Thoughts on the Software/AI Selloff

[Smorgasbord1]

@SlowAndFast

Since they are a large model provider, I estimate that they would have spent up to $30M per year with Datadog, before switching

Reports are that OpenAI is spending $150M+ with DataDog each year.

@SlowAndFast

So, the bottom line is that this situation is actually pretty much standard in the software industry. Do you want to go full DIY or something in between (that is for example Chronosphere), where you still need to do the major lift of extracting intelligence from your observability data, or do you want to focus on getting ahead in this AI race and pay someone like Datadog so your own engineers can focus on getting you to your actual goal?

AI changes this equation dramatically. For instance, Anthropic is using Claude Code to create 90% of Claude Code.

I see no reason why OpenAI, or any of the “LLM Hyperscalers” (my term), can’t use their product to first instrument the code and then a) provide the dashboards they want, followed by b) analyzing the problems and suggesting, if not actually authoring, fixes.

I know very little about the space, but considering there’s already an Open Source telemetry gathering standard, it would make sense for these LLM companies to use that for instrumentation, then first use a commercial product like Chronosphere, later transitioning to its own dashboards, created using itself.

How many engineers at OpenAI would you estimate are today focused mostly on using DataDog to find/fix issues? Pulling one out to create similar dashboards and alerts using Codex seems like a small pull to me.

I believe it’s a question of when, not if. OpenAI and Anthropic are going after software development first, because it speeds up their own development. Next, they’ll look at what else they use for operations and go after those. And then they’ll look at other verticals. Actually, they’ve already started some verticals, like contract review.