How AI impacts software development at different layers

A year ago I wrote that AI-assisted programming had reached a tipping point and spoke about the spectrum of AI-assisted coding: from smarter tab completions to fully autonomous app builders.

A year later, the tipping point has truly been passed. By the end of 2025, the frontier LLMs were good enough to use in really complex projects and adoption skyrocketed. Today, in mid-2026, it is clear that AI is profoundly changing how software development works — and it's here to stay, with all its good and bad consequences.

As organizations adopt AI more widely, it's becoming clear that this new technology brings different challenges and opportunities at various levels of software organizations: individual, team, and organization-wide.

Individual developers

The most visible changes are at the level of an individual software developer. These are also the most talked about, and, at the extremes, the discussion focuses on whether "AI will replace programmers" or whether "AI can ever write code as good as a human's".

Both extremes are misguided. AI can already write better code, on many tasks, than many humans. But it can also get spectacularly stuck on a seemingly-trivial thing, doesn't self-reflect, doesn't understand what's important and what's not, and cannot be held responsible for its decisions.

This won't change soon, if ever, so AI won't replace software developers. But it will redefine what they need to do, how they do it, and how they see their jobs.

Historically, people (especially non-devs) usually associated software development mostly with thinking about code, typing out code, and looking at the code trying to spot a bug. Less visible parts were thinking about the code structure, data structures, program flow, edge cases, maintainability, scalability and sensible error handling. These are at the intersection of business requirements and deep technical expertise, are a poor fit for today's AIs, and will be increasingly the type of challenges software developers will have to think about. The AI will do the "boring" coding tasks.

This is where senior developers have an edge, because they have the field experience, deep knowledge and understanding of the systems they work on. They can better control and guide the AI, spot potential problems and wrong turns.

Junior developers lack this depth of experience and wisdom and are less likely to spot subtle errors (of intent, not code) that the AI might make. Juniors are also most vulnerable to job cuts (or, more realistically, future job prospects that never materialize) because AI is good at exactly the kind of work in the entry-level roles they'd start their careers with.

How to solve that remains a big, critical, unsolved question.

Team dynamic changes

The second level that AI-assisted development changes is team dynamics.

AI tools don't fundamentally change the structure of a software development life cycle (SDLC): discovery, planning, implementation, testing, review, deployment and maintenance are still there. But they do change how each of them is done and what gets done faster.

The implementation (coding) phase traditionally took the largest chunk of time for most tasks. With AI, its duration can shorten dramatically: claims of 10x speedup are not unheard of. However, this will put more pressure on other phases which can't be sped up easily — most notably the code review.

This shift can create pressure to "move fast, if it breaks the AI will fix it", resulting in declining software quality and team morale, and increased technical debt and cognitive debt (lack of understanding of the codebase).

To address this, at least some of the productivity gains at the implementation phase should be invested back into planning, QA, review and maintenance phases. Better code review, better automated tests, and more active maintenance can keep these potential problems at bay and even increase software quality.

But the speedup won't be 10x or even 5x: a more realistic, sustainable improvement is 10% to 50%.

The CTO dilemma

This brings me to the organizational level changes, or rather, the leadership's conundrum: how to leverage AI in software development, but minimize the downsides.

Part of the problem is all the AI hype. When organization leaders hear unrealistic claims of AI performance and believe they can 10x their productivity, avoid hiring for (or even downsize) their developer teams, it's hard for bottom-up IC-driven realistic assessment to get through.

They push for "all-in-AI" initiatives, encouraging "tokenmaxxing" (using AI token usage as a KPI) and paying only lip service to quality. This leads to exploding costs, plummeting morale and a downward spiral of software quality.

To get the AI transition right, leaders must set realistic expectations, understand both the opportunities and challenges when adopting AI in their software development processes, educate their teams on the responsible use of AI and have a simple and clear strategy on how and when to use it.

AI as a multiplier

An oft-repeated phrase is that AI acts as a multiplier: in a functional organization, it can increase productivity, while dysfunctional organizations will just be dysfunctional faster.

It's already a cliche, but it's true. The primary challenge when adopting AI (whether in software development or otherwise) is to prepare your processes and data governance, and then to understand how these processes are changed by AI. This requires conscious and continuous effort, but is the difference between having a successful AI transition and being yet another failed AI pilot.