Drew Bailey

Token allocation has been a disaster. Uber announced $1,500/month per employee for AI tools, the result of months of uncapped spend with no way to track what any of it produced. The engineers who were actually delivering just got the same cap as the ones who weren't. 𝗛𝗼𝘄 𝗨𝗯𝗲𝗿 𝗿𝗼𝗹𝗹𝗲𝗱 𝗼𝘂𝘁 𝗔𝗜: 1: Announce you're all-in on AI, celebrate as the spend climbs, and call it a culture win. No tracking, no way to know what's making it to production. 2. Four months later, the CFO wants to know what changed. Engineering says it's working but can't point to anything specific, and the monthly bill just doubled again when the new model dropped. 3. Cap everyone. Across the board, because ROI is unclear. That cap means your best engineers, the ones who’re genuinely shipping faster than ever, just lost their budget. So did the engineers spending just as much and producing nothing that made it to production. Two completely different situations, one decision, and your top performers paid for it. 𝗛𝗼𝘄 𝘁𝗼 𝗽𝗿𝗼𝗽𝗲𝗿𝗹𝘆 𝗱𝗼 𝘁𝗼𝗸𝗲𝗻 𝗮𝗹𝗹𝗼𝗰𝗮𝘁𝗶𝗼𝗻: 1. Measure output per AI dollar, per person. Use an analytics platform (like Weave) so you know the return that each dollar of AI your are spending. Some engineers produce 3 units of output for every dollar they spend. Others produce 0.5. 2. Coach the 0.5 before you cut the 3.0. Look at their prompt history, their model usage, their workflow, and find where the disconnect is. 3. Give the 3.0 more budget. Your best reps get your best leads. It’s the same logic. AI is intelligence doing work inside your org and should be managed like any other resource. Measure the productivity, measure the cost, look at whether the ROI is there, and coach up or out. These decisions have major impacts on employees and shouldn't be taken so lightly.