I would like to express my deep frustration with using advanced AI assistants (like the model I use) for developing and debugging complex workflows. The promise is of an autonomous coding partner, but the reality has proven to be a machine for burning money (tokens) and, above all, time.
I’ve been dealing with the same automation scope for weeks, and the AI’s modus operandi is constant and exhausting: it analyzes the scenario, writes code patches, and returns the task to me claiming with 100% certainty — and even a certain arrogance — that “now the problem is definitively resolved.” However, when it comes to actual testing in production, nothing works.
On my last issue alone, the AI guaranteed success and failed miserably at least 4 consecutive times. With each failure, an apology, a fancy new technical justification, and another failed attempt.
The biggest problem is that it’s glaringly obvious the AI wasn’t trained to understand and solve the root cause of the problem structurally. Instead, it acts reactively, creating “workarounds” or trying to plug holes with superficial solutions. And worst of all: it has an absurd inability to test and validate its own work before delivering it. In one of its “fixes,” it broke the system simply because it swallowed a special character when injecting the code, resulting in a basic syntax error.
The toll of trusting this tool? Weeks of my time thrown in the trash, my patience completely exhausted, and a massive financial cost in millions of tokens consumed by an infinite loop of incompetence, false solutions, and robotic apologies. If you need a concrete resolution and someone who guarantees that the code was actually tested before deployment, don’t rely on these AIs. They are still not good for fixing real problems, only for creating new ones.