The Anti-Vendor Manifesto

A stack of subscriptions

A software company can rent authentication, email, analytics, search, storage, monitoring, jobs, and the database before it has a customer. Each line item can look reasonable. Together they are the product.

You build around their API. Their outage is yours. Their price sits in your unit economics. Their terms constrain how you operate. The invoices add up. Switching costs matter more.

Fine, until the capability sits at the center of the product, the integration runs deep, and there is no path out.

What you are actually buying

A lot of a vendor's value is operational: reliability, support, documentation, compliance, someone awake at 3am. That work is real. The technical core is often smaller than the brochure.

They sell “we can operate what you would rather not.” Fair thing to sell. It is a price, and a price needs a comparison. The thesis is why more of those comparisons now come out differently.

What coupling costs

The subscription is the visible price. The deeper costs do not require the vendor to behave badly.

Availability. When vendors sit in the request path, their failures become yours. The tool meant to report an outage can be inside the same outage.

Control. Every integration uses a contract you do not own. Endpoints change, limits move, product tiers disappear. HashiCorp's Terraform license change was a legitimate business decision that still forced users to make one of their own.

Data. Each service that receives user data adds another retention policy, security boundary, and compliance relationship. Your duty to users does not shrink as their data spreads.

Maintenance. Credentials, retries, error handling, mapping, tests as the API changes. That code is part of your product even though its schedule is not yours.

What we own

We prefer to own what is central to the product and bounded enough to operate. A smaller implementation for the cases in front of us, not the whole market a platform has to cover.

What ownership buys is optionality. An implementation you own does not renew, reprice, or deprecate an endpoint on a Tuesday. It can still fail. The path to a repair stays yours.

Some capabilities cannot be made independent, because the authority or the physical network is the product. Payment processing depends on financial networks. Public domain registration and certificate issuance depend on recognized authorities. No amount of engineering removes those relationships.

Buying also makes sense when a customer requires a provider, when you cannot operate the capability safely, or when another priority matters more. The answer can still be “buy.”

The interesting question was never how many vendors are in the stack. It is how many of them you could not leave.