The few that did felt like real achievements because every step took real work. Today, you can get surprisingly far with an LLM API and a few hundred lines of code. The technology is objectively more powerful. The barrier to entry is lower. More people can build useful software than ever before.

So why does it sometimes feel less rewarding?

We are terrible at appreciating the normal

Part of the answer is that humans are terrible at appreciating things once they become normal. The first time you saw a Large Language Model write code, answer questions, and hold a conversation, it felt like science fiction. After seeing it a thousand times, the magic fades. The capability stays the same. Your expectations change.

The same thing happened with smartphones, GPS, cloud computing, and the internet itself. Technologies that once felt revolutionary eventually become infrastructure. Nobody wakes up amazed that they can instantly communicate with someone on the other side of the planet. That stopped being remarkable a long time ago. AI is going through the same transition.

When your edge becomes a feature in someone else's API

There is another reason. Many engineers built their careers around solving problems that were genuinely difficult. Years of learning, experience, and persistence created an edge. Watching those tasks get dramatically easier is a strange feeling. You are happy about the progress, but part of your brain remembers how much effort those skills once required. What once felt like a hard-earned advantage becomes a feature inside someone else's API. That realisation can feel unsettling.

Yet this is also how technology has always progressed. The engineers who manually optimised assembly code watched higher-level languages arrive. The people who managed physical servers watched cloud platforms automate large parts of infrastructure. Database experts saw managed services remove a lot of operational work. Every generation experiences some version of this shift.

The difficult problems never actually disappear. They move.

The hard problems did not disappear. They moved.

Building a demo is easier than ever. Building a system people depend on every day is still hard. Reliability is hard. Evaluation is hard. So are data quality, security, scalability, cost optimisation, and product thinking. The hard part just moved to a different layer of the stack.

That is why many experienced engineers find themselves less impressed by flashy AI demos and more interested in questions that are not popular on social media.

• How do you evaluate thousands of outputs?
• How do you detect silent failures?
• How do you control costs at scale?
• How do you build systems that remain useful six months later?
• How do you make AI behave predictably inside messy business processes?

Those questions are less glamorous than getting a good response out of a model. They are also most of the actual engineering.

Magic becomes infrastructure

Maybe the reason it feels less exciting is because we are living through the moment when magic becomes infrastructure. The first generation is amazed. The next takes it for granted. The engineers in the middle spend their time making it actually work.

That has always been where the most interesting problems are.