by Anuj Garg
AI economics and engineering judgment.
A model can give you an answer. You still have to decide what to do with it.
What AI is worth building. What it costs. When to trust it.
A weekly letter for engineering leaders and builders who have to make those calls.
AI decisions involve more than model choice. There is the bill, the workflow, the failure you missed, and the person who has to check the result. The Token Diaries follows those details to the decision they change.
One free issue each week. Unsubscribe anytime.
Three questions worth asking
Is it worth building?
What problem are we solving, and does this design earn its complexity? From a single model call to a council of models, examine what each extra part contributes.
What does it really cost?
Look at the whole workflow: token spend, retries, latency, and human review. Connect the bill to a useful outcome.
When should we trust it?
Decide what to test, where to keep a human involved, and which failures would change the rollout decision.
Start with the first diary
The opening essay looks at how the same AI unit can mean different things to the people building, buying, and governing the technology.
🪙 Why I write this
I’m Anuj Garg, a computer science engineer near Mumbai with around 13 years in IT, including DevOps and release management. I build in Rust and Python and am exploring systems that bring several AI models into one decision process.
The Token Diaries is where I connect that builder’s perspective with a bigger question: what makes AI useful enough to justify its cost and earn our trust?
Get the next diary
One idea each week to bring into your next architecture discussion, budget conversation, or rollout decision.
One free issue each week. Unsubscribe anytime.