What Building Infrastructure Taught Us About Trust
Years of building infrastructure taught me that trust is earned quietly, honesty about limits beats big promises, and locking people in is a slow death.
Years of building infrastructure taught me that trust is earned quietly, honesty about limits beats big promises, and locking people in is a slow death.
Open-weight models let you run capable AI on hardware you control. Here is why I think that matters more than any leaderboard, for privacy, cost, and independence.
We chose to run our own AI infrastructure instead of building on closed APIs. Here is the case for owning your stack, and the honest cost of doing it.
VPS, cloud, or dedicated? What each one is, what it costs in money and effort, and how to pick based on traffic, resilience, and budget.
The AI energy cost is treated as an inconsequential while the power demand is at the size of some countries and why efficiency, not scale, is the future.
The first time something I built went down at 3 a.m., I learned more about trust in one night than in the year I spent building the thing.
Nobody applauds infrastructure when it works. That is the deal you sign when you decide to run things other people rely on. When a server hums along, a database returns a query in nine milliseconds, a network route holds under load, the reward is silence. No thank-you note arrives for the outage that did not happen. The only feedback loop you get is the bad one, and it is loud, and it is at 3 a.m. So you start to wonder why you bother chasing a standard nobody will ever see.
Then you sit with it for a while. And you realize that silence IS the applause. The whole point of infrastructure and trust is that the better you do your job, the less anyone has to think about you. Dependability is not a feature you ship. It is the absence of reasons to leave.
I have built a lot of things from scratch. Some lived, some died, most taught me something I did not want to learn at the time. And the lesson that kept coming back, in different costumes, was always about trust. Where it comes from. How fast you can burn it. Why almost everyone in this industry gets it exactly backwards.
Here is what nobody tells you when you start building for other people. Trust is not built in the demo. It is not built on the pricing page or in the launch tweet or in the slick onboarding flow you spent three weeks perfecting. Those get you the first yes. They do not get you the second year.
Trust is built in the boring parts. The migration that did not lose a byte. The bill that said exactly what last month's bill said. The support reply that came from someone who could read a stack trace instead of a script. The Tuesday afternoon when nothing happened, again, for the four-hundredth Tuesday in a row.
This drives me a little mad about how the industry talks about itself. We celebrate the launch. We post the funding round. We hand out awards for the thing that is six weeks old and shiny. And we have almost no language for the quiet, unglamorous discipline of keeping a promise to the same person for a decade. That discipline is the entire job. Everything else is marketing.
When I think about who I trust as a customer, it is never the company with the loudest promise. It is the one that has been quietly correct for so long that I stopped checking. That is the bar. Be so dependable that people forget to worry about you.
You want to know the fastest way to lose someone? Promise them something you cannot hold.
The instinct, especially early, is to say yes to everything. Unlimited this. Five-nines that. We can do it, sure, absolutely, whatever you need. It feels like confidence. It is the opposite. It is a loan against trust you have not earned yet, and the interest comes due the first time reality disagrees with your sales copy.
I have come to believe the most powerful thing you can tell a customer is what you are NOT good at. "This plan is genuinely fine for what you are doing, you do not need the bigger one yet." "We do not do that, here is who does." "This will be slower than you want, and here is why, and here is the honest tradeoff." Every time I have told someone the unflattering truth about a limit, the relationship got stronger, not weaker. Counterintuitive. Also the most reliable thing I know.
On saturated consumer hardware, you do not buy small-model quality with a faster framework. You buy it with better data and a better training objective, with numbers to back it.
Because honesty about limits is a costly signal. Anyone can promise the moon. Only someone who plans to be around in five years tells you where the edges are. When a company is straight with you about what it cannot do, it is quietly telling you that it expects to keep you by being good, not by having oversold you in the moment. People feel that. They may not name it, but they feel it, and it compounds.
Don't get me wrong, ambition matters, and you should build toward more than you can do today. But there is a difference between "we are working on it" and "yes, we already do that." One is a roadmap. The other is a lie with a delivery date.
There is a word for what happens when a company stops earning your loyalty and starts engineering your captivity. Cory Doctorow coined it in 2022, and it was named Word of the Year by the American Dialect Society in 2023. The word is enshittification, and once you see the pattern you cannot unsee it.
Doctorow's definition is brutal and exact. Platforms, he wrote, are "good to their users; then they abuse their users to make things better for their business customers; finally, they abuse those business customers to claw back all the value for themselves" (Wikipedia). You know the arc. The app that was great until it wasn't. The service that got worse the moment leaving got expensive. The platform that stopped competing for you the day it figured out how to trap you.
And here is the thing that should terrify anyone running infrastructure. The trap usually works. For a while. The metrics look fine. Churn is low because leaving is painful, not because staying is pleasant. Egress fees that cost a fortune to move your own data out. Proprietary formats that do not export cleanly. Contracts that auto-renew into a price three times the one you signed. Each one buys a little more captivity at the cost of a little more goodwill, and the spreadsheet says it is working.
It is not working. It is a slow death with good quarterly numbers.
Because the day a customer realizes they are staying because they are stuck, the relationship is already over. They are just waiting for an exit that does not hurt. And the moment a competitor makes that exit cheap, they are gone, and they tell everyone why. Lock-in does not retain customers. It launders resentment into a temporary revenue number and calls it loyalty.
So what is the alternative? Build something they want to keep using, and then make it easy to leave.
I know how that sounds. Make it easy to leave is not the line you expect from someone who runs infrastructure for a living. But it is the only version of retention I respect, because it is the only one that keeps you honest. If your customers can walk out the door tomorrow with all their data, in a format they can use, with no penalty, then every month they stay is a month you earned. There is no captivity propping up the number. Just the work.
Think about the company everyone points to when they talk about long-term customer trust. Costco has sold its hot dog and soda combo for the same $1.50 since the mid-1980s, roughly forty years, through every wave of inflation that touched everything else. Earlier this year, asked whether that would ever change, the CEO said flatly: "The hot dog price will not change as long as I'm around" (Fortune). That is not a hot dog strategy. That is a promise made visible, held for decades, at a cost the company chose to eat. People do not trust Costco because of the hot dog. They trust Costco because the hot dog proves that some prices, some promises, are not on the table for renegotiation. That is what durable trust looks like. Boring. Consistent. Slightly stubborn about the things that matter.
That is the standard I want to run infrastructure by. At Carpathian we build and run our own US-based cloud rather than reselling a hyperscaler, and the part I care most about is not the hardware. It is the contract underneath it. Predictable pricing you can plan around. Direct access to the people who built the thing. Your data being yours, portable, no hostage situation. We are a young company, and I am not going to pretend we have a forty-year track record we have not lived yet. But the principle is the whole point: keep people because they want to stay, never because they cannot afford to go.
Back to that night. The one that taught me more than the year before it.
What I learned was not how to prevent the outage, though I learned that too. What I learned was that trust is not a thing you have. It is a thing you are constantly, quietly redepositing, and the balance only ever shows up when you overdraw it. Every uneventful day is a deposit nobody notices. One bad night is a withdrawal everybody remembers. The math is unfair. It is also the job.
So I stopped chasing the applause that infrastructure will never give me. I stopped measuring the work by who noticed it. The win condition is not a thank-you. The win condition is that years from now, someone is running their business on something I built, and they have forgotten my name, because they have never once had a reason to think about me.
Build the thing people want to keep using. Tell them the truth about its edges. Never lock the door behind them. Then earn the next year, and the year after that, in the boring parts where nobody is watching.
Trust is not what you say at the launch. It is what is still true at 3 a.m.