Building software is essentially free now. Good engineers are 10x-ing their output, and that without touching a single line of code. If entire products ship in weeks, what can actually remain defensible?
I recently talked about this with Maciej (Matt) Małysz from Inovo VC on my podcast. He invests at pre-seed and seed in Central & Eastern Europe, so he thinks about this constantly.
The code isn’t the moat
Faster code means faster products. Sintra, one of Inovo’s portfolio companies, went from $0 to $12M ARR in 12 months providing AI agents to small businesses. Maciek says “$3M ARR in 6 months” is becoming the baseline for the best companies in their portfolio. Teams I work with can prototype in hours—but so can everyone else.
In our conversation, we kept coming back to two things: product and distribution. Deep domain understanding, UX that compounds over time, network effects—and the deeper you integrate into a customer’s workflow, the harder it becomes to switch. Product and distribution are of course tightly connected. Build something that embeds deeply into a customer’s workflow, and it becomes hard to rip out. Build something hard to rip out, and it starts to spread on its own.
Did you 10x too?
The popular narrative says non-technical founders should be thriving—they can finally ship without a CTO. Maciek flipped this on me.
“The technical founder just got 10x stronger over the last year. Did you in your non-tech?”
He’s right—the gap didn’t close. Engineers are 10x-ing their output with AI, and this has accelerated dramatically since around December 2025. If your co-founder can prototype in hours what used to take weeks, the pressure to match that can feel uncomfortable. Non-technical founders should be looking at agents too. I already use them well beyond code: agentic workspaces for writing proposals, managing projects, producing my podcast, even home admin. This is still early—everyone’s figuring it out as we go. But if 2025 was the year agents became reliable for software engineering, 2026 is shaping up to be the year they do the same for everything else.
Integration depth as the new defensibility
Maciek backs domain-specific solutions that handle entire workflows end-to-end—not a tool that does one piece, but the full job to be done. Switching costs from integration depth, not code complexity.
Take Victor, a Slack bot that understands company context, connects to internal systems, and does real work through voice notes. The more you use it, the more it learns about your company—and the more valuable it gets.
Agents talking to agents
I keep thinking about what’s next. I’m betting on personal AI assistants (I’m building one). Maciek mentioned the next layer: agent-to-agent interfaces—services, protocols, and products redesigned for agents rather than humans.
“There’s a new world where all the interactions are not designed with human in mind, but with agent in mind.”
Andrej Karpathy built an agent that controls his entire home—lights, HVAC, shades, security cameras—by talking directly to the APIs of six different smart home systems. No apps, no UIs. Just exposed endpoints with an agent as the intelligence layer tying them together. I can’t stop thinking about this.
Right now, we’re squeezing agents through interfaces built for humans. That won’t last. The companies that redesign for agent-first consumption—APIs over UIs, structured data over dashboards—will have a real edge as this plays out.
For the full conversation, listen to this episode of Hidden Layers:
Interested in bringing these practices to your team? I run hands-on workshops on AI-assisted development—including Vibe Coding for CEOs. Book a call here.