We continue to observe how new KPIs are driving IT management crazy. Recently I wrote about the birth of Tokenmaxxing and corporate madness with AI dashboards at Disney, where developers are pushed to completely abandon handwritten code for the sake of pretty statistics.
Now the baton has been picked up by Amazon.
Inside the company, they rolled out the MeshClaw tool — an orchestrator for AI agents that can call internal systems, deploy code, and respond in Slack. Managers set a target: at least 80% of developers must use AI weekly. And, of course, they added leaderboards for the volume of tokens burned.
The outcome is predictable. Smart lazy people do what they have to do with dumb metrics. Engineers started automating non-critical tasks and running agents idle to inflate their tokens before performance reviews. Management, of course, swears that the statistics "don't affect evaluation," but everyone understands. Goodhart's law in full effect: when a metric becomes a target, it ceases to be a good metric.
Meanwhile, Google has officially allowed the use of AI assistants (its Gemini) in technical interviews.
The pilot was launched for the code comprehension stage. Now you will be evaluated not on how you mentally balance red-black trees, but on your "AI fluency." Interviewers look at prompt engineering skills, validation of LLM hallucinations, and debugging of someone else's generated code.
The reason is clear: Google openly admits that 75% of new code in the company is generated by neural networks. Asking textbook algorithms in interviews has become pointless when the real work of an engineer has turned into managing a horde of hallucinating agents.
In general, all interviews are shifting towards open-ended tasks and system design, even for juniors. And that's a good thing.
Comments
0No comments yet.
Sign in to join the discussion.