Two or three days ago, I was asking how to approach learning languages.

You listed many websites and applications that are helpful in your learning process to acquire a new skill.

I can bet you tried many solutions and have an overview of them, so I will ask a few more questions to observe this area:

I ve been thinking about where the line actually is with Kubernetes. When you re a small team, something like Render, Railway, Fly, or a managed cloud setup can take you pretty far. You push code, it runs, and nobody has to become the accidental infrastructure person.
Then the product grows. You have more services, workers, cron jobs, environments, traffic that doesn t behave nicely, maybe different scaling requirements. At some point someone says, We should probably move to Kubernetes before this becomes a mess. But moving early can create a completely different mess.

Now someone needs to understand the cluster, networking, autoscaling, deployments, observability, upgrades, Helm/YAML, and what happens when something breaks at 2 AM. On a five-person engineering team, that can very quickly become one person s unofficial full-time job. Waiting too long doesn t sound great either. If you eventually need Kubernetes, migrating when the existing infrastructure is already struggling is probably the worst possible time to do it.

So I m curious what the actual trigger has been for teams here.

Was it traffic? Number of services? Cloud cost? Needing more control over deployments? Hiring a platform/DevOps person? Or did you get quite far without Kubernetes and realise you never actually needed it?

For years, there s been one place where big-name tech products could enjoy a second life, albeit without most or all of the people that got them there: Bending Spoons.

The Italian company, which is valued at $11B and reports 300M+ monthly active users, acquired Eventbrite for $500M this week.

When we set up credits, one credit meant one generation. Easy to explain, and wrong.

A generation isn't one thing. The same request routed to a cheap model versus an expensive one can differ by an order of magnitude in what it actually costs us, and a retry after a bad output costs full price again. So a flat credit means the people doing simple work subsidise the heavy users, and the users you most want, the ones pushing the tool hard, are the ones losing you money.

I work with B2C products most of the time. I have experience with B2B in sponsorship.

But I can t compare it most comprehensively.

That s why I think some of you have more relevant experience and can compare your work in B2C and B2B.

AI pricing today doesn t fully reflect real cost.

Some reports and industry estimates suggest that real inference cost can be 5 10x higher than what companies currently charge, depending on model size, context length, and infrastructure load.

So the market is not really stable yet.
It s still in a growth phase, supported by subsidized pricing and scale.
And usage is already massive.

Most of you probably saw this report: