> ping hkon.org

The Siren Song of AI 2025-07-31 by hkon

I'll focus on software development here, but this theme likely applies to other knowledge work domains as well.

Short-Term Gains

The coding copilots are doing an increasing amount of typing and thinking for developers. As AI will get better at reading, explaining and writing code, we developers will also be using them more to reap benefits such as increased efficiency and quality. Sure, not everything is perfect now, but AI agents will likely improve each year, and we are already seeing early signs of shrinking software teams.

Companies are rushing to use AI to become more efficient, deliver more for less and cut overall costs, which will mean reducing staff. We already read that technical writers, testers, designers and UX professionals are being replaced by AI and let go [1]. But further down the line developers will increasingly also be replaced by the copilot/developer duo.

Strategic Dependency, Long-Term Costs

Companies are currently reaping major benefits from AI by reducing cost and increasing profits. But what they must realize is that they are increasing their dependency on a tool whose price they cannot control. As companies become more profitable because of AI usage, the AI services in themselves will be more valuable and the costs of using them will increase accordingly. So instead of hiring and training new developers, they lay off developers and pair the remaining staff with a chatbot that reduces the depth of their creative work to prompting an AI. The lack of junior positions and media's nonstop focus on how AI is replacing developers will scare everyone but the most dedicated away from choosing IT-related studies.

So when the AI companies eventually increase the prices of various services, say, by 10% overnight, it will be a challenging time for the companies. The companies will now face what they knew would always come, only what they may not have realized is that there is no way back. They can't revert to the old ways, there are no longer competent people available. Given the trajectory we're on, in 20-30 years there will probably be very few builders left. At that point, companies will have no choice but to pay or risk ceasing to exist.

Competition

But what about different vendors, won't there be competition? Sure, you can probably select between a few similarly priced vendors. But the problem with AI is that you can't really just build your own. Like blockchain it requires enormous energy, not to mention vast amounts of data for preparation and training. It will simply be unfeasible to do anything, but pay. Many organizations are also cautious about who they entrust their data to, and these self-imposed or regulatory limitations will in practice make competition, even if it exists, more difficult.

Cloud Caution

There has been a similar scenario developing in cloud computing for some years now. Companies have now mostly completed their move to either Google Cloud, AWS or Azure. The cloud vendors, having been successful in their strategy of getting companies to hand over their infrastructure concerns to them are now capitalizing on their success. 13%, 15% cost increase overnight is not uncommon. Needless to say, by owning your own hardware you mitigate this risk. And some are now moving to self-hosted servers again, even famously so [2], but unless this becomes a common trend, I fear we'll see a similar development here.

Resist the Sirens

Do we have a choice? Yes! Companies can decide that they don't want to be put into that position. Sure copilots are nice and the added benefits will gladly be accepted, but NOT at the cost of being held hostage by the AI companies. The companies can maintain, recruit and train a staff of software engineers who are allowed to build and develop that skill with AI as a tool, but not as the builder itself. So when prices increase drastically, they will at least have the option to say no or explore alternative solutions, because the company's value will lie not in its tools but in its people.