Big tech companies are desperate to find a use for artificial intelligence in their companies. I’m not talking about those already all-in on AI, like Sam Altman’s company, Google, Microsoft, and others. Nor am I referring to companies like Salesforce, whose artificial intelligence solutions tend to… just not work.
Instead, I mean companies that have no business adding artificial intelligence to their offerings. These implementations, like Uber Eats adding prompts to its ordering process, actively harm their existing models. Uber learned that the hard way earlier this year. Next up is DoorDash, an international food delivery service that’s about to learn the same lesson Uber (and several others) just learned.
Ask DoorDash: why?
The company has launched something called Ask DoorDash. The ‘feature’ lets users “tell DoorDash what you’re in the mood to eat, share a recipe link or cookbook photo, or describe the reservation you’re looking for to get personalized results in seconds.” Seems convenient, right? Sure, if your business model involves burning money for no good reason.
The AI implementation is tailored to users who treat their restaurant app like Netflix. You probably know the syndrome. You fire up the app and scroll for twenty minutes before picking something you don’t really feel like watching. Ask DoorDash will apparently counteract this by letting an AI model take the lead. Just spout out your vibe for the evening, and it’ll come back with an answer. Any answer. It has to have an answer, because otherwise it’s broken.
This isn’t fantastic for users. Outsourcing meal choices to an artificial system only sounds good on paper — and that’s true for both DoorDash and its customers. See, AI gets things wrong. Often. AI queries also cost money.
The ‘convenience’ theory this company is counting on is half-baked. Sure, some queries will be met with the desired answer. Others… won’t. That means fighting with an artificial intelligence system that really doesn’t want you to stop using it to get it to predict your answer correctly.
Surely it’s easier? Cheaper?
It’s easy to see how this company’s executives are thinking about AI. Less analysis paralysis equals faster orders. This means happier customers, right? Sure it does. And that means more orders. Spend a little on artificial intelligence, and get a substantial return in the form of a larger, happier customer base. What could possibly go wrong?
Quite a bit. Should Ask DoorDash hallucinate, even a bit, it’s not going to match a user’s energy. That means those users will need to query. Again. This extends the process to the point where it would have been easier (and more accurate) to make a choice on their own. Customer frustration goes up (not always, but sometimes). Customer satisfaction goes down.
And then there’s the expense. The food app doesn’t get those AI queries for free. Smash your entire customer base up against an LLM, even a dedicated one, and you’re burning tokens. That, as Uber knows extremely well, gets expensive. Fast. And if someone figures out how to make the in-house AI write code for free… well, then your delivery service chatbot is vibe-coding (at your expense) when it’s supposed to be making you money. The entire system involves far more downside than upside.
It’s not local, though?
Well… no. But it’s part of a discernible pattern, one that’s easy enough to spot. Not every service needs AI. Most products don’t. Microsoft’s Windows doesn’t need artificial intelligence stuffed into every corner. Nor do Google’s products. Uber Eats, DoorDash, and others really don’t need to add AI — even when the idea is to streamline things for users.
Most of the ‘what are you thinking’ scenarios are obvious. They also all follow the same process. Something works. It makes money. Someone in the executive seats decides that including the hot new tech trend will somehow make more money. All that matters is what was sold to them. Even if, like Salesforce’s Agentforce, the product doesn’t actually do what its sellers claim. So it rolls out, despite what it actually (or will eventually) costs.
OpenAI and Microsoft’s Copilot recently switched to token-based billing, which (slightly) more accurately reflects the actual cost of artificial intelligence. Those already deeply entrenched… were not happy. And those price increases, even those which were walked back, are going to arrive. If these systems live inside something that doesn’t need it, those increases will either be eaten by the host company (bad) or by the end user (worse). Especially when most of these implementations… don’t really benefit anyone other than the folks selling the AI solution.
