idk maybe this is obvious to people already working in bigger teams, but the AI coding tool cost thing feels like early cloud all over again.
Everyone keeps saying tokens are getting cheaper, which is true, but then somehow companies are still freaking out about AI bills. And I think the reason is pretty simple: people are treating these tools like normal SaaS seats when they are really more like metered infra.
Like with a normal dev tool you kind of know the cost. X users, Y dollars per month, done. But with agentic coding tools one small request can quietly turn into a bunch of model calls, context loading, tool calls, retries, verification, more retries, etc. From the user side it looks like “fix this bug” or “write this function” but underneath it may have done a whole mini workflow.
And then there is the other cost which I feel people don’t talk about enough: reviewing the generated code. Sometimes the code works but it adds weird duplication, misses existing abstractions, or creates stuff that someone has to clean up later. So the bill is not just tokens. It is also review time + maintenance + future tech debt.
Not saying these tools are bad btw. I use them too and they are obviously useful. But it feels like the industry is moving from the fun phase of “look what this can do” to the boring phase of “who is paying for all these calls and did this actually ship anything useful?”
Curious if teams are actually tracking this properly yet. Like cost per PR, cost per resolved ticket, cost per workflow etc. Or is it still mostly hidden under “AI productivity” and vibes.
[link] [comments]