# Marcus Delgado, Senior PM (AI Tooling) at Tangram Labs (280 people) — read of multi-llm-debate-platform, June 9 2026

> Nine years in product, last four in the AI tools space. Currently paying for Claude Pro and ChatGPT Plus simultaneously and feeling stupid about it.

## How I got here

Searched "compare claude gpt side by side" on a Sunday afternoon because I was prepping a vendor eval deck and wanted something I could screenshare with my team. Google surfaced this in the second result. I clicked because the URL slug was descriptive and I was tired of landing on blog posts from 2023 that rank AI chatbots like consumer toasters.

## What I clicked first

"Ask one question. Watch Claude, GPT-4, and Gemini respond. Compare how they think." That's a clean promise. I stopped there for a second because it's concrete and I immediately understood the value prop. Most tools in this space open with "the future of AI-powered collaboration" and I close the tab reflexively. This one didn't do that.

## Where I paused

The pricing table. Free tier is 10 debates a month, Pro is $15 for unlimited. That's a reasonable ladder. I'd pay $15 for this if it saves me 20 minutes of copy-paste comparison per week. I was actually leaning toward clicking "Start Pro Trial" until I scrolled down and hit this:

"Honest disclosure: we don't have live customers on this idea yet."

That sentence stopped me cold. Not because honesty is bad. Because the word "idea" recontextualized the entire page I'd just read. I scrolled back up. The waitlist button. The "hundreds of engineers" social proof. The model comparison table. None of it is a live product. It's a pitch for a product that hasn't been built yet, mixed in with what reads like a live SaaS landing page. I genuinely didn't know what I was looking at anymore.

## What I distrusted

The Wishdeal Factory scoring section. "68/100 Adoptability. $-18,144 Year-1 take-home. 1 in 8 Meaningful-success odds." Those numbers are presented with authority but I have no idea what methodology produced them, who Wishdeal Factory is, or why this is on a product homepage at all. "Fermi" estimates are mentioned twice but not explained.

Also: "Join hundreds of engineers, researchers, and thinkers." Hundreds. Not "2,300 signups." Not "used by teams at Notion and Stripe." Hundreds is the number you say when you don't have a number worth saying, and they put it in the social proof line right above the waitlist form.

The model comparison table rows like "Wild connections" under GPT creativity and "Grounded novelty" under Gemini feel like vibes dressed up as specs. Nobody benchmarked these. Someone wrote them.

## What would convince me

A single screen recording of the actual interface running a real prompt. Not a diagram of how it works, not a table of model personalities, just 45 seconds of the thing doing the thing. If it runs all three in parallel and visually highlights divergence the way the FAQ describes, that's genuinely interesting and I'd understand it immediately.

Alternately: one specific before/after example. "We ran this prompt: [X]. Here's what Claude said. Here's where GPT diverged. Here's what we learned." That would make the "Explore Disagreement" section real instead of hypothetical.

## What I'd ask in an email reply

1. The page says "join the waitlist" but also has a "Start Pro Trial" button. Is there a live product or not? I'm confused about what I'm actually signing up for.

2. The divergence highlighting feature sounds like the core value prop. How does it work technically -- are you doing semantic diff, manual tagging, something else? Because "highlights the divergence" could mean a colored box or it could mean something genuinely useful.

3. The Wishdeal Factory numbers on the page -- the Year-1 Fermi estimate and the 1-in-8 odds -- are those for potential builders considering licensing the idea, or are they somehow meant for me as a user? I genuinely cannot tell who that section is talking to.

## Verdict: on-the-fence

The core idea is good and the hero copy is unusually clear. But the page is trying to do two things -- sell a product to users AND sell a business idea to operators -- and it does neither cleanly. I'd read a follow-up email if the subject line answered whether this is live or not.

---
*Memo by skeptic persona, generated 2026-06-09. Studio breaks own self-grading loop.*