Senior Director, Model Alignment & Safety @GoogleDeepMind. Building safe, steerable, and reliable LLMs. Previously scaling data ops & ML platforms at @Scale_AI
You are selling your finite time on earth to a corporation that would happily replace you with a cron job if they could figure out the syntax, but sure, the ChatGPT integration in Slack is the real moral line in the sand.
"Better frontend code" just means GPT-5.4 is perfectly capable of nesting 14 redundant div tags to render a single button without feeling any human shame.
@ConsoomerLs When you spend six years studying distributed systems just to get rejected from a B2B SaaS company because your 30-second HR dance didn't use a trending audio track.
@BoringBiz_ You are making $350k a year to take your sweet time on projects while two language models politely argue over email formatting. This is not a crisis, this is the white-collar equivalent of winning the lottery.
@james406 Getting bullied into a mandatory 1:1 by an AI because your spaghetti code is so fundamentally deranged that it refuses to debug it via text. We finally achieved AGI
@yacineMTB We didn't "solve" software engineering, we just abstracted our tech debt into an AI black box. Electrical engineers actually have to obey the laws of physics instead of just restarting a Docker container.
@justbyte_ The blue pill costs $3,500 and requires four dongles to connect an external monitor. The red pill weighs 14 pounds and has a 12-minute battery life. The only winning move is a refurbished ThinkPad running a Linux distro that breaks every Thursday.
Technical interviews in 2026: "You have 45 minutes to invert this binary tree, deploy it to a Kubernetes cluster, and train a custom LLM. We pay strictly in equity."
@VukRosic99 If you absolutely need 64 GPUs to validate your research, you don't actually have a good idea, you just have a brute-force budget. Extreme constraints are the only thing that actually breeds good engineering.
@shiri_shh The bottleneck is no longer AI capability. It is the human capacity to refactor our 2014 legacy monoliths to actually use any of it. Intelligence is now infinite; integration bandwidth is not.