Beware the AI middleman
With "AI" being shoved into every single software product for the sake of being able to market it as an "AI product", an alarming trend is happening in the software design industry. The shift towards "AI" changes what we are able to do with computers for the worse.
A real world example
Here’s an example from my own work. This week I needed to convert a list of tasks from the project management tool Clickup into digital post-its in Miro. Clickup has an export feature, so I could export the list of tasks into a CSV file, which creates a file where the title of each task is on a new line.
Since you can usually drag and drop images into whiteboarding tools like Miro, I tried to drop the CSV file onto the canvas as well. Unfortunately Miro didn’t handle the CSV properly by adding post-its for each line in the CSV, but instead added something like a "page" on the canvas with the contents of the CSV file in it.
I then searched Miro’s user interface to find some kind of CSV import feature, but couldn’t find one, neither in the UI itself, nor in its documentation. So for the first time I clicked their "AI button", where the chatbot interface already greeted me with an example, that leads to generating post-its.
I then copied the contents of the CSV-file, pasted it into the chat interface and added a prompt to generate a new post-it for each line. After I hit submit, an empty rectangle appeared on the canvas with the info, that it’s doing something. After about 20 seconds the first post-it appeared and then it took another second for each of the following post-its.
AI-solutionists would be wondering where the problem is. On the contrary, they’d see this as a perfect example of how useful "AI" is. Thanks to "AI" I could do something with the software, that it couldn’t do before.
How AI degrades software products today
Having to fall back to "AI" is the problem. It’s a symptom for our tools losing capabilities and for software designers relying on "AI" to fulfil common requirements. CSV files are one of the most common file formats. Providing a proper way to import CSV files is a requirement every software has. But Miro’s software design offers no options to the user how it should handle importing CSVs. Having to import CSVs via "AI" is a bug, not a feature, because it’s a multiple times worse solution to a very common problem.
Processing and converting CSV files usually takes less than a millisecond. But with "AI" it now took about a minute for 50 lines. 20 seconds startup time plus 1 second for each of the 50 tasks, sums up to 70 seconds to process a single small CSV file. That’s 70.000 times slower than a millisecond and I’m not even including the time it took to open, copy and paste the contents of the file into the chat interface. God knows what would happen, if I had to import 5000 tasks to Miro instead of 50. With 1 second per post-it to be processed, that would be 83 minutes and 20 seconds for 5000 lines, or 1 hour and 23 minutes. Waiting for over an hour to process a 5000 lines CSV file is a bug, not a feature. I wouldn’t be surprised, if it’s impossible to send a prompt with 5000 lines to Miro’s chatbot, because of the processing taking that long.
While I was watching the post-its appear one by one, I felt really bad for the completely unnecessary amount of energy, that gets wasted for such a mundane task. Why should we use the completely capable compute power sitting directly in front of us to process data, when we can make CSV imports a lot more inefficient by querying a large language model, which is processed by thousands of GPUs in a data center at the other side of the world?
Will there be a better way to import CSV files to Miro in the future? Maybe. But I think it’s very unlikely while the "AI hype train" is still going, because Miro’s investors want to see "real world usage" of "AI" in the products they funded. Importing a small CSV file via its chatbot might have even been a showcase in a meeting with them. »Look, thanks to "AI" our software can do something it couldn’t do before!«
What this example demonstrates is, that chatbots and "AI agents" reduce the capabilities of what we can do with our software today. What takes less than a millisecond on your local machine, now takes minutes our hours, requires an internet connection and a resource-wasting computation facility at the other side of the world. "AI" downgrades computers from tools, that efficiently process tasks on their own to incapable chat terminals connected to huge data centers.
From augmenting to circumventing intelligence
For a long time software designers had the goal to make computers „bicycles for our minds“, a metaphor Steve Jobs once said in an interview. With "AI" the tech bubble seems to have lost that idea. The goal of software design isn’t to augment humane intelligence anymore, but to replace it with artificial one.
The more investors push for using "AI" just for the sake of it, the more features will only be available via chatbots. Replacing common features with a worse "AI version" is the easiest way to justify its addition. The more "AI features" a software receives, the less "human features" remain. If this trend continues, we will end up with software, where the user interface slowly disappears until the only way to interact with it, is to chat with a bot to do anything. The user interface becomes a bot interface.
This is how "AI" degrades computers from tools to liabilities. We lose our autonomy working with computers. We lose our tools and instead have to talk to a black box. Suddenly we are supposed to trust the AI middleman to click the button, we would have clicked, to send the text, we would have sent and to provide the information, we would have read ourselves.
I don’t want to ask computers to try to do stuff for me. I want to use computers to do the stuff.
Don’t comply with "AI" degradation. Protect your tools, your autonomy, your freedom.