1. Core Haven Labs
2. Blog
3. Two AI providers back to back

Why we put two AI providers back to back

30 August 2026 · 3 min read · Technology

An AI provider has outages, like every service on the internet. For a business owner who gets an appointment request at 22:30, that must not make any difference. That is why our systems always have two providers back to back.

One interface, two providers

Auto-Reputatie and BarbrFlow do not talk directly to one AI company. They talk to one layer of our own that knows exactly what a reply or a classification must deliver. Behind that layer sit two providers: in Auto-Reputatie OpenAI with Anthropic as backup, in BarbrFlow Claude with OpenAI as backup.

If the first returns an error, a timeout or an unusable answer, the same request automatically goes to the second. The customer notices nothing; at most the reply takes a few seconds longer.

A backup that is the same as the first choice is not a backup. That is why they are two different providers.

An outage must never mean silence

With a WhatsApp receptionist, the worst that can happen is that a customer hears nothing back. That is why BarbrFlow is built so that a failed classification never results in silence: if neither provider manages to understand the message, the customer still gets a proper confirmation that the message has arrived, and the owner is brought in.

With Auto-Reputatie the opposite principle applies: a reply that fails the check is not published. There, doing nothing is the safe outcome, and it stays a draft for the owner.

Everything traceable

Per reply we record which provider, which model and which version of the instructions were used, plus how much text was processed. That way, when an owner asks, we can look back precisely at what happened, and we measure per version whether replies get better or worse. For that, Auto-Reputatie is run against a fixed set of 52 test reviews on every change.

Also read: why an AI reply never goes live unchecked.