For years, the advice was simple: move fast and break things. Then the market got tired of broken things. Today, I think the real question is not fast or safe. It is: fast at what, and safe where?
I lead Nexcar, a company that builds fraud detection technology for automotive transactions in Mexico. We work with insurers, lenders and other players in industries that took decades to move from paper to PDF. And this is exactly where the speed dilemma becomes interesting.
Three years ago, building a new workflow for a client took months. Today, with AI models and good tooling, a small team can ship in days what used to take a full quarter. With one of our main clients in the insurance sector, we went live in production in three weeks. In another moment, that same project could have taken months.
Your competitor knows this too. If you wait to have the perfect version, someone else will be in the meeting with a working demo. Execution is still the first moat.
But execution alone can be copied. What is harder to copy is what you learn after you execute: which patterns repeat, where things go wrong, what the client really needs versus what they asked for. Over time, that knowledge and data becomes a second moat. Speed gets you in the door. What you learn once inside is what keeps you there.
The Part Nobody Likes to Talk About
The first cost is people. AI models now update every few weeks. Every new version promises better accuracy, lower cost, longer context. The honest truth is that a startup cannot test every model properly. You end up choosing between two bad options: stay on an older model you trust, or switch fast and hope nothing breaks in production. Your engineers feel that pressure every week, and burnout is not a bug in the system. For many teams it is the system.
The second cost is security. And 2026 has been a brutal reminder. In Mexico, a group called Chronus leaked 2.3 terabytes of data in late January, affecting more than 36 million people and reaching federal institutions like the tax authority and the social security institute, as well as state and local governments. In the following months, other public agencies reported new attacks. Outside Mexico the list is just as long, from a cyberattack that caused a global disruption at Boston Scientific in August to ASOS confirming a breach in October after customers saw “HACKED” notifications in its app.
When you move fast with sensitive data, every shortcut is a door. And someone is always checking the doors.
Slow Industries: The Advantage and the Bottleneck
Industries with a long history of slow transformation are a gift for startups. Processes are manual, and the gap between what is possible and what exists is enormous. You don’t need to be ten times better. Sometimes being twice as fast is enough.
AI also made something possible that used to scare us: personalization. One client wants results by WhatsApp, another wants a batch file, another wants everything inside their legacy system. A few years ago the answer was “this is our product, take it or leave it.” Today a startup can say yes and still keep one platform underneath.
But the same slowness that creates the opportunity also limits it. You can build a feature in one week and then wait six months for approval. Procurement, legal, compliance, IT committees, systems that only accept files from 2009. Your product is often ready long before the industry is. So the startup lives in two speeds: internet speed inside, institutional speed outside.
A Real Test: Launching to Consumers
This quarter we launched our first product for consumers. Since day one we have only sold to companies, and going to consumers forced us to face the dilemma in a very direct way.
In B2B, a client tells you something is wrong and you fix it together. In B2C, a person just leaves. So the pressure to launch fast and learn from real users is huge. At the same time, a consumer’s trust is easy to lose and very hard to win back. There is no “we’ll fix it later” when a real person is on the other side.
The debates inside the team were very honest. Wait for every edge case or launch and iterate? Add one more feature or protect the core first? In the end, we launched with things we know are not perfect: some screens, some copy, some flows. Those will change many times in the coming weeks. What we did not rush was the core: security, payments and access. That part took longer than we wanted, and I think it was the right call.
That is the rule we try to follow: fast on the surface, slow at the core. Interfaces, workflows and reports can change every week. Security, payments, access and any decision that affects someone’s money get reviewed twice and tested longer, even if that makes us “late.”
So, Fast or Safe?
Both. Fast where mistakes are cheap and reversible. Safe where mistakes are expensive and permanent. Speed without iteration is chaos. Iteration without a stable core is risk.
In slow industries, the winner is not the fastest runner. It is the one that is still running when the industry finally arrives.
