I Use AI 100 Times Per Hour
Tunguz documents his personal AI usage patterns, finding he invokes AI approximately 100 times per hour across dictation, coding, and analysis tasks. He presents a cost-benefit analysis comparing Claude Opus 4.5 versus Sonnet 4.5, demonstrating that model selection significantly impacts both cost efficiency (86% savings) and performance (36% faster execution). The article argues for a hybrid AI strategy that uses cheaper models for routine tasks while reserving expensive models for complex analysis.
Metrics in this report
0.064USD per task
average
Autonomous task execution in Asana agent monitoring
0.009USD per task
average
Autonomous task execution in Asana agent monitoring
59seconds
average
Autonomous task execution time
38seconds
average
Autonomous task execution time
18words per call
average
Dictation transcription in email and analysis tasks
51calls per hour
average
Speech-to-text transcription during knowledge work
2456tokens per task
average
Input and output tokens combined
1797tokens per task
average
Input and output tokens combined