# Ravi Sundaram, CTO at Briefly — read of LLM Provider Failover, June 13 2026

> 13 years in backend infra, last 3 building an AI meeting-notes product on Claude. We run about 40k API calls a day. I have Twilio on my resume so I have opinions about reliability layers.

## How I got here

Googled "claude api outage fallback production" at 7:15 AM on the Caltrain, the morning after Claude went down for 19 minutes and killed our live product for a customer demo. Found a Reddit thread, someone linked this in a comment as "a tool that handles exactly this." Clicked through expecting a product landing page. I was not prepared for what I found.

## What I clicked first

The headline "Never lose an API call" is exactly the right hook. That's the phrase I would have typed into Google myself. The specs table underneath looked like a real product: instant failover, zero code changes, cost transparency. I was leaning in. "Drop-in replacement for Claude client libraries. Add resilience without refactoring." That's a specific promise from someone who understands the actual pain.

## Where I paused

The scoring widget. "57/100 Adoptability. $-17,855 Year-1 take-home (Fermi). 1 in 8 Meaningful-success odds." I stopped and read that three times. I thought it was some kind of investor-facing internal dashboard accidentally exposed. Then I read: "Honest disclosure: we don't have live customers on this idea yet. We shipped the strategy package; you ship the customer conversations."

That's when I understood what this actually is. This is not a product. This is a packaged business idea for sale. The whole page is selling me a dossier for building this thing, not the thing itself. That realization took me embarrassingly long to reach, which tells me the page is structured in a way that front-loads product signals and buries the actual offer.

## What I distrusted

The "financial upside: 1/10" score is listed as a concern, right next to "uniqueness: 9/10." They're scoring their own idea a 1 out of 10 on financial upside and then offering to sell it to me for $99. I genuinely don't know what to do with that. Either the score is honest and they're selling me a bad bet, or the score is a hedge and they don't stand behind it. Neither reading is good.

Also: "the Wishdeal Factory scores every idea against 10 Adoptability axes." I don't know what this factory is. I don't know who built these scores. The Fermi estimates have no sourcing. "$-17,855 Year-1 take-home" implies I'd lose money. They say that plainly. And then ask me to pay to proceed.

## What would convince me

If this were an actual infrastructure product: a traffic graph showing failover latency (sub-100ms matters here), a named customer who's gone through a real outage, and the actual SDK wrapper in a public repo. That's all I need. The problem is real. A real solution would sell itself in one look.

If this is staying what it is (an idea dossier): a single example of a previous idea they sold where someone actually built it, shipped it, and got paying customers. Not a testimonial. A URL to the product and a rough MRR range. Show me the model works before I test it with my $99.

## What I'd ask in an email reply

1. Is there a working prototype anywhere, or is the code starter in the $99 tier literally a scaffold I still have to build from scratch?
2. The scoring says financial upside is 1/10. You're the ones who built the score. Do you believe it? Why would I build something your own model says has low upside?
3. Have you or anyone at Wishdeal actually deployed LLM failover in a production environment? What did the failure mode look like when the primary provider went down mid-stream?

## Verdict: dismissive

I came here looking for a product to plug into our stack by Thursday. This is a kit to start a company, priced like a Gumroad download, with a 1-in-8 success rate disclosed right on the homepage. Wrong page for where I am today.

---
*Memo by skeptic persona, generated 2026-06-13. Studio breaks own self-grading loop.*