AI switching costs will be real, but the true cost of switching may not be what you expect. Let me explain.
Welcome to Lead Prompt // executing leadership from the root. I’m your host, John Collins.
In the world of software engineering, we often talk about "technical debt" or "refactoring costs." We understand that moving from one database to another or shifting cloud providers isn't just a matter of clicking a button; it’s a gruelling process of migration, testing, and troubleshooting. But as we move deeper into the age of Generative AI, a new type of friction is emerging that is far more subtle, more psychological, and perhaps more expensive than the monthly subscription fee on your corporate credit card.
The best engineers on my team are already deep into AI. They aren't just using it for simple code generation or writing boilerplate. They are using it as a high-level partner for debugging legacy monoliths they’ve inherited from the past that lack documentation and are held together by hope and spaghetti logic.
As I observe them working away each day, it is becoming clear that they each have their preference for which model is their daily driver. One engineer is a Claude guy; he swears by its nuance. The other prefers Kiro for its speed and specific logical structure. I myself? I prefer Gemini, largely due to its integration and the way it handles massive context windows.
It’s not that we believe one is objectively "the best". It is a matter of professional preference. In the same way that I would not force a senior engineer to use a specific IDE like VS Code over IntelliJ, or dictate their operating system, I would not force them to use a specific AI model. They are professionals. As leaders, our job is to enable them to choose the tools that make them most effective.
The Illusion of Low Friction
At first glance, switching an LLM feels like it should have zero cost. You just open a new tab, right? You copy the prompt from GPT-4 and paste it into Claude.
Recently, however, I started to think deeper about the actual switching costs. If one of my guys decided to switch from their "daily driver" to a competitor tomorrow, what would it actually cost the organization? Apart from the obvious financial considerations like enterprise seats and API credits, I believe the real costs will be something altogether harder to measure. They are cognitive, contextual, and emotional.
Let’s look at each one in turn:
1. The Muscle Memory of Prompting
Just as a carpenter has a favourite saw, or a mechanic a favourite wrench, a software engineer develops a "feel" for their AI. Every model has its own vibe.
If you use Gemini long enough, you learn exactly how to phrase a request to get the reasoning chain you want. You know its hallucinations; you know when it’s being too "chatty" and how to shut that down. When you switch to a model like Claude or Kiro, that "muscle memory" fails you. You find yourself frustrated because the new model doesn't respond to the same shorthand. You have to "re-learn" how to talk to your tool. That period of re-calibration is a productivity drain that most managers aren't accounting for.
2. The Context Moat
The second major factor is context. This is perhaps the most significant "sunk cost" in the AI era.
If you have spent several weeks "training" an AI on your specific domain by uploading documentation, sharing snippets of your unique architectural patterns, and providing feedback on what "good" looks like in your specific company: you are building a moat.
An AI prompt isn't a one-off transaction; it’s a conversation. Every file you provide, every screen-shot of a stack trace you upload, helps the model build a "mental map" of your problem space. If you switch models, you are essentially firing an assistant who knows everything about your project and hiring a genius who knows nothing. Unless you have a robust mechanism to transfer that context: not just the data, but the nuance of the previous interactions; you are starting from zero. That effort becomes a sunk cost, which is a massive switching cost in disguise.
3. The Resonance of Tone
Finally, there is the "vibe" factor. It sounds unscientific, but in leadership, we know that team chemistry matters.
Many users find that each AI has a certain tone that they either resonate with or find incredibly annoying. Some models feel overly subservient; others feel clinical; some feel like a partner that is trying a little too hard to be your friend.
Just like a human colleague who rubs you the wrong way, you will be less inclined to work with an AI that annoys you. Conversely, you will stick with a model that provides a sense of "flow" and pleasure in the work. When an engineer says, "I just like the way this model thinks," they are talking about a reduction in cognitive friction. Forcing a switch breaks that flow, and in a high-output environment, flow is everything.
The Leadership Takeaway
As leaders, we need to stop viewing AI as a generic commodity. It is not "oil" that you can buy from any station. It is becoming a personalized extension of the engineer's own cognitive process.
When we look at the "surprising" switching costs of AI, we see that the real lock-in isn't just the contract but it is the relationship between the creator and their co-pilot. If we want to keep our best talent, we have to respect their choice of "intellectual partner."
So, I want to ask you: What is your favourite AI model right now, and more importantly, why? Is it the logic, the speed, or just the "vibe"? Tell me in the comments below; I would love to hear how you're navigating these invisible costs.
I’m John Collins, keep executing from the root. Talk to you next time.
Download audio
File details: 8.7 MB MP3, 6 mins 36 secs duration.
Title music is "Apparent Solution" by Brendon Moeller, licensed via www.epidemicsound.com
Subscribe
Sponsor
Five.Today is a highly-secure personal productivity application designed to help you to manage your priorities more effectively, by focusing on your five most important tasks you need to achieve each day.
Our goal is to help you to keep track of all your tasks, notes and journals in one beautifully simple place, which is highly secure via end-to-end encryption. Visit the URL Five.Today to sign up for free!