Basically no one is using AI agents for their DoorDash orders

People coming up with these tech use cases live in a reality where price sensitivity isn't a thing.

"You can call an Uber with a voice command"

"You can ask alexa to buy you things"

"Your AI will order food for you"

Seem like magical ideas that will remove friction and increase sales. Except that regular people compare 13 different options to pick the cheapest one. No one wants to see food appear on their doorstep followed by a bill they cannot pay.

     95% chance they're lying
    4.5% chance they're that foolish
    0.5% chance they're so rich they don't care.

Something that is perhaps “underreported” is that a lot of AI solutions aren’t easier to do than just doing the thing.

My budget app, for example: It’s not easier for me to ask the AI bot “show me all my transactions from McDonald’s” than it is to click on the search bar/payee filter and type “McDonald’s.”

But if I use the AI bot, I’m burning tokens.

“Get me a burrito. Ok cool are there any other places? Ok but how about ones under 50 minutes delivery? Ok now of those which has higher than 4.5 stars? Alright that one looks good, can I get a large lamb burrito with extra salsa and guac? Ok they don’t do lamb that’s fine, beef is good. They don’t do a large size, that’s alright go with a small. Yes confirm payment.”

Text is just not a good format for ordering things. You want to see pictures and just click on the options you want - we’ve had people paid 300k working on the optimal visual interface for this for probably 30 years, just trying to improve the sale rate 2%.

Agents are good for open ended problems which require stringing together lots of different data sources, which couldn’t be further from ordering food.

Just because text isn't the best interface for agentic commerce doesn't mean agentic commerce is dead. The way I'm building my open-source marketplace[0], each store's storefront and cart are rendered right in the chat through MCP UI. So you can ask the agent to change something in your cart or add something, but you also have a real cart in the UI that you can tap and edit directly.

0. https://github.com/openshiporg/marketplace

Ordering off DoorDash seems to me like the worst possible case for agents (or Alexa voice speakers before them.)

I need to see the full menu, with prices and picture of the food and options. Having an agent in the middle is just going to slow me down.

I would not trust the AI to order food off of these websites for me because the labeled data itself is pretty much shit. Lots of dishes are unlabeled, have really bad pictures or no pictures at all, and the worse is the ones that have AI generated descriptions which are often wrong or not helpful. I'm totally fine with letting it just use my saved credit cards and delivery address (I assume the AI doesn't see them) but I can't trust it to order the food I actually want.

E: As an example (n = 1 etc): I am vegetarian. I have never ordered meat dishes, ever. Yet all of these food delivery apps suggest meat dishes for me.

Recently I needed to book a table for 16 for a rehearsal dinner. I used ChatGPT work to find availabilities and create an html doc of only restaurants that could handle our party size, with pictures, menu comparisons, locations, and approximate prices. It knocked the task out of the park and I went with one of the recommended restaurants.

That said, the last mile, the bit where I paid, I called the kitchen manager and talked to them. Even if I trusted ChatGPT with this part, I like building a rapport with who’s serving me and my family food, and it’s a special occasion so I know they want to nail the experience. ChatGPT saved me so many hours of work and research and replaced Google, yelp, and so many other random services.

I don’t think agents are going to work with DoorDash they’re going to replace DoorDash - taking their core competency of helping me discover and choose what’s new and interesting to me, what restaurant is open and meets my criteria and let me stay connected with the people rather than building a wall between me and them

Remember when Amazon thought that people would say "buy me more paper towels" to their Echo? And of course they didn't because that's insane.

> DoorDash only rolled out its Model Context Protocol (MCP) connector less than a month ago. ... Without that kind of linkup—which is still only available to a small number of “early beta partners” and mainly used to streamline corporate lunch orders

So it's available to almost no one anyway. Misleading title.

"I want to order food from this restaurant"

"Sure, here's 10 options for different food from this restaurant that you totally couldn't have found just by searching for their website instead of writing this prompt!"

I think eventually demand for this is gonna pick up. I order the same thing every single time(though on a local alternative & not doordash) and I totally wish I could have an agent drive that. At my lunch time ask me if it can order the usual, I say yes it places the order, monitors it - essentially make sure I get my lunch brought to me without - the cognitive overload from interacting with these food delivery apps or in some cases speaking with the delivery people. Yeah - chat might not be the right interface for that there's the MCP and I think more people if they knoew how powerful that was would be excited about the possibilities of building custom workflows

Similar concept as with shopping, where the selling point is "your agent will do all your shopping for you"

But the wide consensus from most of the industry (including AI people) is that agent shopping is useful for stuff you don't like to shop for (like groceries, computer parts, ad-hoc supplies like paper towels)

But for clothes, and interesting things, where the experience is akin to window shopping, scrolling and imagining what you look like in a particular shirt is actually an integral part of the experience.

Guessing that doordash/ubereats has a part of that appeal, where you get to imagine/visualize the food before you order it.

The problem with AI assistants in general is that humans barely ever know what they want and the promise is that AI can intuit/predict what we want. It can't. People like to shop. It's easier to shop with our eyes than our ears. Personal assistants, as far as I can tell, have one use case and that is doing mundane chores like e-mail summaries. But to do that, you have to give them access to everything, so it becomes a transaction of privacy. And I think a lot of people are paranoid. Plus each one is a little token bonfire that is probably costing them money.

And I feel like the labs must all know this, so it makes me wonder why they're actually trying to push these assistants onto us. The answer I'm left with is training data — data theft. They'd all violate their privacy terms if it earned them an extra nickel.

For me, the best use of AI is making bespoke software connecting systems and devices that have no business being connected but I find value in so I can have a unique interface for all my tech and hobbies. Not sure how that gets translated to the everyday man but it sure isn't "bring me food".

When I have demoed the use of agents, I usually do it with an agent combining multiple applications' results. Something like: "We're having a family gathering at grandma's. Which of us can stop in at the drugstore to pick up her prescriptions on the way without adding a lot to our travel time?"

Agents are great at moderately complex combinations of data and apps. If I had to do this every day or every week even, it might be worth creating a script suited for a task like that using some automation system. Agents are great at providing a disposable saw when it doesn't make sense to stop and sharpen the saw.

I'm not a doorDash user, but they've been doing it long enough to have an app that's probably highly polished. There's not going to be a lot of upside to speaking or typing an English language command to get what you want. Trying to do that is a waste of a well designed app and the agent's token budget.

I built https://bit.recipes to facilitate my agent ordering my groceries (saves ~60 mins of family annoyance a week). it’s really meant to apply more rails and human/family collaboration, rather than a pure AI play or anything

This has always been an ai use case that made no sense to me. The UI for ordering food is good, it's visual, it shows you the food you're getting and how much it costs. There's no reason for a chat interface to be an upgrade here. Chat interfaces only have an advantage when you need the nuance of language. It's good at the things where trying to build a UI for it would collapse into an incoherent jumble of visual language.

Some of these use case feel so imaginary that lot of people just don't get why would some other people do them. Like with food delivery I want to select specifically what I want and evaluate if the price is right. Same goes for ordering most stuff or deciding on vacation.

Maybe I could get train ticket from AI... Or order taxi. At least until it went wrong for the first time...

Bots and tech are testing humanity's levels of desire for laziness and how much they are ready to pay for buying that laziness. Turns out that the desire is limitless. As someone said, the ultimate luxury is in having maximum laziness - not having to move, talk, think or use senses - the state of being a rock.

I would use it if they let me, but you have to be approved for access to their cli and mcp is only for companies. I've had my agent place orders but clicking around with computer use takes longer than ordering by hand.

Not surprising in hindsight.

It's not a super repetitive problem (e.g.respond to 100s of messages) and human judgment is highly valued here (what am I hungry for?) and visuals are sometimes a factor.

Agents are good for complex purchases with lots of hidden fees like hotels. Less useful for burgers, though I'd definitely try it for a large catering order.

This is kind of a moot article because up until a month a go when DoorDash released their MCP, their bot detection was getting in the way of agentic ordering.

My hot take is capitalism can't deliver a "order one pepperoni pizza" product.

This is literally Pizza Hut flow right now: - Go to website, look for deals page where pizza is $10 instead of $20. - Click order, upsell on breadsticks, etc, etc. - Check out, fees, delivery charge, tip take price from $10 to $20. - On the receipt page, get SiriusXM offers.

Sellers are just too incentivized to upsell, and over-charge the less transparency there is.

Same for recipes/shopping carts. Can't trust sellers to optimize for minimum price/maximum value.

I know people that have been automating their Doordash orders even before modern AI was a thing. When done right the server cannot tell.

Because the DoorDash app is so simple, delightful, intuitive?

Look, there's a lot of AI excitement and a lot of it is dumb. But if there's a number I can text "hey get me a pizza" and it re-orders the same pizza order I make every Friday night, I'm going to use it. Would I pay for it? As a standalone service? Obviously not.

But if I could delete half the apps on my phone with their horrid dark-pattern-filled interfaces and constant upsells, I'd love that. I'll happily delegate the interfacing to an agent.

And I'd 100% pay that agent to drop the ads and upsells when DoorDash tells it to pass them along.

DoorDash seems like a resource anti-optimization that would be made irrelevant by an AI agent managing one's meals. An unofficial HEB MCP [0]—that's cooking with gas!

0. https://github.com/mgwalkerjr95/texas-grocery-mcp

* No affiliation with the above repo, but I'm considering trying it out.

2026, re-inventing APIs with 3 levels more indirection \_(ツ)_/ and calling it "innovative"

While I'm sure there is a place for AI on the food prep side of things, I think it misses the point on the ordering side. While a lot of people want something different, they don't want it to be that different and it's just easier and faster to choose it themselves.

[dead]

[dead]