Spend and memory limits across three AI agents, and workarounds
A video flags limits in Grok Bots, OpenAI dots and Jev 1.13, then proposes a chain: agent proposes, code does math, Jev picks from a reordered list, human approves every send.
THIS IS F*CKED UP: YOUR AI AGENT'S SPEND LIMIT ISN'T A LIMIT. THE VENDOR WROTE IT DOWN. 3 AI VENDORS, 3 FAILURE MODES, ALL ON PAGE 2 OF THEIR OWN DOCS. Grok Bots: all your bots share 1 cloud computer, with the same files, browser sessions and logins. The spend limit is not a hard stop either, a bot can keep running past it mid-task. The fix is 1 job per bot, with the low-risk jobs going live first. OpenAI dots: disconnect an app and your dot still keeps what it learned from it, and saved memories only go with a full reset. The fix is 2 or 3 apps per dot, because connect 20 and that is 20 apps it cannot forget without wiping everything. Jev 1.13: it reads instructions word for word, cannot count, cannot compare 2 dates and leans toward the first option in a list. The fix is to do the math in code, then reorder the options and ask again. The chain that survives all 3: agent proposes → code does the math → Jev picks from a reordered list → a human approves every send. All 3 vendors say their agents make mistakes even when they follow your rules, so anything that sends gets approved by a human, 0 exceptions. The honest part: these limits come from each vendor's own docs as of today and they change, so recheck page 2 before you hand over a single login. You connected 20 apps to an agent because the onboarding screen made it look easy. One-person desk: the bots do about 90% and you only give the human yes on every send. Steal this: 3 apps max, 1 job per bot, math in code. Ready to copy, and send this to your Grok Bot and ask which of your jobs are low-risk enough to go live first. Save this. Follow for more one-person desks.
This post quotes another post: x.com/donatelloxpro/status/2106803135071035580