
Every moment in business happens only once. The next Bill Gates will not build an operating system. -
Peter Thiel (
Zero to One: Notes on Startups, or How to Build the Future
by Peter Thiel)
Five weeks into Antler, I’d finally hit my stride. Looking back at my initial strategy, it was simple, maybe a bit unconventional: become what I now call an
idea sampler
.
I told Richard (my coach in the early days of the Antler programme) exactly this before I joined. My plan wasn’t to immediately find the perfect team or chase the hottest idea. Instead, I’d talk to everyone, work with everyone, and deliver tangible value as fast as possible.
Some founders needed technical validation. Others just wanted someone to build quick scripts to scrape data (which is actually how I met my current co-founder). I’d spend 30 minutes here, a couple of hours there, just getting shit done.
In just six weeks, I’d worked with seven different teams. That pace forced me to do some unusual things to move closer to the right fit. Some teams didn’t really believe in what they were building. Some loved the
feeling
of working on a mission, but weren’t the right people for the mission or it wasn’t just the right time. And some were solving the wrong problems altogether.
I have a lot of empathy for every single founder in these categories but I do not need the market to punch me in the mouth if I can just leave the boxing ring.
Working across so many teams also forced me to confront something I’d been resistant to: fully embracing AI in my workflow.
I’d always been pro-AI in theory, but Antler’s pace demolished my mental block. I realized AI could handle tons of the grunt work I didn’t want to do anyway. It could write unit tests, generate boilerplate code, and even draft entire functions.
This freed me to focus on architecture and system-level thinking. And honestly, that’s what a CTO should be doing anyway.
When I think about what a CTO actually does, it feels less like writing code and more like translating. Taking the latest tech, shaping it, and showing the team how it can become real value.
At a very abstract level, we build tools that enable customers to be dramatically more productive. We come into an industry and ask:
what old problems could be solved 10x better with new technologies?
This question became a flashlight when things were a bit unclear at Antler. How can we make our customers 10x more productive while having them spend 10x less time using the same tools?
Recommended by LinkedIn
[H3 Startup Team Triangle
💡Michael Quatrini
9 years ago](https://www.linkedin.com/pulse/h3-startup-team-triangle-michael-q-quatrini)
[Thinking about starting a startup but don’t know how…
Mihir Deo
10 years ago](https://www.linkedin.com/pulse/thinking-starting-startup-dont-know-how-code-here-answers-mihir-deo)
[Your Code is no Longer Defensible: Lessons from Niall…
Philip Smith
4 months ago](https://www.linkedin.com/pulse/your-code-longer-defensible-lessons-from-niall-moran-philip-smith-fizaf)
I’ve been thinking about time-to-value a lot. At first I thought it was just about giving customers quick wins. Now I see it more as a
compounding value flywheel.
The best products feel useful right away, but the real magic is how that value quietly multiplies the longer people use them.
Pricing plays a role here as well. Should you really be the cheapest option on the market if you are truly providing 10x value? I’ve noticed that when you really create 10x value, you feel less pressure to be the cheapest. Probably even a proxy to confirm that you’ve won over the market and delivered on your promises. Price too cheap, and you may never know if you’ve actually made good on your value proposition.
What I’ve found, talking with others and testing myself, is that AI moves fastest when the context stays small. When you prune the context before feeding LLM it forces you to stay close to the code yourself while letting AI focus on deep reasoning instead of wasting tokens on noise.
My process has become simply:
research, plan, then execute with AI in tight loops.
I can sometimes jump straight to plan and execution except there are more custom/business logic I have to flesh out or nuances I have to take care of.
So yes, AI now does about half my work. But the trick isn’t asking it to do more. It’s keeping the context lean so it thinks harder and delivers sharper results.
Next week, I’ll break down my current setup for solving this in practice.
We are at an inflection point where new companies can fully take advantage of AI agents, at Souk,
we move at
Souk Speed from idea to prod, because of this so many processes don’t work for us; we don’t even write PRDs
.
We are even in the process of orchestrating our AI Agents to write unit tests as we push code to prod at light speed, review our PRs, even spin up entire modules in parallel.
If you’re a software engineer or ML engineer and you’re excited about using AI Agents to get 2-3x more done? we should talk!
Originally published in Blackbox Thinking on LinkedIn.