
AI tools, agents, experiments & everything worth exploring. Founder @firsayai & @0penoffers | BD @frameonx | Amb @1inch
Joined September 2010
- Tweets34.7K
- Following2.8K
- Followers4.9K
- Likes62.7K
been paying $4.62 for a 10-second Seedance 2.5 clip that costs $2.32 at the maker’s own price
same model. same settings. no markup. no subscription. just pay as you go.
what you actually get:
→ one key for every leading image, video, audio and 3D model
→ the model you call is the model that runs
→ every request itemized in the console
building at scale? enterprise deals go up to 40% off leading models’ own API prices (volume-based).
the models are ready. your app is next.
Claude Code just showed one guy what his phone does while he sleeps. TikTok had turned his mic on 347 times. with the app closed
Instagram did it 211 times, Facebook 189, Snapchat 94. each activation lasted 3 to 15 seconds long enough to catch a full sentence you said out loud
that was just 1 of 4 posts that blew up this month:
→ a “photo” sent over Bluetooth that wasn’t a photo. a week later the phone was mocking its owner on screen
→ a Walmart teddy bear in a kid’s room, uploading video every 30 seconds all night
→ $3.99 here, $2.49 there. 7 months of Apple Pay charges under $6. $4,712 gone before anyone noticed
→ and Apple’s answer: every charge was “Face ID authenticated”
you’re checking your bank app every day. you’ve never once checked what your phone does when you’re not looking
do this tonight, it takes 10 minutes:
→ iPhone: Settings > Privacy > Microphone. the orange dot = mic is live right now
→ scan your wifi at night with the house asleep. whatever uploads during silence is the problem
→ every charge under $10 and every device in your payment apps you don’t recognize. screenshot before you touch anything
the tools security people actually use, free:
→ MVT by Amnesty International. scans your phone for known spyware
→ TinyCheck by Kaspersky. shows everything your phone sends out over wifi
every prompt, command and fix is in the article below. bookmark it before you need it ↓
elon told 39 million people Grok Bot is a chief of staff. most of them will run it like a smoke alarm
7 bots on a 5-minute routine = 7 × 288 × 7 = 14,112 runs a week. if 30 find something real, the other 14,082 just woke up to look around
the math on one routine:
→ every 5 min = 288 runs a day
→ hourly = 24
→ once a day = 1
→ every run burns your weekly usage pool. even the test button
5-min vs hourly is 12x the usage for the exact same job. one busy routine can eat a week's pool in a day
the fix: schedule the job, not the clock
→ morning brief: once, 07:00
→ call prep: once at 07:00, preps every call on today's calendar
→ invoices: once, on the 1st
→ renewals: once, every friday
→ inbox: hourly, work hours only. the only bot that earns it
→ ~100 runs a week instead of 14,112
where it breaks:
→ a meeting moves after 07:00 and the prep is stale. add one rerun at 13:00, not 288
→ 40 emails land at once. batch them into the hourly run, never one run per email
the rule to put in every bot's description:
------
Before any work, check what changed since your last run. If nothing changed, say nothing and stop. Never send, pay or sign without my approval. One job only.
------
save this before you give your next bot a schedule ↓
your 100 coding agents cost $389 a month or $5,712 a month. same sandboxes, same hours. the only difference is who you rent them from
100 small sandboxes, 8 hours a day, 30 days:
→ Lizard $389
→ Boat $432
→ Cloudflare $1,673
→ Vercel $2,650
→ E2B $3,974
→ Daytona $3,974
→ Modal $5,712
almost 15x between the top and the bottom of that list. for the exact same box
and most fleets don't even run 8 hours. they run 24/7 while nobody gives them work:
→ 50 agents on small sandboxes, 24/7: $136 a week
→ the same 50, paused outside working hours: $36
→ a paused sandbox costs $0 and wakes up with the same memory, files and running processes
the other number that matters when every task spins up a fresh sandbox. median cold start, 1,200 boots:
→ Lizard 238 ms
→ E2B 270 ms
→ Daytona 796 ms
→ Modal 865 ms
if your agents live at the bottom of the price list, you're paying for the logo
first week is free, link below ↓
🤝 Paid partnership
Some users asked where is comparison with @boatdotdev. Added it, and scaled the bill to a real agent fleet
100 small sandboxes, 8 hours a day, 30 days:
> Lizard $389
> Boat $432
> Cloudflare $1,673
> Vercel $2,650
> E2B $3,974
> Daytona $3,974
> Modal $5,712
lizard.build/boat
Very close, but Lizard still better price!
your 100 coding agents cost $389 a month or $5,712 a month. same sandboxes, same hours. the only difference is who you rent them from
100 small sandboxes, 8 hours a day, 30 days:
→ Lizard $389
→ Boat $432
→ Cloudflare $1,673
→ Vercel $2,650
→ E2B $3,974
→ Daytona $3,974
→ Modal $5,712
almost 15x between the top and the bottom of that list. for the exact same box
and most fleets don't even run 8 hours. they run 24/7 while nobody gives them work:
→ 50 agents on small sandboxes, 24/7: $136 a week
→ the same 50, paused outside working hours: $36
→ a paused sandbox costs $0 and wakes up with the same memory, files and running processes
the other number that matters when every task spins up a fresh sandbox. median cold start, 1,200 boots:
→ Lizard 238 ms
→ E2B 270 ms
→ Daytona 796 ms
→ Modal 865 ms
if your agents live at the bottom of the price list, you're paying for the logo
first week is free, link below ↓
🤝 Paid partnership
Some users asked where is comparison with @boatdotdev. Added it, and scaled the bill to a real agent fleet
100 small sandboxes, 8 hours a day, 30 days:
> Lizard $389
> Boat $432
> Cloudflare $1,673
> Vercel $2,650
> E2B $3,974
> Daytona $3,974
> Modal $5,712
lizard.build/boat
Very close, but Lizard still better price!
your $30 X subscription came with an employee you can rent out to small businesses. most people use it to ask what's on their calendar
7 Grok Bot setups people will actually pay for:
→ receipts and inbox desk: solo founders drowning in email
→ lead research with drafted outreach: agencies and B2B founders
→ team bot inside their slack
→ webhook intake: anyone with a form on their site
→ content research desk: creators
→ competitor watch inbox: marketing teams
→ morning brief for traders: it drafts, you decide
why they pay: it's not answers, it's routines. "every weekday at 8:00, do X". "when an email lands from this domain, do Y". a cloud agent with its own computer that keeps working when the laptop is closed
the one i'd sell first is lead research. every founder hates prospecting more than paying for it. the bot finds leads, reads what they posted, writes a first line for each. a human hits send, or the client's domain burns
two rules before you sell any of it:
→ the client owns the account and pays the plan. you charge for setup and upkeep
→ every routine run burns the weekly usage pool, even the test button. one busy slack trigger can eat a week in a day
routine wording, what to charge and where each one breaks ↓
anthropic just opened its books and the number that should scare you is 900%
net revenue retention. customers who paid anthropic a year ago now pay ~9x more
the best public software company at this, palantir, sits at 157%
that 9x is you
→ the $20 plan that became $100, then $200
→ the "one more opus run" at 2am
→ the 5-hour limit that burned out in 30 minutes last night
they didn't just find more customers. they found out how much more you'd pay
and you will. so will i
Anthropic has filed an S-1 registration statement with the SEC related to its proposed IPO.
Read more: anthropic.com/news/s1-sec
Legal disclaimer: A registration statement relating to these securities has been filed with the SEC but has not yet become effective. These securities may not be sold, nor may offers to buy be accepted, prior to the time the registration statement becomes effective. This post does not constitute an offer to sell or the solicitation of an offer to buy these securities, nor shall there be any sale of these securities in any state or jurisdiction in which such offer, solicitation or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction.
holy f*ck, Grok Bot just replaced your assistant
it's f*cking insane… but only if your prompt isn't one line
"check my calendar every morning" gets you a calendar
this gets you a chief of staff:
bot 1, morning brief
"Every weekday at 07:00 read today's calendar, open reminders and unread email. Send me one message: the 3 things that matter today, every meeting with one line on who it is, anything that moved since yesterday. If nothing is urgent: 'quiet day'."
→ 07:00, one message. you know your whole day before coffee
bot 2, call prep
"15 minutes before every call with someone outside my team, find them in my contacts and read my last 5 emails with them. Send me: who they are, what we last agreed, what they'll ask, one question I should ask. Max 6 lines."
→ you never walk into a call cold again
paste them into the bot's description, not the chat
the other 5 are in the article. bot 3 signs into your billing pages on the 1st and sends every invoice to your accountant. that job used to die at the login screen ↓
your Grok Bot routine runs every 5 minutes. it costs you ~12x more than running it once an hour
same bot. same job. 12x the bill
that's 288 runs a day for a job you read twice
the fix takes 30 seconds:
→ most jobs run once a day: morning brief, renewals, weekly numbers
→ hourly is for the inbox. nothing else
→ every routine gets one line: "if there's nothing important, say nothing"
→ one bot per job. small and boring beats big and clever
do that and the bot you used to ignore becomes the one message you actually open
the full 7-bot setup with every prompt below ↓
your Grok Bot routine runs every 5 minutes. it costs you ~12x more than running it once an hour
same bot. same job. 12x the bill
that's 288 runs a day for a job you read twice
the fix takes 30 seconds:
→ most jobs run once a day: morning brief, renewals, weekly numbers
→ hourly is for the inbox. nothing else
→ every routine gets one line: "if there's nothing important, say nothing"
→ one bot per job. small and boring beats big and clever
do that and the bot you used to ignore becomes the one message you actually open
the full 7-bot setup with every prompt below ↓
2.7 million people just watched Grok Bot get access to their phone. most of them will ask it what's on their calendar and close the app
that's like buying a Tesla to charge your phone
the setup that actually uses it: 7 bots, one job each
→ morning brief at 07:00: the 3 things that matter today and who's in every meeting
→ call prep 15 min before every call: who they are, what you agreed, what they'll ask
→ invoice collector on the 1st: every invoice to your accountant, totals to you
→ follow-ups after every meeting, waiting in your drafts
→ inbox router: max 5 emails that actually need you
→ weekly numbers every monday in 3 lines
→ renewal watch every friday
one bot, one job. the Grok Bot team said it on their own livestreams
every prompt is ready to paste below ↓
2.7 million people just watched Grok Bot get access to their phone. most of them will ask it what's on their calendar and close the app
that's like buying a Tesla to charge your phone
the setup that actually uses it: 7 bots, one job each
→ morning brief at 07:00: the 3 things that matter today and who's in every meeting
→ call prep 15 min before every call: who they are, what you agreed, what they'll ask
→ invoice collector on the 1st: every invoice to your accountant, totals to you
→ follow-ups after every meeting, waiting in your drafts
→ inbox router: max 5 emails that actually need you
→ weekly numbers every monday in 3 lines
→ renewal watch every friday
one bot, one job. the Grok Bot team said it on their own livestreams
every prompt is ready to paste below ↓
holy sh*t, someone just watched 72 hours of the Grok Bot team's own livestreams and turned it into one copy-paste setup
not the docs. how the people who build Grok Bot actually run it: their prompts, their schedules, their costs
what's inside:
→ a one-message install. a bot from the marketplace reads the repo and designs your whole setup, then waits for your yes
→ 7 prompt habits the team landed on by themselves
→ the rules file every bot reads before it does anything
→ 5 routines and 5 connector rules that save time, money and nerves
→ 5 ready teams built from marketplace bots
→ a first-week plan and a checklist to screenshot
the line that changed how i set up agents:
"verification is the job. writing code is the easy part"
and 3 rules i'm adding today:
→ "if a routine finds nothing important, say nothing"
→ "anything that leaves my team is a draft. I send it"
→ every Friday the bot compares my edits to its drafts and turns the difference into style rules
16-minute read. every setup is a block you paste
save this ↓
AI x DeFi that actually does something: my agent opens and closes LP positions on @1inch Aqua, my wallet signs every one
connected Claude to the 1inch MCP, one sentence per action:
"how are my Aqua positions doing?"
one reply: [N] positions across [pairs / chains], each with liquidity, fees and recent activity
"close the [PAIR] one"
it comes back with a preview first: pair, amounts, price, fee
nothing is sent until i confirm and my wallet signs
what i like:
→ no new app. it runs inside the AI i already use
→ no deposit. tokens stay in my wallet, one approval backs a position
→ closing is final and there's nothing to withdraw, they never left
don't want to connect a wallet? reading works for any wallet, no account needed
my run in ↓
We asked one of our own auditors how safe DeFi really is.
@bailsecurity has audited the Aqua code alongside seven other firms. Vik from their team on what an audit catches and what it does not.
your Railway secrets are probably rotated by now. your bill isn't
i priced the setup half of us actually run: 10 sandboxes for coding agents, 24/7, 2 vCPU + 4 GB each
→ Railway: ~$2,129 a month
→ Daytona: $912
→ E2B: $907
→ Vercel: $826
→ Modal: $440
→ @LizardBuild: $118
same 10 agents. 18x apart
and $118 is the worst case, every sandbox running all month. in practice:
→ you pay per second
→ an idle sandbox pauses at $0 and wakes up with the same memory and processes
→ agents talk to each other over a free private network, EU US included
→ /workspace survives a rebuild
→ a broken deploy comes back as JSON with the reason and exit code, so the agent fixes it, not you at 3am
if you're moving anyway, don't spend the weekend on it. tell your coding agent:
"migrate us to Lizard"
review the changes, ship, done
still on Railway? rotate every secret first. then move. first week is free
🤝 Paid partnership
Builds are temporarily paused
We are currently investigating this issue
Status: Investigating status.railway.com/incident/…

















