Five Subscriptions for a Handful of API Calls
Take a look at the SaaS bills of an average marketing team. There's a tool for £99 a month that schedules social posts. Next to it an SEO suite for £199 that tracks rankings. Then a reporting tool that pulls ad spend into a dashboard. A fourth for image variants, a fifth for email campaigns.
What these tools have in common: none of them owns the data it works with. The posts go to Meta. The rankings come out of Google's index. The ad numbers sit in ad accounts the company already controls. Every one of these tools is a layer between the company and an API the company could have been talking to all along.
Those intermediary layers had a clean economic justification for years. Not anymore. And the reason isn't "better APIs" – it's AI.
What the Middle Layer Actually Sells
The key thing to understand: a typical SaaS tool rarely sells a unique product. It sells saved integration work.
The Meta Marketing API is public. The Google Search Console API is public. The Stripe API is public. Any company with the right credentials is allowed to call them. What a tool like Hootsuite, an SEO suite, or a billing dashboard actually delivers is the work that sits between "raw API" and "usable workflow":
- OAuth flows and token refresh that nobody wants to build twice
- pagination across tens of thousands of records
- error handling for rate limits and short-lived fields
- a dashboard that turns raw data into something readable
- a scheduler that fires things at the right time
That's real work, and it used to be expensive. Which is exactly why it was rational for years to pay £99 a month rather than put a developer on a three-week in-house build that then needs maintaining forever. The middleman's moat was never the feature. It was the glue code – the invisible adhesive that turns an API into a tool.
Why the Maths Flips Now
Glue code is precisely the kind of work language models excel at. It's well-documented, well-bounded integration work that appears a million times over in the training data. No conceptual breakthrough, just diligent wiring of known building blocks.
Five years ago, "I'll call the Meta Marketing API directly" meant someone reading through 200 pages of docs, working out the OAuth dance, learning the SDK's quirks, building pagination and error handling by hand. That last mile was 80% of the effort of an in-house build – and therefore the entire reason you bought a subscription instead.
Today you describe what you want and the model writes the integration: pulls the docs, builds the auth flow, handles pagination and rate limits, refreshes the tokens. What's left is a short script that does exactly what you need – not 200 features of which you use three.
That's the real shift. The APIs didn't get better. The cost of calling them yourself collapsed. And with it goes the justification for any middleman whose only value was bridging that last mile.
Three Examples Where It Gets Concrete
Meta Marketing & Graph API instead of a social tool. A team using Buffer, Hootsuite, or Smartly is essentially paying for three things: schedule posts, manage campaigns, pull insights. The Graph API does all of it directly – publish posts on a schedule, create and budget ad campaigns, retrieve reach and conversion data. An AI-written script that drops every ad account's weekly numbers into a spreadsheet or your own dashboard each Monday is a matter of hours, not weeks. And it knows no artificial "5 social accounts on the starter plan" ceiling.
Google Search Console API instead of an SEO suite. A substantial part of what teams rent an SEO suite for is their own search performance data: which queries bring impressions, which pages are slipping, where the gap between ranking and click-through rate sits. That data belongs to the company and lives entirely in the Search Console API. Instead of viewing it through the grid of someone else's dashboard, you pull it directly and build exactly the report that answers your question – including the cuts no SaaS tool offers because they're too specific to your business. (What an SEO suite adds on top – competitors' backlink data, third-party keyword volumes – is a separate, larger dataset; more on that shortly.)
Python instead of Adobe Photoshop. Nobody needs a £60 Creative Cloud licence to resize 800 product images to uniform dimensions, stamp on a watermark, convert to WebP, or generate a batch of banners from a template. Pillow, ImageMagick, or OpenCV do exactly that in a script – programmable, repeatable, version-controlled. Where Photoshop is creative pixel-level work by hand, batch image processing is a job for code. And code that drives these libraries is something a model writes down reliably today, without your ever having read Pillow's API.
The list goes on. Calling the Stripe API directly replaces many a billing dashboard. Email through SES or Postmark replaces a good chunk of what Mailchimp charges for, at least for transactional mail. The pattern is always the same: where the tool is just a surface over an API you pay for anyway, the middleman has become vulnerable.
Where the Middleman Still Earns Its Keep
This is where it pays to stay honest – otherwise you draw the wrong conclusions and cancel subscriptions that are worth the money.
AI lowers the cost of building the wrapper. It doesn't lower the cost of running it. Writing a script that queries the Search Console API is trivial today. Making sure it runs reliably every day, that someone notices when it breaks, that the logic gets maintained when the API changes – that remains real work. That operational reliability is part of what good SaaS is paid for.
There are other cases where the middleman clearly wins:
- Proprietary data, not just an API. An SEO suite is more than a wrapper around Search Console – it has a vast crawl index of its own, backlink data, and keyword volumes no single company could gather. That's not a middleman, that's a product in its own right.
- Compliance and auditability. When a tool is the system of record, provides SOC 2 evidence, and has to be regulated, you're not buying a feature – you're buying accountability. No in-house script takes that on.
- Collaboration and self-service. As soon as non-technical colleagues work in the tool daily, manage permissions, build things together, the interface is the product – and a script helps no one who doesn't want to write code.
The rule of thumb: the more a tool generates data of its own, carries accountability, or has people working directly inside it, the less of a middleman it is – and the more the subscription is justified.
The Test: Wrapper or Platform?
Before you cancel anything, it's worth a sober look at each individual tool through four questions:
- Do we use less than 20% of the features? A tool where only a narrow slice gets used is a prime candidate for a focused in-house script.
- Is the underlying API public and stable? For Meta, Google, Stripe the answer is yes. For a tool whose value rests on a proprietary dataset, the answer is no.
- Is our usage programmable and recurring – or ad-hoc and human? The weekly report is screaming for code. The one-off creative decision on an image stays manual.
- Do we need to own the data? Where data sovereignty matters – and in more and more cases it does – direct API access isn't just cheaper, it's strategically better.
Four "wrappers"? Then you're currently paying for avoidable integration work. Four "platforms"? Then the subscription is money well spent.
The Takeaway
Every SaaS subscription is, at heart, a small tax on a capability you could own yourself. Most of those taxes were worth paying, because doing it yourself – wiring up the last mile to the API – was expensive and risky. That assumption no longer holds across the board.
The tools that hold genuine data of their own, carry accountability, or house people day to day are safe. But the pure middlemen – the layers between a company and an API it pays for anyway, Meta, Google, and Stripe foremost among them – are genuinely exposed for the first time in years.
The win isn't to reflexively cancel everything. It's the ability to tell a wrapper from a platform – and to switch off exactly those layers that do nothing but stand between you and your own API. Companies that draw that distinction cleanly spend less and move faster. Both at once.