14.ai Dispatch #11
From Bellagio suites to system prompts in 48 hours
Dispatch #11 rolls the dice in Vegas, sneaks into a support conference, and somehow ends up running workshops between blackjack hands.
Recap: Prompting for customer support, live from Vegas at the Support Driven Conference
At this year’s Support Conference in Las Vegas, Michael took the stage to break down how to actually make AI agents useful in real-world support.
The session covered:
How LLMs work under the hood (and why it matters for prompt design)
Structuring system prompts so they’re clear, consistent, and easy to maintain
Keeping agents from confidently inventing policies your company never approved
And because it was Vegas, he left the audience with this closer:
“What happens in the system prompt, stays in the system prompt.”
🏝 Live from the Bellagio: Scaling support teams from Series A to Series C
In our latest workshop, Jen Burton (founder of Upper Righters) joined us for a deep-dive Q&A on building support teams for startups.
We covered:
When to make your first (or next) support hire
Balancing generalists vs. specialists as you grow
The most important early-stage metrics
Omnichannel vs. omnipresent support (and why it matters)
Thank you, Jen!
📄 Full recap here.
Workshop: Prompting for AI support agents
Your AI agent is only as good as the prompts it’s fed. Fuzzy, inconsistent, or overstuffed prompts? Your customers will feel it.
Join us for a session with Cyrus Nouroozi, CEO and co-founder of YC S24 startup Zenbase.ai (and former core contributor to DSPy, Stanford NLP’s most popular GitHub project), as we unpack what actually works when prompting AI for customer support.
We’ll cover:
Core principles for writing prompts that AI agents can follow reliably
How to avoid common pitfalls that lead to vague or off-target responses
Strategies for adapting prompts to different support workflows and contexts
Plus, a live Q&A where Cyrus shares real-world lessons from building and scaling AI-powered support.
📆 Wednesday, August 20 at 9 am PT
🔗 Sign up here.
Clippy: The OG hallucinator
The original AI agent. Never forget.
If you get this reference, you qualify for the “back in my day” discount
If you don't, here's a TLDR on Clippy
Born in 1997 as “Clippit,” part of Microsoft’s grand plan to make Office “friendlier.”
Looked like a paperclip, acted like a micromanaging intern.
Main skill: popping up uninvited to announce, “It looks like you’re writing a letter!” - even if you were making a grocery list.
Couldn’t remember your name, preferences, or the fact that you’d told him to go away 47 times.
He was built for first-time users… which made him unbearable by your second or third document.
Survived years of mockery, parodies, and passive-aggressive memes before Microsoft finally retired him in 2007.
Clippy walked so modern AI assistants could run…
📊 Poll of the week: What was the worst escalation to “Tier 2 support” in history?
📌 In case you missed it
We gave the full sensory guide to “arriving in Paris”… plot twist: it was actually Las Vegas.
The AI agents have arrived, and they’re checking into the Bellagio.
Marie and Michael were featured by Finta as our very first founder spotlight.
If you're looking for AI chat support, book some time here with the 14.ai team.
💼 Cool jobs at 14.ai (and friends)
In the market for something new? Or just here for the salary benchmarking?
🛠️ Founding Engineer @ 14.ai (San Francisco)
💻Technical Support Engineer @ Junction (EDT/EST, US East Coast Only)
⚙️ Senior Manager, AI Support Enablement @ Ancestry (Remote)
📚 What we’re reading & listening to this week
Marie: Working in Public: The Making and Maintenance of Open Source Software, Nadia Eghbal
Maya: The Story Behind Owner’s Pivots to a Billion-Dollar Business, First Round Review
Michael: Design Patterns for Securing LLM Agents against Prompt Injections, Beurer-Kellner et al., 2025
Ugo: The Sales Playbook For Founders, Startup School
🫡 That’s all for now.
May your prompts be crisp, your agents well-behaved, and your Vegas metaphors just slightly over the top.
Until next time: bet on clarity, double down on good support, and never let your system prompt spill the tea.
 | A guest post by
|