if you wonder what the difference is between working at big tech and startups, look no further than Google's AI offerings 😐

after getting such great results w/gemini 3.7 flash, i've dusted off my Gemini AI Pro membership. somehow, gemini 3.7 is not yet available in their Jules coding environment.

don't fall for the bait to use gpt 5.6 luna because the benchmarks show its performance is "pareto optimal"

it may perform similarly over a one shot assessment, but luna, terra, and sol are dramatically different in the output tokens and agent steps needed to finish the task.

Ignore the FUD on Gemini 3.7 Flash, it actually sits on the pareto frontier.

5.6 Luna is a much smaller model and while my team cut over production workflows to it, when you factor in raw token throughput, 3.7 Flash is in a league of its own.

openai's 80% price cut actually pareto mogs so hard its absurd

5.6 luna max is literally smarter than opus 5 low and 80% cheaper. with 90% prompt caching, sending a 100k repository map on every turn costs less than a single fresh 10k query.

If you have ever wanted to build an "AI chief of staff," I strongly recommend giving Dayflow a try.

It captures what you are doing throughout the day and gives you visibility into how you actually spend your time. Visibility is the prerequisite for designing high-leverage agent handoffs.

here's my AI model tier list as of rn as a gtm engineer at vercel.

this isn't just pure benchmark performance, but how I feel given cost, latency, reliability, usability, etc. also, just pure dopamine per turn. 👑 S TIER: 5.6 sol pro & Gemini 3.7 Flash.