Opportunity → Agent-First Database
Framework → Decision Type Vs. Time
Tool → Cignara
Trend → Searches Ending Without Clicks
Quote → Just Ship The Best Product
🔗 Houck’s Picks
My favorite finds of the week.
Fundraising
Growth
Katelyn Bourgoin on what you should build instead of an audience (Link)
Dan Martell on how to build an AI agent that doesn’t torch your company (Link)
ICYMI
Alice Bentinck (Founder/CEO at Entrepreneur First) on how EF picks founders before they have an idea (Link)*
Ayman Al-Abdullah on the evolution of a CEO (Link)
Andrew Gazdecki on what startup founders actually need (Link)
Gabriel Jarrosson on the age of the 40-year-old founder (Link)
Kathryn Wu on how founding hiring is a science (Link)
Tyler Denk shares a very underrated habit for every founder & CEO (Link)
Opportunity: Agent-First Database
Agents don’t need flashy UI or feature-laded UX.
They just need function.
Much like how websites are increasingly two-sided, with a human-centric experience built around brand and messaging and an agent-centric experience built around ease-of-navigation, there’s an interesting opportunity to build a database in a similar fashion.
Imagine Notion-level UX (don’t get too mad about this if you hate the block page format) but with an agent-first background layer.
You’d need strong permission controls (for both, agents hallucinate but non-technical users who suddenly get direct DB access would be more destructive otherwise) but if careful guardrails were a first-principle here there’s an exciting product to build.
🧠 Framework: Decision Type Vs. Time
You’re probably familiar with Jeff Bezos’s concept of a one-way door (irreversible decision) vs a two-way door (reversible).
If not, here’s him briefly explaining it:
WHOOP’s founder, Will Ahmed, explored this a bit regarding how to think about time in relation to the type of decision.
Will says that irreversible decisions should be made at the latest possible moment it’s still responsible to be considering it (aka you’re not delaying something critical).
Whereas, for most decisions (which are reversible) you should just go with your gut as quickly as possible and move forward (adjusting if necessary later).
The tradeoff of how long it takes to make the decision is not worth it unless the decision is irreversible.
I use this same principle myself basically all the time. It’s the same idea as saying no contract is real until the money’s in the bank (aka the transaction is irreversible).
🛠 Tool: Cignara
There’ve been a lot of AI-for-customer-service attempts ever since ChatGPT first launched. Most have had issues, but YC-backed Cignara claims to be “hallucination free.”
I can’t personally attest to that, but they are built for enterprise-level customers as a first-class citizen, so you have to assume a claim like that would be vigorously tested by deal evaluators.
If it isn’t too laden with enterprise-level bloat (as a YC company, you’d expect some level of being able to avoid that), it could be worth testing yourself.
📈 Trend: Searches Ending Without Clicks
Remember when you used to use Google to find answers to questions?
I’m joking… somewhat.
These days when you want information, you probably go to an LLM unless you know you want to actually look at a website after your search. Then, Google still wins.
It’s increasingly obvious that people are preferring this no-click experience.
a16z shared that, interestingly, not only is this occurring but it’s been occurring since well before LLMs were commonly used.
While high quality LLM summaries being on Google’s results page have clearly accelerated this, other surfaces in Google’s results like knowledge panels and featured snippets were giving people what they need as well.
Google’s becoming less of a list and more of an interface.
So, contrary to what you may have thought in the early days of ChatGPT, LLMs will likely end up being a net-positive for Google rather than a Google-killer. These no-click searches keep users within Google’s ecosystem rather than sending them out to publishers’ sites.
It’s the publishers who feel the pain, and need to adapt to instead showing up in LLMs directly.
💬 Quote: Just Ship The Best Product
I keep seeing what I can “shoehorned AI” in legacy tools.
Shoehorned AI is when you’ve built a tool that works, but now that LLMs exist, you try and fit it into what your product already does to appear to be “an AI product” and capture user interest.
Inevitably, the shoehorned version in your product is both A) worse than AI-native new products and B) degrades the experience of your non-AI product.
Even good ones do this!
No, I don’t want to use Notion AI, Slack AI, Zapier AI, or even AI Spotify playlists.
I don’t think I’m in the minority on this.
But the product managers at these companies and many others are likely some combination of genuinely excited about working with AI, looking at it as a career-advancer, and getting downward pressure from leadership to implement their “AI strategy.”
The lesson for founders is that your users don’t care about “your AI strategy”… they care about how well your product delivers on its promise.
Certainly, AI can help with that in many cases. But it’s your job to discern where usefulness ends and shoehorning begins.