This is How AI Changes Software Developer Roles

AI coding tools are rapidly moving from “interesting but too unreliable for real projects” to useful, indispensable aids, despite their current problems. Even incremental progress expands the range of tasks they can handle.

We’re also improving the tools and processes that make AI agents more reliable and easier to integrate into development workflows.

What does that mean for developers, especially juniors entering the industry?

Craft vs. glue work

Most software development falls into two categories:

• Finely crafted work of art

This demands pushing performance, scalability, or reliability to their absolute limits. In such cases, experts invest days, weeks, or even months perfecting a few lines of code that deliver massive impact. Research on new algorithms, protocols, or work close to the hardware requires a high level of skill, experience, expertise, and good judgment.

• Standard, repeatable plumbing work

This isn’t glamorous, but it’s necessary for a working product and a good user experience. Things like billing, onboarding, API integrations, CRUD, analytics, custom business logic, and the boilerplate glue code that turns core functionality into a usable product.

The majority of software written today, both in-house and for external users, falls into the second category – ordinary, day-to-day work. The median developer spends most of their time on such tasks.

Those tasks are exactly where AI tools excel. Their training data contains countless examples of similar code, and there are accepted best practices, tools, and frameworks. On these tasks, an AI agent can efficiently generate boilerplate, translate requirements into working code, or scaffold an entire feature.

AI empowers product owners

AI tools now let product owners and product managers be more directly involved with implementation.

While they should avoid drive-by coding, they can experiment and create prototypes, mockups, or proof-of-concept changes to identify priorities, validate user experience, and uncover implementation details before asking developers for a production-quality version.

Meet AI-native product developer

Traditionally, product managers decide what to build, product owners prioritize and flesh out features, and developers implement them. With AI capable of producing more and more code of acceptable quality, the boundary between these roles is shrinking. Someone who understands both the product goals and the technical details can often go directly from idea to working code with AI.

That makes a developer who understands the product side far more valuable, as well as a technical product owner with the skills and experience to grasp the details. In many organizations these roles may increasingly overlap and combine into a unified “product developer” role.

The AI-native product developer understands the business context, knows the user and the problem being solved, and is empowered to define and prioritize work. They are also technical enough to work closely with AI on implementation, guide it, review the code, and take responsibility for the quality of the end result.

How junior developers can adapt

Many junior developers are experiencing existential dread: will AI replace them? Many technology leaders worry about the same thing: if nobody hires juniors because AI does the work, how will anyone gain experience and become a senior?

It’s an important question. The answer is surprisingly positive but requires commitment from the junior developer, their manager, and the organization.

The key is recognizing that one of the junior’s jobs – if not the most important – is to learn. AI is an excellent learning tool. Imagine an infinitely patient, knowledgeable onboarding buddy that can explain a codebase, project, framework, or language. Used this way, AI can be an immense help to a newcomer.

When using AI, ask it to explain how and why something works. LLMs may hallucinate, but they usually give a good first approximation. Dig deeper until you’re satisfied. Experiment to verify. Explore alternatives. Ask “Why not X instead?” Break down problems into smaller parts. Play the detective, treating AI as a somewhat unreliable witness.

Whatever you do, don’t use AI as a shortcut! That’s where the danger lies. If you let AI think for you, the results will be subpar and you’ll miss out on the learning. Above all, never use AI output as the final result without understanding how and why it works.

While it’s easy to blame “lazy juniors” for copy-pasting results without thinking, most of the blame lies with the organization.

If a manager expects a junior developer to be much quicker because they have access to AI, of course the junior will “cheat.” It’s not due to laziness; it’s due to unreasonable expectations.

Even with AI, juniors will still be slower because they’re constrained by the pace of their own learning, not typing speed. Pressuring them will impede learning and hurt productivity for the entire organization.

The tools are here – use them wisely

AI is changing how software gets built. Some roles will shift a lot; others only slightly. But everyone still has agency. Whether you go deeper into specialized engineering, broaden into product ownership, or use AI as a learning partner, how you adapt is the deciding factor. The tools are here. What you do with them is up to you.