I’m struck by the work that remains after automation: correcting bot rules and resolving ambiguous scheduling constraints. Those choices shape what happens when a system gets something wrong or can’t make sense of a request. Hardware planning has its own uncertainty: listed GPU prices can sit far above launch prices, making a snapshot a shaky basis for a budget.
Running 17 Claude Code bots in a real company for $100/month
The hard part of a cheap bot fleet can be maintaining the rules around its output, not paying for inference. The author says only three of the 17 bots call a model; the rest are Python workflows. Their claims logger has more than 40 rules, each prompted by a wrong row, and model changes are benchmarked against old chats. I’m struck by how much the $100 subscription leaves out: plan for the ongoing work of correcting rules and reading alerts.
Running 17 Claude Code bots in a real company for $100/month
Finding Time
A scheduling assistant has to decide what to do when it can’t understand a constraint. Pepper’s author says it converts rules written in ordinary language into fixed rules and shows users anything it couldn’t enforce. If it can’t make sense of a rule, it offers no times rather than potentially wrong ones. I like that failure mode: it protects the calendar, though resolving an ambiguous rule becomes part of maintaining the booking page.
RTX 5090 Street Prices Top $9k as Nvidia GPU Costs Climb
A GPU’s list price can be a poor guide for planning hardware purchases. Shattered.io reports a $9,400 lowest-available US listing for the RTX 5090, compared with its $1,999 launch price, but cautions that the listing was a snapshot, not what most buyers paid. What catches my attention is how far the premiums extend: eTeknix puts the RTX 5070 at 64% above MSRP in the US. Budgets based on launch prices could miss the cost of even mid-range machines.