Invisible and Visible Plumbing Tech Companies

Here's a fun question: what do Stripe, Zapier, and a Belgian cybersecurity startup called Aikido have in common?

You've probably used one of their products without even realizing it. And that's not an accident, it's the whole business model.

There's a term for this: invisible plumbing.

What does that even mean?

Think about plumbing in a house. When it works, you never think about it. You turn on the tap, water comes out, you don't picture the pipes behind the wall. You only think about plumbing when it breaks.

Now apply that to software. Some companies build products where the whole point is that you don't see them working. They sit between two systems that don't normally talk to each other, and quietly make them talk.

A few examples:

• Stripe sits between a website and a bank. You click "pay," and somewhere in the background Stripe is talking to your card, your bank, and the merchant's account, all in about two seconds. You never see any of that.
• Deliverect, a Belgian company, connects restaurant POS systems with delivery apps like Uber Eats or Deliveroo. When you order food, an order from one app magically shows up correctly in the restaurant's kitchen system. That's Deliverect, working invisibly in the background.
• Aikido, also from Ghent, builds security tools that quietly scan a company's code, cloud setup, and pipelines for vulnerabilities, without developers having to do anything extra. It just lives inside the workflow they already use.

These companies aren't selling you an app you open every day. They're selling the connection between things. And weirdly, that's often a really good business to be in, because once a company plugs you into their systems, it's annoying to rip you back out. You become part of the plumbing.

So... is every tech company "invisible plumbing"?

Here's where it gets interesting, because at first it really seems like yes.

The argument for "basically every tech company does this": Almost every tech company connects to something else behind the scenes. Spotify talks to licensing systems and content delivery networks. Notion syncs with your calendar and Slack. Even a simple to-do list app probably pings a server, talks to a database, and maybe sends you a push notification through some third-party service. If you squint, every piece of modern software has some invisible plumbing running underneath it somewhere.

The counterargument, why that's not really the same thing: Sure, but there's a big difference between a tech company using plumbing and a tech company being the plumbing.

Here's a simple test: imagine you stripped away all of a company's integrations and connections to other systems. Would the product still exist?

• Spotify → still works. You'd lose some convenience (like syncing your library across devices), but the core thing you're paying for, streaming music, still works.
• Notion → still works. You'd lose the calendar sync and Slack notifications, but you can still write notes and organize pages.
• Stripe → completely gone. There's no "Stripe app" underneath the integrations. The connection between a website and a bank is the entire product. Remove it, and nothing is left.
• Aikido → completely gone. Same idea: strip away "connects to your code, your CI/CD pipeline, and your cloud," and there's no separate product hiding underneath. The integration was the whole thing.

A few more to run through the same test:

• Uber → still works without most of its background integrations. Yes, it talks to maps APIs and payment processors, but the core product, a driver picks you up and takes you somewhere, is the thing you're paying for, not the connections that make the app function.
• Strava → still works. Strip away the integrations with smartwatches and other fitness apps, and you can still manually log a run and see your stats. The tracking and the social feed are the product, not the syncing.
• OpenAI (ChatGPT) → still works. People pay to talk to the model itself. Sure, ChatGPT can plug into your calendar or browse the web, but the core product, the AI answering you, works completely fine on its own.
• Twitch → still works. It's a place to watch and chat during a livestream. Remove every third-party integration and you can still stream and watch.
• Reddit → still works. It's forums and comment threads. None of that depends on some hidden integration layer connecting outside systems, the content and the community are the product.

Now run the test on a company that's actually massive, one of the biggest in the world:

• Visa → completely gone. There's no app you open for fun, no feed, no content, nothing you'd recognize on its own. The entire company is the invisible network connecting your card, your bank, and the payment terminal at a store. Strip that connection away and there's no "Visa product" left at all, just a logo with nothing behind it. It's one of the most valuable companies on the planet, and most people couldn't actually explain what it does, because being invisible is the entire job.

That's the real dividing line. Spotify and Notion have plumbing somewhere in their stack, but they're not selling you the plumbing, they're selling music or notes, and the plumbing just helps deliver that. Stripe and Aikido don't have a product underneath the plumbing at all. The plumbing is the product.

So a sharper way to ask the question:

1. Does this product connect two or more separate systems?
2. Can someone start using it instantly, without a big sales process or setup?
3. Does it get more valuable the more of the ecosystem it touches?
4. If you removed every integration, would there be anything left to sell?

If the answer to that last question is "no, nothing would be left," that's real invisible plumbing. If the product survives just fine on its own and the integrations are a nice bonus, it's a regular tech company that happens to use some plumbing too.

A useful contrast: visible vs. invisible (within tech)

Notice the pattern: visible tech products are things you actively open, use, and interact with. Invisible plumbing products are things working for you, somewhere behind a button you already clicked.

Why this matters

It's not just a fun classification game. It explains why some of the most valuable tech companies in the world are ones most people have never heard of.

You'll never open the Stripe app the way you open Spotify. But every time you tap "pay" online, there's a decent chance Stripe or Mollie just quietly did its job in the background. Same with MessageBird routing a WhatsApp notification, or TechWolf feeding skills data into a company's HR software without anyone having to log into a new tool at all.

These companies win by being forgettable in the best possible way. The less you notice them, the better they're doing their job.

So next time something "just works" online, an order arrives correctly, a payment goes through instantly, a security check passes without you lifting a finger, there's a good chance some invisible plumbing company is quietly behind it, doing its thing, exactly as designed.

Where the theory gets messy

It's worth being honest that this isn't as clean a split as it might sound so far.

"Invisible" depends on who's looking. Stripe and Visa are invisible to you tapping "pay," but they're extremely visible to the business that signed up for them, negotiated fees, and stares at a dashboard every day. So it's not that these companies are invisible, period, it's that they're invisible to consumers and very visible to businesses. That distinction matters more than the article lets on so far.

Some companies are both at once. OpenAI is the cleanest example. ChatGPT is a visible product, you open the app, you type, you wait for an answer. But OpenAI's API business, which might be even bigger, is pure invisible plumbing, quietly powering features inside Notion, Shopify, Snapchat, and countless apps that never mention OpenAI by name. The same company sits on both sides of the table at once.

The "would it survive" test is a bit circular. Saying Stripe has nothing left if you remove the integration is basically just restating that Stripe is the integration. It's not really proof of anything, it's a rephrasing. Even the "still works" examples are softer than they look. Strava technically survives without smartwatch syncing, but almost nobody manually logs runs, the automatic sync is the actual reason people pay and stay. So some of these examples pass the test only by ignoring the part of the product that drives the most value.

Being invisible isn't always the goal. The whole idea assumes invisible is the winning strategy, and a lot of the time it is, but not always. Intel famously did the opposite on purpose, they took something genuinely invisible (a chip inside a computer) and forced it into visibility with "Intel Inside" stickers, because staying hidden was capping their pricing power and brand leverage. Visible products like Spotify or Reddit build something invisible plumbing usually can't, a habit and a brand people are loyal to. Invisible plumbing tends to win on stickiness and margin, visible products tend to win on brand power and pricing flexibility with consumers. Neither is strictly the better strategy, they're just different bets.

None of this breaks the idea, it just means the line between visible and invisible is more of a spectrum than two clean boxes, and some of the most interesting companies are the ones deliberately playing both sides of it.

What about AI "wrappers"?

This whole framework also explains something people argue about constantly in AI right now: wrappers.

A wrapper is a company that takes someone else's model, OpenAI, Anthropic, whoever, and builds a product around it, a prompt, a UI, maybe some workflow logic. The common criticism is "it's just a wrapper, no moat," meaning anyone could rebuild it in a weekend by calling the same API.

Run it through the same removal test from earlier: if you took away the underlying model, would the wrapper survive?

For a lot of wrappers, the honest answer is no, there's nothing left. A product that's just a system prompt and a nice landing page calling GPT or Claude directly isn't really its own invisible plumbing, it's a thin pipe sitting on top of someone else's much bigger pipe. That's the risky category, because the actual plumbing company can absorb that feature into their own product whenever they want. That's basically what happened to a bunch of early AI writing assistants, once ChatGPT could just write the marketing copy itself, the wrapper's entire job disappeared overnight.

But "wrapper" shouldn't automatically mean "worthless." Stripe is, in a sense, a wrapper around banking rails it doesn't own. Twilio is a wrapper around telecom networks. What made them real companies wasn't secret technology underneath, it was everything built around the connection.

So how does a wrapper actually build a moat instead of staying a thin pipe?

• Data the model doesn't have. Every interaction a user has builds up data the raw model never sees. If that gets fed back in over time, the product keeps improving for that specific user, and a copycat starting from zero can't catch up quickly.
• Having the right context, not just data. A model is only as good as what it's given to work with at the moment you ask it something. The hard part usually isn't the model, it's pulling together the right documents, the right history, the right state of a system, right before the question gets asked. A wrapper that knows exactly what context to fetch and hand to the model, your codebase, your past tickets, your company's specific policies, gets dramatically better answers than the same model used cold. That context-gathering layer is invisible, but it's often the actual product.
• Workflow depth, not just a chat box. A tool that lives inside the place work already happens, the codebase, the inbox, the design tool, is much stickier than a separate tab you have to remember to open.
• Vertical trust and compliance. In regulated spaces like legal or healthcare, the model is the easy part. Years of compliance work, accuracy, and certifications are the hard part, and that doesn't vanish the moment a better model ships.
• Distribution you already own. A company that bolts AI onto millions of existing users has an advantage a brand-new AI-only startup doesn't, it doesn't need to win distribution from zero.
• Becoming model-agnostic on purpose. Some wrappers build their product so it doesn't actually care which model runs underneath, swapping providers depending on price or quality. That flips the power dynamic, instead of depending on one model provider, the wrapper becomes the layer deciding which plumbing to use.
• Network effects. If the product gets better as more people use it together, not just for one person, that's a moat that has nothing to do with which model is being called.

A wrapper with none of those is just a feature wearing a company costume, and it gets squashed the moment the model provider builds the same thing natively. A wrapper with two or three of those stops being a wrapper at all, it quietly becomes its own invisible plumbing layer, with the model as just one ingredient instead of the whole product.

Where this is heading with AI

Here's the part I think is actually the most interesting: AI is turning into the biggest invisible plumbing layer we've ever built.

Right now, most people still experience AI the visible way, you open ChatGPT, you type a question, you get an answer. That's the equivalent of using Spotify: you know you're using it, you opened the app on purpose.

But that's not where this is going. The real shift is AI quietly slipping into the background of products you already use, the same way Stripe slipped into "pay" buttons everywhere without anyone needing to think about it. You'll stop "going to use AI" and start just... using things that happen to have AI running underneath, invisibly, doing work you don't see.

Traditional AI, already invisible

This is already happening with "regular" AI, models that predict, score, sort, or detect, not the kind that writes or creates anything new:

• Fraud detection in banking apps. Every time your bank blocks or approves a transaction in real time, there's almost certainly a model scoring the risk behind the scenes. You never see it, you just notice your card either worked or got declined.
• Spotify and YouTube recommendations. You don't open "the AI," you just scroll your homepage and the next song or video is already picked for you.
• Spam and phishing filters in email. Gmail silently runs AI to decide what lands in your inbox versus your spam folder. You just notice your inbox is mostly clean.
• Google Maps traffic predictions. When your ETA updates because of traffic, there's a model predicting congestion patterns from thousands of phones in real time.
• Duolingo's difficulty adjustment. The app quietly decides how hard your next lesson should be based on your mistakes. You just feel like the app "gets you."
• Photo apps automatically sorting faces and objects. Google Photos can find "all pictures with your dog" without you tagging anything.
• Customer support routing. Before you even talk to a human, AI often reads your message and routes you to the right team, completely invisibly.

Generative AI, becoming invisible too

This is the newer shift: generative AI, models that write, design, or create something new, slipping into products the exact same way:

• GitHub Copilot inside your code editor. You're not "using an AI app," you're just typing code, and suggestions appear inline as if your editor got smarter.
• Notion AI writing inside your notes. You highlight a paragraph and hit "improve writing," and new text appears in the same document. You never left Notion or opened a separate chat window.
• Gmail's "Help me write" feature. You type three words of an email and a full draft appears, right inside an app you already use for something else entirely.
• Canva's Magic Design and background removal. You upload a photo and Canva generates a cleaned-up version or a whole layout around it. You never see the model itself.
• Shopify's AI-generated product descriptions. A small business owner uploads a product photo, and the description, SEO tags, and marketing copy get generated automatically in the background of the store-building flow.
• Klarna's customer service AI. Klarna's support chat is largely powered by a generative model writing full, natural responses to refunds and order issues, without the customer necessarily knowing they're not talking to a person.
• Adobe Photoshop's Generative Fill. You select part of a photo and it generates new pixels to extend or replace it, right inside the tool you already use for editing.
• Salesforce and HubSpot auto-drafting sales emails. A salesperson clicks "draft follow-up" inside their CRM, and a fully written, personalized email gets generated, never opening ChatGPT separately.

In every one of these examples, the model isn't the thing being sold. It's quietly doing the predicting, writing, or designing behind a button inside a product whose name has nothing to do with AI at all.

That's the pattern worth watching: the most successful AI companies probably won't be the ones with the flashiest app. They'll be the ones that disappear completely into other people's products, the way Stripe disappeared into "pay" buttons and Visa disappeared into every card swipe. You won't open their app. You'll just notice that things got a little smarter, a little faster, and you won't be totally sure why.

The companies that get good at removing themselves from the experience, running invisibly, only showing up as a slightly better result, are probably the ones that end up mattering the most. The future of AI might not look like more chatbots. It might just look like plumbing.