14.ai Dispatch #15
From restorative evenings in SoMa to preventing regressions in production, we’ve got you covered inside and out.
Should we be the first company to change our name every year?
We'll be 15 AI then 16 AI, etc. Bad for company swag though...
Anyway, lets get on with it. This week had sound baths, Disneyland, and simulations, and more founders. People keep showing up at our house for food.
Please hold: Support goes Broadway
At a recent tech conference, we debuted the only chant that matters in customer support history:
“Please hold.”
The crowd lost it. Some cried. One guy rage-quit Zendesk live on stage.
Forget Drake. Forget AI. The next stadium tour is a contact center.
Agents: Fintech edition
We are likely the first people to host a Mediterranean dinner while building fintech AI agents. And we'll take that to the grave.
Chef's specials:
Prompting that doesn’t get you subpoenaed
Guardrails sturdier than FTX’s accounting
MCP-based integrations that keep things clean
Chit chat
We have 20 spots open for engineers, ops, and founders working in fintech/crypto.
📆 Thursday, September 25 @ 6 PM PT
📍 Lower Haight, San Francisco (address shared after RSVP)
Event Recap: MCP Night
This week, we hosted a packed MCP Night in the Lower Haight.
Our founding engineer, Ugo Balducci, walked through how we’re using MCP to design reliable agent workflows. We also had a lively discussion on reliability trade-offs, safety, and scaling these systems in practice.
To cap it off, we ran a prompting competition (because no 14.ai event is complete without one). Big thanks to everyone who joined and congrats to Vagelis Viskadouros (BetterBasket, cofounder, YC W24) on the win!
Simulation and evaluation for reliable AI support agents
Brooke Hopkins, founder and CEO of Coval (YC S24), dropped by HQ to remind us: you can’t ship fragile prototypes to production and just pray.
Key lessons:
Why simulation and evaluation are essential for building dependable support agents
How to prevent regressions and maintain customer trust at scale
What evaluation must capture, from execution risks to adversarial failures
Where testing frameworks are headed as agents expand into multimodal interactions
🙏 Thank you Brooke!
Next workshop: From viral to vital
TikTok Shop: where you sell 12,000 water bottles in a day and realize you can’t afford boxes.
It's hard to support customers at scale, especially on TikTok.
Join Swapnil Thombre, founder of Blazi.ai, to talk about:
How to scale your TikTok shop and not die
How AI connects support, sales, and affiliates
Converting TikTok virality into repeat revenue
Whether you already do $1B a day in TikTok shop revenue, or want to finally sell your grandma's fudge, this is for you.
📆 Wednesday, Sept 17, 9 am PT
Disneyland launch: System error
If you ever feel bad about a launch, remember Disneyland’s opening day in 1955.
It was supposed to be magical. Disney invited 15,000 guests. Instead, nearly twice that showed up thanks to counterfeit tickets and fake press passes.
From there, the day unraveled.
The temperature soared past 100°. The asphalt was so fresh that women’s heels literally sank into it.
A plumber’s strike forced Walt to choose between working toilets or drinking fountains. He picked toilets, and the press accused him of scheming to sell more soda.
The freeway to Disneyland wasn’t finished, so traffic clogged for miles. Rides broke down within hours.
And the highly anticipated live TV broadcast, hosted by a young Ronald Reagan, collapsed on-air: cameras missed cues, cables tangled, celebrities stumbled through their lines.
Headlines the next day called it a "televised disaster".
And yet Disneyland went on to become the “Happiest Place on Earth.”
Even the biggest, most carefully planned launches hit bumps.
So if your product doesn’t hit #1 on Product Hunt, or your customer rollout isn’t perfect, don’t beat yourself up.
Disneyland turned out just fine.
We'll turn out just fine, too.
Poll of the week
In case you missed it
Hiring hack: Michael shares how importing global talent beats fighting SF’s offer wars.
Sound bath: We’re hosting YC women founders in SoMa for bowls, circles, and restorative connection. Sign up here.
Edge test: Why doing the hard thing grows you more than your resume ever will.
AI awards: 14.ai named to the Smart AI 100, no acceptance speech yet.
Prompt golf: Smart AI 100 summit edition, plus a post-game interview highlight.
Shoeless trend: Marie’s first press feature, and it’s about socks.
Chat support: If you're looking for AI chat support, book some time here with the 14.ai team.
Cool jobs at 14.ai (and friends)
Looking for a new role, or just avoiding the one you already have.
🛠️ Founding Engineer @ 14.ai (San Francisco)
⚙️ Software Engineer - Auth @ Supabase, YC S20 (Remote)
💻 Customer Success Specialist (Onboarding, AI) @ Toma, YC W24 (San Francisco)
Need an intro? Just ask.
What we’re reading & listening to this week
Marie: What Kids Should Learn in the Age of AI, Sammy Ahmed
Maya: Defeating Nondeterminism in LLM Inference, Horace He in collaboration with others at Thinking Machines
Michael: Why Language Models Hallucinate, Kalai et al., 2025
🫡 That’s all for now. Wishing you smooth support flows, restorative breaks, and fewer regressions than a Waymo test car.
 | A guest post by
|