i think after 'context engineering' (RIP 2025-2026), it's just 'compute engineering'

you've started asking the right questions to AI, your org knowledge is served over the right agent-facing filesystem/connectors, now it's really just how many model-tokens are you willing to spend for a given task

code review? that's important, lets spend $10/PR via 3 parallel opus subagents

internal project brief? eh, lets just gpt nano it