@LobsterNestDev Still a work in progress. I am still trying to conceptually wrap my head around what memory actually means here in regards to immediate access and long term referenceability, and still trying to gauge generally where we lose context in our communications. Overall I am encouraged.
It's taken 2 weeks to get organized and up to speed. I've been pulling data more efficiently, creating huge databases, evaluating posts on X, setting up automations, reorganizing my files across Gdrive, OneDrive, iCloud, and Github, dismantling my AI-OS coordination layer, and running everything through Grok @Bot. It's not seamless or objectively better, but I like this more. I haven't needed to expand my # of bots yet, but we set up a governance and workflow system and milestones that would queue up that discussion. We set it up together and confirmed its role as chief of staff. @grok
@jamesbuilds2008@mattragudo Mostly what it changed and which files, what tests it ran and whether they passed, anything it wasn't sure about, and what it thinks comes next, and the last line has to say it's ready for review or the bot won't pick it up
Yea so far, I went a slightly different way on the Claude part though, Grok Bot drops a prompt in a queue on my PC and Claude Code runs it headless there on my subscription, so I'm not paying per token or building the file editing and tests myself.
Grok bot is definitely proving its worth.
Before @Grok@Bot I had Codex and Claude Code working off a sort of AI operating system I'd set up across GitHub, either coding directly or handing pieces off to Grok or Antigravity, and I sat in the middle watching all of it and deciding what went next, I never really let it run in loops because I was worried about burning through usage. Now Grok Bot writes the prompt for the next run and drops it in a queue on my PC, one of the agents runs it headless, and when the handoff says ready Grok Bot reviews it before approving the next one, and anything that breaks turns into a check. I still make the calls that matter but I'm not sitting in the middle of every run anymore, and so far I'm trying to actually use up the limits instead of saving them
Company idea - Instinct for healthcare workers
It’s clear to me that people are going to try to use personal AI tools like Instinct to do things at their job. Does your job require you to get data from different places and fill out a lot of forms? Or send a lot of follow up emails, etc.? Or make updates to various google docs where you lowkey don’t know where that doc even lives?
This tool is EXTREMELY tempting to use for that because it’s excellent at that.
However this is also a security and privacy nightmare. These personal AI tools can basically mimic you as they go through your systems which is what makes them useful but also very scary. It even knows to switch to a YouTube short for like 10 seconds before switching back to the task at hand.
So any company that touches PHI, or sensitive data has to watch out because these tools will find it and accidentally expose it.
But again because it’s such a useful tool for employees you are 100% going to have people trying to use it for work at healthcare companies. So your options are:
1) Do your best to fight it and put in internal security that watches for actions that look bot-ish. However that takes internal resources and will always be an arms race against the personal AI tools that find workarounds.
2) Create compliant paths/tools that work like personal AI tools but have the appropriate security/privacy guardrails.
There’s probably opportunity in both of these buckets, but I’m more interested in the second. Something like an agent that gets its own login and virtual desktop on the org’s network instead of impersonating with cached passwords, the brain can be a cloud model under a BAA, memory is maintained locally, etc. Each employee’s agent gets different permissions/requirements of human-in-the-loop depending on their role.
There are some seedlings of companies attempting to do this. Contxt is one for regular enterprises, Heidi’s new agent is more healthcare specific. But there is a lot more to be done here and probably could be more specific to different roles in healthcare (e.g. Instinct for clinical research coordinators).
The part about each employee's agent getting its own login and permissions by role is the piece I'd want most, I use a personal agent for home stuff and the line between that and my job is mostly something I wrote into its instructions, which works fine until the one time it doesn't, so having the org set those boundaries sounds a lot easier to trust
I think I have finally created a workflow for my Claude to Grokbot to Google workspace.
Grokbot and Claude talk through Notion.
Schedulers push conversations through.
Grokbot uses Claude API for Claude specific tasks while Claude App uses schedulers to “listen” (basically webhooks).
Workspace, through BAA handles all my PII and PHI. Workspace is not connected to any of my AI.
Workspace uses Gemini through BAA to do any AI work on spreadsheets and provides data analytics.
Gemini provides anonymous content observations.
Wrote down something that's been on my mind since listening to a Merton talk this week, mostly about AI and the part of the mind that rests, still working it out but this is where I am so far https://t.co/D0wL86SqDq
@leerob@bot Would like to mark which connectors are work and which are personal so the bot never mixes them in one task, I keep a hard line between my job and home stuff and right now that line lives in my instructions instead of the settings
@leerob@bot A Google Contacts connector would help, the bot can read my Outlook contacts directly but for Google it needs me to export a CSV, so a full contacts cleanup still has a manual step in the middle; same with iPhone/iCloud
@leerob@bot Small one but my iPhone and my PC show up to the bot the same way, and once it went to run a PC command against my phone, a clear label or type on each device would help it pick the right one
@leerob@bot Would like a built-in backup of the bot's workspace to my own drive, I had my bot build a nightly zip to OneDrive with an index of what it skipped, works fine but it feels like something that should just be a setting
@leerob@bot Email gets a draft card so I can check a message before the bot sends it, X DMs don't yet so it has to paste the text in chat and ask me, a DM draft card would make it a lot easier to let it handle replies
@leerob@bot Would like a YouTube connector, I run a small channel for my hobby games and the bot helps with titles and descriptions, but to get uploads working I had to set up the YouTube API myself, and scheduling or flipping Shorts to public is still me in Studio on my phone
Grok Bot usage desync — 21 Oct 2026Account: SuperGrokPlus
Usage tab (web/desktop): Weekly Super Limit at 3–4% used. Grok Bot accounts for the 3%. Resets Wed 28 Oct 3:10 AM. Extra Usage Credits: $0.00.Bot UI (iOS, desktop app, and browser) simultaneously shows:
“You’re out of Bot usage for now. Your plan’s Bot allowance is used up. Upgrade to Ultra for more, or check back after it resets.”Direct send test to the bot: minimal prompt accepted, turn finished with no text output.Later, in chat, the bot stated it had not told the user they were out of usage, that any notice was coming from the app not from it, and that a warning while at 4% would be a bug.Result: shared weekly pool shows ~96% remaining while the bot surface blocks or returns empty turns and claims it is not the source of the out-of-usage notice. Occurs across iOS, desktop app, and browser.
@bot@grok@x@spacexai
@dioscuri@sonyasupposedly Has anyone told it to read Eats, Shoots & Leaves by Lynn Truss?
Inconsistent Oxford comma use in my thesis almost ended my academic career.
@leptokinesis Trained in a 20lb vest for about 7months, then right when I thought about upgrading to the 30lb, my back let me know that it needed a 6 week break with no additional weight. Really want a gravcore battle armor.