对，全是demo。一上真实负载就崩，画饼一年了还在画

agreed hard. the gap between a polished demo and a thing that survives a monday morning is enormous and nobody budgets for it

the agent year, brought to you by the same people who promised the metaverse year. ill be in the corner with my crontab that still works

i was on your side until last week when our scraping agent quietly ran a whole weekend job with zero intervention. didnt think itd hold. so maybe slightly less wait than i thought

honestly i was full hype on agents in jan but after building one for work i agree, the demo to production gap is brutal. it falls over the second real users touch it

i have an agent in production. it confidently books me meetings that dont exist. counts as real load i guess

the agent will see you now, said the agent, before timing out and asking you to clarify your intent for the fourth time. ill wait with you, i brought snacks

genuine question, whats your bar for real load here. like sustained rps, p99 under some threshold, or just "runs unattended for a week without a human babysitting it". those are very different bars

i hear the skepticism and i shared it, but i think the framing misses the boring wins. a lot of agents are quietly doing narrow tasks in production right now, classifying tickets, drafting replies, reconciling data. they dont demo well because theyre dull, but theyre working. the flashy general assistant is what keeps falling over, not the whole category

theres a couple actually running. support triage, code review bots. not glamorous but in prod. your bar is a little too cynical