How developers can keep their jobs (3)

Here are links to the posts in this series:

Our job is changing fast

The job we call software engineering is changing a lot, and fast. Frank Buchholz, security architect at SAP, wrote on LinkedIn in October 2026 (translated from German):

Despite 45 years of programming experience in many languages, Claude AI has now outdone me in every respect: a better understanding of other people's (and my own) code, up-to-date programming skills, accuracy, freedom from errors... and it does everything in a fraction of the time I would need. I literally fell off my chair when I started carrying on an old program with the AI four weeks ago.

Trotz 45 Jahren Programmiererfahrung in vielerlei Sprachen hat mich die Claude AI nun in allen Belangen überflügelt: besseres Verständnis fremden (und eigenen) Codings, Aktualität der… | Frank Buchholz | 31 comments ↗Trotz 45 Jahren Programmiererfahrung in vielerlei Sprachen hat mich die Claude AI nun in allen Belangen überflügelt: besseres Verständnis fremden (und eigenen) Codings, Aktualität der Programmier-Skills, Genauigkeit, Fehlwrfreiheit,.. und erledigt alles in einem Bruchteil der Zeit, die ich brauchen würde. Ich bin buchstäblich vom Stuhl gefallen, als ich vor 4 Wochen angefangen habe, mit der AI ein altes Programm weiterzuführen. | 31 comments on LinkedInlnkd.in

So you might get the impression that you, as a developer, won't be needed much longer. That isn't so. Your old job won't be needed any more, but you can go and get yourself a new one.

The industry's future winners

The winners are software engineers who can diagnose a broken architecture, set the target and use AI agents as refactoring tools to get there.

Nearly half of this ability is diagnosis and decision. Handling the tools with confidence is the easier half and is becoming a matter of course. So take care that you don't only practise working with AI agents. Very many people will soon be able to do that.

Make sure you know how to do the following, and that you can do it, both on your own and together with your agent:

Derive the architecture from the requirements
Design structures that stand up to the requirements and deliver quality
Write technical concepts (e.g. persistence, logging, security, scaling)
Help your colleagues (humans and agents) implement them
Evaluate and improve your architecture regularly
Keep learning

That's your new job, the one you should go and take. Power and skill are taken, not given. If you need help with that, just ask me.

The supply paradox

A deep understanding of architecture used to come from years of exactly the grunt work that is now being automated. Because hardly any juniors are hired any more, the group of people who can do rescue work is shrinking, just as the demand for it grows. These people will be scarce and in high demand, and their pay premium will grow.

So make sure you gather these increasingly rare skills, practise them and keep them fresh that way. Work with others, perhaps especially with people older than you who still come from the days when code was written by hand. And if you're still one of those who write it by hand, show the younger people, the "AI natives", how it's done.

Open-source projects help a lot here. In October 2026 I came across the Immich project on GitHub:

Immich ↗Self-hosted photo and video management solution. Easily back up, organize, and manage your photos on your own server. Immich helps you browse, search and organize your photos and videos with ease, without sacrificing your privacy.immich.app

Immich is a system for managing photos and videos. You can host it on your own hardware, so you don't depend on a provider. If you want to practise architecture on a real example, contribute to projects like Immich, especially when the maintainers have tough rules like this in their CONTRIBUTING.md:

We ask you not to open PRs generated with an LLM. We find that code generated like this tends to need a large amount of back-and-forth, which is a very inefficient use of our time. Even a one line change can have significant impact. We cannot have any confidence in an LLM, so if it's non-trivial for us to verify it works and you don't show that you fully understand all implications of the change, reviewing the PR is not worth our time. If we want LLM-generated code, it's much faster for us to use an LLM ourselves than to go through an intermediary via a pull request.

Does that challenge you? Then get going, and put it on your CV afterwards.

Where I might be wrong

My hypothesis is a bet on the limits of the models, and when I think three or more years ahead, those limits are what I'm least sure of.

But the timeless version is this: whoever knows the context that isn't written down and answers for the result stays valuable, even if models one day take over a large part of architectural thinking.

Some companies are already hedging their bets

In February 2026, IBM's chief HR officer, Nickle LaMoreaux, presented some remarkable measures at the "Leading with AI Summit" in New York.

IBM is tripling its hiring of people at the start of their careers, after rewriting its junior roles around judgement and AI oversight. A lot is happening there right now:

IBM looks beyond short-term AI gains, tripling entry-level hiring ↗The tech giant argues that increasing entry-level hiring is the only way to avoid a hollowed-out talent pipeline five years from now.www.cio.com

Here is what IBM plans to do:

Triple hiring: In 2026, IBM wants to hire three times as many entry-level people in the US, explicitly including software developers and other roles that AI can supposedly take over.
Rewrite job profiles: LaMoreaux has changed the entry-level profiles so that they focus less on work that can be automated, such as programming, and more on contact with customers.
Use junior developers differently: They still program and test, but spend much more time with marketing teams and customers, to get feedback for product planning. They also build new kinds of products instead of only maintaining existing ones.
Oversee what AI produces: According to reports, junior developers also check AI-generated output. In HR, newcomers step in when the HR chatbots can't resolve a request, and correct their answers.
Target roles: IBM is hiring for roles around AI and hybrid cloud, among them software developers, cybersecurity analysts and AI engineers.

The reasoning:

A missing middle: Whoever saves on entry-level people today will have too few experienced leaders later, and will have to poach them from competitors at a higher price.
A bet on the future: LaMoreaux expects that in three to five years, the most successful companies will be the ones that hire more entry-level people now.

So if you want to take yourself a new job, find companies that have heard this "AI gong" too. Check whether they respond by hiring, as IBM does. Those companies are wise. The others, who only have savings on their list, are less so.

My invitation to leaders

This isn't a call to do without agents. It's a call to measure. Watch what an architecture change costs over time, how many people can explain each critical system without asking the AI, and whether rescue work gets cheaper with each generation of models.

If rescue costs fall with every model release, the optimists win, and I'll say so. If they stay the same or rise, the bill is an architectural one, and it will come due.

If you're a leader and have read this far, write and tell me what you're seeing right now. I'm very curious, because a lot is changing right now.