Ya as someone who is a builder in the fintech space... it doesn't work to wave hands and stitch together a bunch of third-party software into one offering.
Regardless of how neat the integration seems on paper, in practice the general "law" is that it won't work in the way you specifically want it to work. Often users get frustrated because they see the third-party has a feature, but that specific feature isn't setup to be integrated well, and instead the third-party "integration" hub is just the least common denominator functionality, which achieves "logo integration" but barely delivers value to users.
Developers are spoiled thinking about super well documented & flexible APIs like stripe, whereas in many industries b2b software providers are nowhere as robust. If you're not directly their customer, they won't put in the extra effort to support your integration.
No one wants dark/cloud kitchens, and I don’t think anyone wants dark software either.
It’s extremely hard to do many things well. If a restaurant specializes in pizza, they’re going to get good at doing that — their employees will know the best way to cook them, their recipes will slowly evolve over time, etc. If a restaurant rarely cooks a pizza, none of that experience and refinement is there.
Not to mention, DoorDash doesn’t have any obvious health ratings visible for restaurants in NYC. I’ll happily order from somewhere I know, but not some unknown restaurant. And what’s to stop a poorly rated dark kitchen from closing shop and reopening the next day under new branding?
The whole model feels gross. It’s centered around profit and questionable tactics, rather than making a genuinely good product that people come back for.
> I’ll happily order from somewhere I know, but not some unknown restaurant
Exactly this.
Part of why I go to (or order from) a particular restaurant is that I've been there before and I have an expectation of the taste, quality, service etc. I could also, in theory, go look into the kitchen and see if there are safety issues etc (or outsource this to an inspector). Feels a lot tougher to do this if the "restaurant" is just a label on top of the dark kitchen product.
On a side note, this is why chain restaurants were so successful: you could go into one in any part of the country and have a predictable experience with some base level of quality. The mom and pop restaurant in the town you've never been to wouldn't have that same offering to you (it might for locals though).
I just wish the food problem was completely solved so I would be happy with dark kitchens. There should be a service that I don't have to think about bringing me healthy foods with taste thay I can bear. That is all I care about.
> AI will cause these dynamics to spiral even faster, making software even cheaper and distribution more expensive. It’s still early, but there are plenty of fascinating examples of AI acting as an increasingly competent replacement for engineering.
I've been hearing this for years and I'm still waiting on it. The most I've seen is some employees using ChatGPT or similar to produce some initial code. But even those employees reported you have to spend the same amount of time painstakingly analyzing each section of code because it contains unusual or bizarre bugs in edge cases.
Ive been working in AI for about a year, and Ive been working as a web developer for about 20 years. It appears to me that everyone that thinks AI is going to handle development for companies (in the reasonably near future) are also folks who have never had high level engineering responsibilities for a single project for a multi-year period. That is, theyve never been exposed to the realities of how nuanced managing a software product is at a technical level. The bots can barely write code-bootcamp level scripts. Im not confident that we will see AI solving the kinds of (coding) problems that engineering leaders are handling over multi-quarter projects.
I’m an AppSec engineer, and work with 300+ devs and Software Engineers. I’ve worked through start phase and two acquisitions. You spot on at this point in time. I’m on the team responsible for testing m365 copilot before it rolls out to our org. A month ago I would have agreed with you 100% but now i’m leaning more to theirs a 50% chance of large scale automation happening within 5 years.
What AI was missing is the larger business context. I doesn’t know the politics behind why things are the way they are and why fixing and issue might cost the company 50k every minute or if library is updated it would break 15 business critical products without proper coordination.
M365 Copilot is bridging that gap. Right now it’s dumb and only access what you can see on OneDrive and sharepoint. With plugins and connectors it’s going to integrate into every development platform sooner or later.
I still think it’s some years out and will require a lot of human interaction before these generalized agents can be onboarded.
It’s a security nightmare for me. We basically just automated the recon for any attacker that has compromised a 365 Account. In my opinion it’s moving to fast even when it’s dumb as bricks and has the context of a 2 year old.
I’ve been using it to compare static analysis findings and m365 copilot returns a lot of the same findings with mitigation suggestion. It’s still not 100% though, but either is any secuirty testing.
I give it two years before the grunt work is fully automated
My theory is that they know it won’t work, but they’re scaring developers into using and improving AI and taking pay cuts, so that they can eventually replace them. Every penny they can take from humans and invest in AI research is being used in order to be able to get rid of the humans, eventually.
The AGI that might eventually come won’t free humanity from labor, it will free the wealthy from having to manage and support laborers.
I think right now ai does enable code velocity, though the engineer still has to be an expert to direct the ai for what to write, lookup error codes, understand what the code is doing enough to make edits and ask followup questions, and so forth. The 'future promise' is having an agent-based model act as that person in the middle, e.g. Devin's feature release video. It's really hard to predict how far away that is from becoming good enough to work in the real world, though it certainly isn't here yet.
I kept waiting for the Dark Kitchen section of the article to end with "...and this is a miserable, alienating model that makes nobody happy," but it never came.
Right, the ghost kitchens make (generally) terrible food, but are able to get orders by spamming the DoorDash/UberEats search pages. And if a particular brand gets too many bad reviews, no problem. Shut down and start a new one with another punny name.
Traditionally restaurants (like many software companies, incidentally) live and die by their "brand" and by their ability to get repeat customers. This more-or-less aligns incentives between the restauranteur and their patrons. But under the ghost kitchen model, there's no reputation, no real brand, so the restaurant is incentivized only to make food as quickly and cheaply as possible.
It’s the same with news and media. The traditional media companies (news, television, newspapers), etc all had to be massive brick-and-mortar businesses with huge capex (printing presses, broadcast equipment, etc). Very high barrier to entry, high cost to play, and you had to organize all of it under a real brand tied to real plants and offices… which meant you had to care about your reputation… which meant you had to make sure your editorial department was doing a good job.
Now the web is full of bootstrapped digital media outlets that have no skin in the game in terms of real-world presence. They don’t have printing presses to run. They don’t have delivery trucks. They don’t have inventory to manage. It’s just air. There’s nothing there. Maybe a single open-plan office, or just a group of digital nomads working out of a corner of a coworking space. You can build and tear down an outfit like that in a week. Whereas to spec out, order, install, house, operate, and amortize/depreciate a single printing press would be a multi-year to multi-decade undertaking by a real business.
The cloud kitchens I've ordered from were great and often had food that you wouldn't be able to get whatever, like just basic roasted chicken. I think it's a fantastic model where a talented chef can just focus on cooking.
There's nothing inherently wrong with the disaggregation of kitchen and dining rooms, so certainly you can get good food out of a commissary kitchen. In my area there's an amazing korean katsu place that's run out of one.
But restaurants like that are fighting against the incentive structure, and the incentives always win.
That's also my issue with dark kitchens. Apps list actual, physical restaurants, next to what is just people making food in their own homes. And I've found instances of the name changes you mentioned.
There's this "premium" burger place I like has alternative "delivery first" menu with lower cost offerings and a whole different name and branding. You can also order from this menu in person but you need to ask for it. I find this to be an interesting strategy.
How doesn't it? I've ordered from some; they were fine. Expensive, but food delivery always is. This probably depends on your city, its regulations, the ratio of dark to normal kitchens, and whether enshittification-as-a-service providers have entered your city yet.
Disaggregation of food-making services and seated-dining services makes a whole lot of sense, actually. Just look at any food court - why should each food-maker have its own seating area, when seating areas are interchangeable and cheaper to deliver at scale?
However, disaggregation of food-making services and branding absolutely doesn't make sense for anyone except for people trying to confuse you to make a buck.
Because dark kitchens / ghost kitchens aren't just outsourcing their seating area, they're outsourcing everything but the logo. Ghost kitchens will frequently "cook for" more than one "restaurant," but these separate restaurants will have the exact same items on their menu, because they're coming out of the same kitchen. It's dishonest and done exclusively to put more "options" on delivery apps that funnel orders to the same business. There's a dozen other reasons ghost kitchens are bad for everyone involved (except the business owners).
> Because dark kitchens / ghost kitchens aren't just outsourcing their seating area, they're outsourcing everything but the logo.
Which means that you thought your food was going to be prepared by some people you don't know, but ended up getting food prepared by other people you don't know.
> It's dishonest and done exclusively to put more "options" on delivery apps that funnel orders to the same business.
How is that any more dishonest than multiple OEMs rebranding the same manufacturer's consumer products or conventional restaurants using the same recipes and buying the same bulk ingredients from Cheney Bros. or Sysco?
Oh, yeah. Maybe it's worth differentiating between kitchens that are otherwise legit but exclusively serve via delivery apps, and kitchens that are spam. The former can be "ghost kitchens" and the latter can be "dark kitchens".
At the root of it, the author is arguing that everyone needs to reach for common tools to patch together feature-packed solutions for a narrow set of customers. This somehow translates into it being cheaper to build and more appealing to customers. The problems I see:
1) Stuffing a bunch of third party services together to make a product is great for a startup looking to launch quickly, but in practice it is usually horrible long term. Quality integration between your features becomes difficult, and you will often sacrifice the UX as a result.
2) Outsourcing everything makes you a thin wrapper. Talk about commoditization.
3) The author seems to assume that the brand is the business, not the underlying product. That doesn't even hold for hamburgers, let alone specialized software verticals.
4) In an era of software being dumped on the market in droves, the solution is not to dump into the market even faster. I can tell you as someone coming from the medical side of things that people try doing this to us all of the time, and it's just endless heaps of garbage software. People would be smart to take the time to really think through our problems in the fine details where they matter. I know that doesn't mix well with a VC-backed growth at all costs model, but frankly that's not my problem.
The article feels like it was written by an MBA, not an engineer.
> The article feels like it was written by an MBA, not an engineer.
I'm trying to come up with a name to describe this style of writing. The key components are how hyper-confident the author is in The Future and how everyone who has even a day's more experience than the author is old fashioned and will certainly be roadkill this time next week.
So the future is more repackaging of existing products? Yippy! Stand back Vercel, I'm sewing you and a few other human centipede startups together into a new saas with 1000% markup. YC25 here I come.
The irony is that Vercel is already doing that. You're making a human centipede out of human centipedes. And each layer adds 1000% markup! Vercel adds a 1000% markup on AWS, which adds a 1000% markup on servers. You could add a 1000% markup too!
Dark kitchens are not designed to scale. In fact, they are the opposite - they only implement part of a company. There is no problem with sass as a business model. It's quite valuable to have a company manage software and data for you. The pernicious part is that they don't allow you to access your data unless you pay them in order to make it sticky. Consumers should demand access to their data and that would enable a new type of company that provides a service which will switch you between different saas services.
The current equilibrium: data are co-located with processing, both the storage and the processing ("service") are things that clients are happy to pay to offload. Bringing data in is easy and often free. Taking the data out is hard / expensive, so customers stick unless the service becomes truly poor.
Imagine a player which makes it easy to take the data away. Now customers have easier time leaving, and will leave more often. They will go to the "sticky" SaaS players, and will get stuck there. There could be a bigger influx of new customers because it's less risky to try, but "sticky" providers also offer trial periods and free tiers.
Curiously, the YouTube algorithm recommended that I read "Count Zero", so here I am.
I think Neuromancer absolutely nailed the future of business: giant ledgers holding records of every communication, event and transaction affecting the organization. And all of that information processed by AIs under the control of a small number - one even - of owners.
The AIs will decide whether they need to write and compile code to process the data or whether the task is small enough that they can just do it directly. Exactly like a human deciding whether to write a script or just use a spreadsheet. In this future the only B2B market is for AIs.
This strikes me as a rebrand of the "value-added" reseller or consultant implementation model; particularly in the middle-market space. I'm sure there are imagined differences in theory but in practice I can't imagine that it's much different with the exception of private capital coming from VCs rather than PEs.
Mobile games are already using the Dark Kitchen model; so are dating sites/apps. I remember that match.com operated almost all dating apps - around two hundred of them - at some point in time.
Many founders are realizing that the last decade’s model for building SaaS and fintech products just doesn’t work anymore. Stagnant or declining growth, high and rising CAC, declining LTV… all are symptoms of this breakdown.
.....
What’s causing this breakdown? There’s too great a supply of undifferentiated software chasing increasingly picky demand that’s also locked behind a small number of monopolistic platforms extracting ever greater margin to access that demand (search, social, app stores, etc).
I'm sorry but that's not what's causing the breakdown. Most fintech products fail because they're solutions looking for a problem. Crypto currency is a perfect example. 'Dark kitchens' work as a business model for the same reason restaurants work as a business model. It solves a problem people are actually having i.e. they're hungry and need something to eat but don't have the time to make it themselves.
Regardless of how neat the integration seems on paper, in practice the general "law" is that it won't work in the way you specifically want it to work. Often users get frustrated because they see the third-party has a feature, but that specific feature isn't setup to be integrated well, and instead the third-party "integration" hub is just the least common denominator functionality, which achieves "logo integration" but barely delivers value to users.
Developers are spoiled thinking about super well documented & flexible APIs like stripe, whereas in many industries b2b software providers are nowhere as robust. If you're not directly their customer, they won't put in the extra effort to support your integration.