Paying for AI Without Knowing What You’re Getting
- Apr 14
- 3 min read
Have you ever hit a usage limit on a tool without understanding why? You’re not alone.
Here are my thoughts on today’s AI tools, particularly paid models, in a context where you’re paying for AI without knowing what you’re getting. Open-source and free models do exist, but the real issue lies with subscription-based tools.
What stands out is a widespread lack of transparency. It affects usage limits, computing power, and token consumption, which remains a vague and poorly understood unit of measurement.
After several years of experimentation, the same conclusion still holds: performance isn’t consistent. With OpenAI, Anthropic, and even Google to some extent, the same issue keeps coming up.
There are usage limits, but they’re poorly defined. They’re often expressed as a percentage, without any clear indication of what that percentage actually represents.
Unlike a clear metric like gallons of gasoline, here we’re dealing with “100% usage” without ever specifying what that actually means. It becomes impossible to determine what we’re truly consuming or to compare offerings across AI providers.
Context of use when you’re paying for AI without knowing what you’re getting
In practice, these limitations become even more apparent. Some platforms, like Claude, impose restrictions over specific time windows, such as five-hour cycles. These limits can be reached in the middle of a work session, creating a direct interruption in your workflow.
Beyond these limits, there’s also a noticeable variation in quality and performance. For the same prompt, using the same information, results can range from excellent to very poor. This isn’t just normal variation due to the probabilistic nature of AI models. In some cases, the drop in quality is significant enough to be obvious.
This creates the impression that providers are dynamically adjusting model performance based on overall demand, similar to pressure management in a network.
The more users there are, the more performance seems to degrade. The issue is that this behaviour is never explained. Users have no visibility into how much processing power they’re actually being allocated at any given moment. In a professional context, this instability becomes a direct problem. No one wants to rely on a tool whose quality and availability fluctuate without warning.
Economic context and dependence
Moreover, this issue is becoming both economic and strategic. Users, particularly on Reddit, report that at the same price, their usage has decreased over time. In other words, for the same subscription, the value is declining.
In this context, paying upfront becomes risky. If the service degrades without transparency, it becomes impossible to guarantee the true value of your investment. The market is dominated by a handful of players, which limits alternatives. As a result, users have virtually no leverage. They lack clear data for comparison, and the only real option left is to unsubscribe.
However, unsubscribing isn’t a viable solution when you rely on these tools for work. And when payment has already been made, it creates a form of dependency. Without going as far as calling it a trap, there is very little room to maneuver.
Reflections and suggestions
Given this situation, it becomes relevant to consider alternative approaches. One strategy is to use multiple AI tools in parallel and distribute your budget accordingly. This helps reduce dependence on a single provider and allows you to adjust usage based on your needs.
Another option is to turn to open-source models or local solutions. These approaches offer greater stability and full control over performance, since everything depends on your own hardware. They may require more technical effort, but they provide a level of transparency and consistency that current AI providers don’t guarantee.
The real issue with paying for AI without knowing what you’re getting
Ultimately, the discussion comes down to one central point: as long as paid models remain opaque about their limitations and performance, they will pose a clear problem for serious, professional use. To learn more and structure your use of AI tools with a clear and controlled approach, visit my website or contact me at info@jimmygilbert.com
Jimmy Gilbert, Technology Consultant and AI Trainer . A computer science instructor at Cégep de Sainte-Foy, Jimmy works at the intersection of education, software development, and artificial intelligence. He trains the next generation by combining technical rigor with a practical understanding of the job market, while actively integrating AI tools into his teaching. In parallel, he also works as a technology coach, helping individuals and professionals effectively adopt technology, automation, and AI.
