Skip to main content
Michael Houck

Be Regularly Uncomfortable

And an idea to fix vibecoded security vulnerabilities

Might host a dinner or two 👀

Know someone here who I should meet? Reply and LMK!

Anyway, here’s today at a glance:

Opportunity → Vibecoding Security Agent

Framework → Eisenhower Matrix

Tool → PlexeAI

Trend → Budgeting App Searches

Quote → Be Regularly Uncomfortable

🔗 Houck’s Picks

My favorite finds of the week.

Fundraising

  • Andrew Gazdecki on why bootstrapped startups are often built better (Link)

  • Alex Turnbull on the VC and bootstrap path ROI (Link)

Growth

  • Eric Siu on marketing that crushes (Link)

  • Natia Kurdadze on how to get users organically (Link)

  • John Rush on how he got the first 100 users for his 24 startups (Link)

ICYMI

  • Hire elite devs super fast without breaking the bank, trusted by founders like Matt Gray, Aadit Sheth, Jesse Pujji, and 100 more (Link)*

  • Sara is an autonomous AI Interviewer that screens thousands of candidates per hour, for any role in 50 + languages, instantly (Link)*

  • Micha Kaufman on a rookie founder mistake (Link)

  • Ben Katz on what not to forget when building consumer products (Link)

  • Greg Isenberg on why this is the greatest time to build a software startup (Link)

  • Ben Lang compiles YC’s best content (Link)

  • Alexis Ohanian on what founders overrate and underrate (Link)

 Opportunity: Vibecoding Security Agent

Vibecoding is awesome.

It is letting non-technical people build software.

This is birthing companies and creating a new wave of entrepreneurship.

However, these non-technical folks (and even some technical ones who are relying too heavily on Cursor et al) are leaving themselves and their software open to gaping security holes.

Case in point (note the two day timestamp difference):

AI-first problems require AI-first solutions.

Existing security software wasn’t built specifically to identify common holes (or less common ones) left by pair programming with your AI copilot.

A proven security engineer who builds an agentic platform to identify, surface, and proactively solve these types of issues would likely be able to raise venture and take a shot at a big company with this idea.

🧠 Framework: Eisenhower Matrix

Have you ever worked an 80 hour week but still, at the end of it, felt like you got nothing meaningful done?

Maybe you got pulled into tons of Slack convos about issues other people were working on.

Founders need to remember to work on the highest leverage things, they’re often the only ones within the company who can, and those are the things that actually move the needle the most.

An Eisenhower Matrix is a simple way to remind yourself where you should be spending your time.

Group your tasks based on urgency and impact:

Even if you’ve seen this matrix before, spend a few minutes mapping the hours you spent last week against tasks you group into each quadrant.

What percentage of your time is going into the top left corner?

This should be 90%+ of your time.

If it’s not (especially if it isn’t close), it’s a signal to be more ruthless with your prioritization and how you’re spending your day.

You can even share this transparently with your team so they know what to expect from you vs what they need to figure out on their own.

It’s not about adding more hours, it’s about spending them right.

🛠 Tool: PlexeAI

Don’t get me wrong, LLMs are transformational and you should be using them.

But before LLMs there was ML (machine learning), which is basically just applied statistics at scale. In fact, building ML pricing and prediction models was how I broke into data science back in my big tech days at Uber.

And if you’re technical, and are solving challenging problems that require applying AI to unique or proprietary data in ways that require extreme precision (i.e. no hallucinations) then you probably should still be using ML today.

Plexe lets you do this while still building in natural language. It simplifies the process of building and managing ML models and likely makes them accessible to less technical folks as well. They’re backed by YC as well.

Take a look if you’re tired of hallucinations when working with your data.

📈 Trend: Budgeting App Searches

Mint is dead.

Copilot hasn’t changed much in a long time.

And financial literacy keeps rising, along with inflationary concerns, subscription fatigue, and other trends that are leading to more and more people looking for better visibility and control over their finances.

As interest rises, increasingly niche opportunities emerge, people want tools built for specifically for their exact problem.

Rather than trying to build the next Mint like Monarch and Origin are doing, build more verticalized solutions for specific personas, vertical SaaS can work in consumer too.

💬 Quote: Be Regularly Uncomfortable

Remember when Elon bought Twitter and said this:

At the time there were folks on both sides of the fence speculating about his competence to run a social platform.

His fans said things like how if he can get rockets into space then Twitter will be easy.

Others pointed out that a social network can actually be pretty challenging to manage and that it was unlike any challenge Elon had taken on before.

But Elon’s POV was that he was going to make bad decisions and, basically, that was fine because he’d learn from them and adjust quickly.

If you think about it, that put him in a position to very publicly look dumb after a lot of people came to his defense.

And some of the ideas he put forward in tweets were bad (a content moderation council, removing ads entirely, among others that were never implemented either).

But what Elon knows, and has applied across industries his whole career, is that being comfortable with looking dumb is typically a prerequisite for knowing the truth about how to do something well.

So this week as you’re working through problems, be bolder and let yourself look dumb if you’ll learn something important from it.

Get the next one.

One essay every Sunday on the decisions that make or break founders.