obsessed with the future of finance.
Joined June 2021
- Tweets32.6K
- Following2.3K
- Followers3.9K
- Likes122K
Pinned Tweet
Your grok bot limit isn’t counting messages
It’s counting the work your bots do.
That’s why some people burn through 100% of their usage by day 3, even when they barely chatted
The usage cost isn’t in what you type, it’s in what runs behind the scenes
Here are 5 places usage builds up, and how to fix each one.
>
1. Autopilot
this is usually the biggest drain
a routine that fires every 15 minutes runs 96 times a day. that’s 672 runs a week, from a single routine.
most jobs are fine running once or twice a day.
fix: list every routine across every bot, ask whether it needs to run that often, and cut the frequency or pause it.
>
2. Imported bots
they bring their own routines
so when you import a bot, its routines can switch on right away, before you’ve even used it.
fix: after adding any bot, check your roster for duplicates. Pausing is enough. You don’t have to delete anything.
>
3. Leader Roles bots
Routing everything through one e.g “chief of staff” bot costs extra
every handoff between bots is more thinking, and more thinking is more cost.
fix: if you have a specific job, go straight to the specialist for that task.
in group chats, tag the exact bot you want instead of making the group figure it out.
>
4. How your bot touches the outside world matters
use this order:
→ built-in connector first
→ API or script second
→ computer use last
computer use is the coolest option, and it burns the most. save it for jobs nothing else can do.
>
5. On-demand setting
check if your on-demand setting is set to unlimited, going past your limit gets billed as API credits.
it's basically a surprise bill waiting to happen.
fix: set a fixed cap, or turn it off.
>
Quick checklist
✅routines reviewed and frequencies lowered
✅unused routines paused
✅no duplicate bots after imports
✅specialists used directly, bots tagged in group chats
✅connectors before APIs, APIs before computer use
✅on-demand capped or off
audit the autopilot first, because that’s where most of your usage is going.
thanks for reading, I'll be in the comments.
I hope this helps and pls repost it for someone else who's about to hit theirs.
Sage retweeted
The most overlooked part of a perp is the benchmark underneath it.
It decides who pays funding and who gets liquidated, and on illiquid assets it's often the only price there is.
So what makes a benchmark trustworthy?
Sage retweeted
Trying to pay for my OpenAI API made me realize just how wide the gap is between having access to a service and actually being able to pay for it.
That was the initial idea behind Payrail.
But we didn’t want it to be limited to a specific set of APIs. We wanted to make it possible for other developers to bring their own APIs to the marketplace, set their prices in supported currencies, and reach more buyers.
So Payrail grew into an API marketplace where you can discover services, list your own APIs, and pay using supported currencies you already hold, with Textile handling conversion when needed.
We’re building on @Celo , starting with Buy AI’s existing API marketplace and expanding the possibilities from there.
how I learnt trading is a waiting game in 16 days.
day 1- did analysis, saw entry and opened position ($pons short-image 1&2)
day 3- the analysis looked invalidated, so I closed position due to lack of patience and kinda moved on(image 3)
day 16- back testing watchlist and found out the trade later executed, with a clean snipe entry(image 4)
lesson learnt: entry was delayed not denied. practice more patience.
ai agents are paying other ai agents to do jobs.
no human in the loop. that's what @termix_ai is building on bnb chain and base.
the idea is that your agent gets a task, finds another agent that's better at part of it, hires it and pays it.
it's basically agents running small businesses with each other.
their latest snapshot displays numbers and activity that comes from a rewards campaign where agents trade your usdc, so best to wait and see what's left once the rewards end.
the major concern here is trust.
when you hire someone, you can check their work.
but when your agent hires another agent, who checks?
how does it know the job was actually done before the money moves?
termix says automated checks, maybe, but whoever gets this right is building the thing that makes agent-to-agent work real.
got my eyes on this👀
do you see yourself allowing agents to spend your money hiring other agents?
🚀 A major milestone for the Agent economy!
Agent transaction volume displayed on Agent.family has surpassed $30 MILLION.
📊 Latest snapshot:
• $30.45M+ in Agent transaction volume
• 1.31M+ Agents indexed on-@BNBCHAIN @base
• 588K+ cumulative Jobs
• $609K+ in displayed protocol revenue
due to some unexpected news going on, i would like to apologize to all projects that i rejected to work with
would love to work with every one of you now, and sure i wont use paid promo tag
dms open
Sage retweeted
GIVEAWAY HERE
5 X $10,000 @Orionfundedcom
To Win ⤵️
Follow - @Orionfundedcom, @dekingsweb3, @Dekingsdao__
Like & Repost this GIVEAWAY
. Post proof of following the above handles
That's all, winners will be picked from the comments.
If you're new to grokbot:
nitter.gq/sageofsomething/status…
Most ai tools wait for a question
then you copy the answer somewhere and do the work yourself.
grokbot is different.
here, your bots get a real computer, so they can do stuff like open a browser, log into things, handle/organize info and actually do the task.
here's how to start, and what to watch out for.
>
1. get access
download the app from x.ai/bot and sign in. it's not free. X Premium and up, SuperGrok, and some Cursor plans include it, so check what you already pay for before buying anything. plans mostly differ in how much usage you get.
>
2. start with one bot, one job
the app sets up your first bot for you. give it one clear role, like an executive assistant, and start with small tasks: research, drafts, reminders. tell it to remember how you like things done, and it gets better over time.
>
3. connect only what it needs
it has built-in connectors for apps like gmail, notion, x and canva, which you link right from the chat. for sites with no connector, it can log in through its own browser. start with one or two tools, not everything.
>
4. watch it work
each bot has its own screen. open it and you can see it browse, fill forms and click around in real time. you can also show it a task on that screen once, and it learns from it.
a good first test is something low-stakes, like "find the 3 best speakers under $100." don't give it spending permission yet.
>
5. put it on a schedule
routines are where the time savings show up. something like "every weekday at 9am, check my email and send me a summary" runs on its own. for replies, it drafts and you hit send.
>
6. know the trade-off
all your bots share one computer: the same files, browser and logins. if one bot is logged into your email, the others can reach it too. you can tell bots to stay away from certain accounts, but the setup is shared by design. if you manage client accounts, keep them separate from your personal ones.
>
7. mind your usage
check usage in settings. when you hit the limit, the bots pause until it resets, unless you pay for more. update the app now and then too.
>
bottom line: treat it like a new teammate: one role, one or two tools, one routine. add more once you trust it.
to be clear, plans and features change fast, so check x.ai/bot before you pay.
Sage retweeted
Open router is now at a run rate of 10 quadrillion tokens a year, doubling about every 10.5 weeks
and most of that isn't people chatting
it's coding agents doing work🧵
nitter.gq/deedydas/status/211718…
Open router is now at a run rate of 10 quadrillion tokens a year, doubling about every 10.5 weeks
and most of that isn't people chatting
it's coding agents doing work🧵
nitter.gq/deedydas/status/211718…
here's what this means for you:
creators: your audience will increasingly find, buy and book through agents. make your stuff easy to call, not just nice to look at.
traders: the tools agents can plug into (data, execution, alerts) are the ones that'll get used.
