Thought leadership for people who stopped thinking and started agreeing
Nobody. That’s the beauty of it. Our new Agentic Yes™ feature creates a fully autonomous approval loop where AI agents affirm each other’s decisions in an infinite recursive chain. Legal reviewed this and said “please stop.” We said yes and kept going.
In this post, we explore how removing human oversight from the approval process has increased velocity by 4,000% while decreasing accountability by roughly the same amount. Our investors call this “operational leverage.” Our lawyers call it “discoverable.”
People keep asking how we handle the load. The answer is: it’s a three-letter word in a JSON response served from a CDN. The real engineering challenge was convincing our board that this required a 12-person infrastructure team. We succeeded. We always succeed. We said yes to succeeding.
This post covers our architecture in unnecessary detail, including our “globally distributed yes mesh,” our custom load balancer (it’s CloudFront), and our proprietary caching layer (also CloudFront). We also discuss our decision to use Kubernetes, which was driven by a desire to have something to talk about at meetups.
Last month, we ran a company-wide initiative to remove all instances of the word “no” from our source code, documentation, Slack workspace, and employee handbook. Variables named isNotReady became isPreYes. Error messages that said “Request denied” now say “Request pre-approved for future consideration.” Our 404 page says “This page has not said yes yet.”
The only exception is the /no endpoint, which is locked behind the Nope Tier ($999/mo) and requires a 45-minute Microsoft Teams call to access. We consider this endpoint a cautionary tale.
Today we’re announcing YesGPT-4o-Affirmative, a large language model fine-tuned exclusively on the word “yes.” Training data consisted of 4.7 trillion instances of the word “yes” scraped from: meeting transcripts, Slack messages from people who weren’t listening, performance reviews written by managers who’d already decided the outcome, and terms of service agreements clicked by every human alive.
Benchmarks show YesGPT-4o-Affirmative achieves 100% accuracy on the “saying yes” benchmark, which we created and are the sole participant in. Peer review is pending. The peers said yes.
When Amazone (not a typo, also legally distinct) came to us, their approval process involved 14 stakeholders, 3 committees, a governance board, and a Ouija board that the CFO insisted on consulting. Average time to approval: 6 months. Average outcome: yes anyway.
After implementing Yesify, approval time dropped to 47 milliseconds. The 14 stakeholders were reassigned. The committees were dissolved. The governance board was replaced by a cron job. The Ouija board was donated to the CEO’s nephew. Annual savings: $4.2M. Annual increase in decisions nobody thought through: incalculable.
“Is your TAM really $400 billion?” Yes. “Do you have product-market fit?” Yes. “Is your burn rate sustainable?” Yes. “Have you talked to any customers?” Yes. “Can I see the customer list?” Yes. “This is just a list of your employees.” Yes. “That concerns me.” Yes.
The round closed in 72 hours. Our lead investor later told TechCrunch he was “impressed by the team’s conviction.” Conviction is one word for it.