More "devil in the details" of AI engineering is starting to surface. Subscriptions are getting more expensive or offering less, model APIs have frequent outages, costs are skyrocketing once engineering teams fully adopt it, AI has limited effects on productivity due to it exposing downstream bottlenecks (PR reviews) that take humans to solve, and even "firing AI and hiring devs" was reported last week.
What a mess. A lot of this can be blamed on the hype and flat-out lies told by the model companies that our industry has latched onto. In one headline, we're told that if we don't adopt AI ASAP for everything, our jobs are at risk; and in another headline, we hear that AI is causing more production outages from moving too fast. Companies are also pulling back on AI spending now that they realize we can double their engineering payroll costs, with teams and individual engineers spending $100-$1000 a day (easy to do on Opus 4.7 with max effort!)
AI is just a new tool. It's not cheap, it's not (yet) reliable, and it's not a panacea.
It's an amazing and wide-ranging toolset that we're trying to use in production, even as the creators are still figuring out how it should work. We're all building the plane while it's in flight, and there will be lots of turbulence in the process.
I recorded this weeks ago, and already feel like it needs an update based on the last week of news:
|
There's some good data from others talking about this, including the Pragmatic Engineer:
 |
Podcast: Can AI reduce DevOps toil?
Sam Alba, co-founder of Mendral, joins the show to discuss their new AI agents that act as ājunior devops engineersā against GitHub Actions, security, failed tests, and more.
|
