Mistakes Are a Human Feature

In the age of AI, it’s good to be human.


Tools for AI-augmented software development have become the de facto standard for writing any kind of code today, much like IDEs were a decade ago and compilers a decade before that. AI tools take your requirements, implement them, integrate them into an existing codebase, and generate test suites and documentation. Agentic AI goes even further, releasing and deploying new features, monitoring for issues, fixing and deploying patches, and performing automatic maintenance operations.

It seems like there is nothing left for a human to do, right? On the contrary! It has never been more worthwhile to be a human in software development.

As I explained in my previous post, current approaches to AI will not achieve human-level reasoning anytime soon. This is good news, because it eliminates most of the doomsday scenarios where a Skynet-like AI decides to terminate humanity, and it leaves plenty of room for human judgment and creativity while freeing us from the tedious heavy lifting.

Understanding the big picture, bringing new ideas to the table, communicating them effectively, fostering organizational culture, and helping your peers are skills gaining importance over traditional engineering tasks that AI already performs well. In the words of legendary developer Kent Beck:

The value of 90% of my skills just dropped to $0. The leverage for the remaining 10% went up 1000x.

Indeed, it seems that soft skills are the new hard skills, even in a technology-focused discipline like software development. Now that any junior can generate high-quality code even for complex features, the added value of a senior developer lies in their role on the team as a leader, communicator, and facilitator. Without code reviews, juniors will never develop good judgment. They will simply produce output without learning, failing to grow their talent.

AI output often appears perfect because AI plays it safe, defaulting to averages stripped of any bias or individuality. It avoids obvious mistakes because it never makes bold choices; it never experiments unless explicitly told to do so.

Mistakes are more likely to happen when you leave your comfort zone, try something new, explore uncharted territory, or propose a novel hypothesis—things that AI rarely does. By making a mistake, you challenge your own knowledge and learn something new. So don’t strive for perfection. Be yourself: individual and interesting. And feel free to make a few mistakes along the way!

When you embrace being human, you can also accept mistakes as natural and stop worrying about them so much. There’s no need to compete with the near-perfect phrasing of LLM-generated text. Human-sounding statements often deliver important information much more effectively, even with minor grammatical errors. AI-generated text usually sounds too formal and stuffy for everyday communication with peers. Using your own words without unnecessary fluff is easier and less time-consuming to read, especially for the people around you who already know your voice and style of reasoning.

Accepting our human imperfection can be tough for many when compared to polished AI output. Understanding that AI’s perfection is also its biggest drawback can help us shift to a healthier mindset and turn our flaws into strengths.

Happy stumbling!

✱ ✱ ✱


P.S. #1: This text was drafted entirely by hand and checked for errors by an AI. Some imperfections were kept.

P.S. #2: I included the single em dash (—) in the text on purpose when writing the first draft myself.