Integrations that actually work — a field guide

If you spend a week evaluating AI tools, you'll notice a pattern. Every landing page claims 100+ integrations. Every product demo walks through six. Every customer story uses three. Those numbers are not lying, but they're also not telling you what you need to know.

This is the field guide we wish existed when we started building Vezra.

The three kinds of "integration"

Not all integrations are the same. In reality there are three distinct tiers, and knowing which tier you're looking at predicts whether the integration will save you time or cost you a weekend.

Tier 1 — the real ones. The product has first-class support for the tool. Auth is handled. Common actions are wrapped in tested functions. When the provider changes their API, someone on the product team fixes it. These are the integrations you can count on.

Tier 2 — the proxies. The product exposes a generic "HTTP request" action and lists every API-accessible tool as an integration. Technically true. Practically, you're on your own — you have to read the provider's docs, figure out auth, handle errors, and maintain the calls as the API drifts. A list of 900 tier-2 integrations is a list of 900 homework assignments.

Tier 3 — the aspirational ones. Mentioned on the site, not shipping. Sometimes has a "request early access" link. Sometimes just there for SEO.

When someone says "we have 500 integrations," it's almost always 30 of tier 1, 400 of tier 2, and 70 of tier 3.

Why the numbers feel dishonest

It's not that anyone's lying on purpose. It's that the incentives are bad. Everyone in this space is measured on integration count. Adding a tier-2 proxy takes a day. Adding a tier-1 integration takes two weeks. If you're racing to fill a grid on a landing page, you ship tier-2.

We did this briefly. We added a handful of proxies because we wanted the logo wall. Within a month, support tickets told us the real story: users were hitting the tier-2 integrations, thinking they were tier-1, and getting lost. The logo wall was hurting us.

We took the tier-2 ones down. The integrations page got shorter. Satisfaction went up.

Where Vezra draws the line

An integration is in Vezra's catalog only if all of these are true:

• We handle OAuth (or the equivalent) on your behalf. You click a button, you sign in, we take it from there.
• We maintain a typed set of common actions for the tool — not just "make an API request," but real operations like "create a Shopify order" or "reply to a Gmail thread."
• We test those actions against the live provider on a schedule. If they break, we hear about it before you do.
• The action count on the integration's page reflects reality, not ambition.

Every entry on our integrations page meets this bar. If it's listed, it works. If something we want to build doesn't meet the bar yet, it's not on the page at all.

But what about the tools you don't list?

Here's the thing that makes this position workable rather than limiting: Vezra agents each run on their own computer with a real browser. If your tool has a web interface and a login, your agent can use it — connector or not.

Is this as clean as a first-class integration? No. You'll give the agent credentials, and you'll point it at the right workflows, and occasionally the UI will change and the agent will adapt. But the baseline capability is there. Your agent can use any SaaS tool a human can use.

The difference is that our integrations page only lists the tools where we've done the work so you don't have to.

What this means for you

When you evaluate an AI platform, look at the integrations page and ask three questions:

1. How many of these handle auth for me, end to end?
2. How many have actions I could demo in the next five minutes?
3. What's the plan when the provider changes their API?

Most landing pages can't answer these questions without losing 80% of their logos. The ones that can are the ones worth building on.

Our job, going forward

The integrations catalog is going to keep growing. We're adding one or two a month, slowly, by our bar. We'd rather ship fewer integrations that work than hit a landing-page number that doesn't.

We'd rather you tell a friend that the integration you needed just worked than scroll through a logo wall and wonder.