What a difference a year makes: AI in 2026

At the start of this year, I wrote an article about AI, and I wasn’t kind. Neither was much of the industry. “Hallucinations” were a big problem that it felt would never go away; many said could never go away. After all, AI didn’t actually think; it was a glorified prediction engine and could be nothing more.

For many industries and use cases, this was challenging. For software development, even more so. We went from Stack Overflow to AI as a “moderately underperforming, sometimes confused intern” and things got messy. Linus Torvalds wasn’t sold, productivity for developers appeared to be a lot worse than even they themselves thought, and managers and CEOs were worried not only about bugs and security, but the very software foundations that their businesses were being built on. Context, code complexity, architecture, user requirements and UX were all thought to be beyond what AI could comprehend.

But technology and development continue to move at a relentless pace, and in 2026 the AI companies have shown just how relentless that pace can be. With global competition, the race to be first to AGI, and enormous product-market fit for software development, the pace has exploded.

Which brings us towards the end of 2026. What a difference a year has made.

I’ve spoken to many junior developers over the past few years as part of my job, many at the start of their careers, many worried about what impact AI will have on their ability to make a living doing the work they’ve trained for. Until the start of this year, many of us hoped that the impact would be modest, even though Stanford researchers had already spotted early warning signs for young developers. But now? In the UK, Computer Science degree placements are down. In the US, tech companies have announced more job cuts this year than in the whole of 2025. And some AI company bosses, such as Mistral’s Arthur Mensch, now say their engineers no longer write a single line of code. That isn’t yet the norm across the industry; most engineers still write much of their own code. And where it does happen, those engineers haven’t stopped working. They direct the AI, then review and edit what it produces.

Even my consulting for interviews has become a somewhat peculiar affair. Getting prospective candidates to develop a standalone app or a code snippet simply no longer makes sense. AI can do that in seconds, and do it very well indeed, tests included. Its ability to architect sensible projects, understand context and nuance and even closely mimic existing code conventions makes such entry barriers obsolete. The skills for people wanting development jobs are now sliding inexorably away from pure code generation towards AI code prompting and review, and interviewing has changed accordingly. In January I wrote about “Jim”, a developer who couldn’t effectively review the code AI produced for him. That skill will increasingly be the job of software developers as we head towards 2027 and through into 2028, and there’s no way to avoid it. Sure, we aren’t yet at feeding Claude a Jira ticket and getting release-ready code out, particularly on large, complex codebases, but hoo boy are we close, and you can guarantee it’s on the agenda of many people (Proton being a case in point).

So, my discussions with juniors and with interviewers are now very different from earlier this year. Technical skills are going to have to evolve, and some will likely diminish in importance, as soft skills and business acumen start to take a more front-row seat. The days of developers left alone all day to churn out code are numbered (some say they’ll be gone by 2030). They’ll be replaced by people with holistic, team-focused views, a focus on customer requirements and delight, a knack for getting AI to write what’s needed, and the skill to review that code to ensure production-ready output and long-term codebase health. That’s where we’re headed, and that’s how I’ll be conducting my interviews. I’ll be encouraging interviewers and candidates to take all of that on board as we move forward.

For businesses with apps in the App Store, this cuts both ways. Code is cheaper and faster to produce than ever, but someone experienced still has to make sure it’s production-ready and that the codebase stays healthy. That review role is the one AI hasn’t replaced.

Times they have most certainly changed in 2026.