This is getting a little too easy: your call with Ivan starts in 10 minutes, and a $20 Claude already knows he's the design lead who wants short async notes
The board:
- 1 card per person: who, meetings, key details
- 3 folders: what you know, who you know, what you do
- 3 root files for the long view: memory, lessons, decisions
- 5 minutes to set up, 0 apps needed
The trick is distilling. A meeting goes into that person's card, a fact goes into notes/, and nothing gets dumped into one giant file
Watch second 4. Ivan's card fills in line by line
Honestly, no CRM I know has a field for "prefers async, short notes". That's the line that saves the call
Bookmark the full breakdown ↓
This is getting a little too easy: your call with Ivan starts in 10 minutes, and a $20 Claude already knows he's the design lead who wants short async notes
The board:
- 1 card per person: who, meetings, key details
- 3 folders: what you know, who you know, what you do
- 3 root files for the long view: memory, lessons, decisions
- 5 minutes to set up, 0 apps needed
The trick is distilling. A meeting goes into that person's card, a fact goes into notes/, and nothing gets dumped into one giant file
Watch second 4. Ivan's card fills in line by line
Honestly, no CRM I know has a field for "prefers async, short notes". That's the line that saves the call
Bookmark the full breakdown ↓
This is getting a little too easy: $20 a month gets you Claude Code, and 1 prompt turns its pixel crab into 1,860 voxels you can pull apart. Watch second 11
The board:
- 24×14 pixels in, 14 layers out
- 5 parts: legs 96, body 1,576, arms 32, eyes 12, lid 144
- the arms clip on at layer 7
- 2 models: Opus 5.5 plans and checks, Sonnet 5.5 prints the layers
The trick is the split. One model decides, the other one stacks. Swap either and the crab stays the same
Watch second 11 again. Nothing was ever glued
Honestly, "never smooth a pixel" is the whole look in 4 words. I keep rereading that line
Bookmark the full breakdown ↓
This is getting a little too easy: $20 a month gets you Claude Code, and 1 prompt turns its pixel crab into 1,860 voxels you can pull apart. Watch second 11
The board:
- 24×14 pixels in, 14 layers out
- 5 parts: legs 96, body 1,576, arms 32, eyes 12, lid 144
- the arms clip on at layer 7
- 2 models: Opus 5.5 plans and checks, Sonnet 5.5 prints the layers
The trick is the split. One model decides, the other one stacks. Swap either and the crab stays the same
Watch second 11 again. Nothing was ever glued
Honestly, "never smooth a pixel" is the whole look in 4 words. I keep rereading that line
Bookmark the full breakdown ↓
This is getting a little too easy: $20 a month gets you Claude Code, and 1 prompt turns its pixel crab into 1,860 voxels you can pull apart. Watch second 11
The board:
- 24×14 pixels in, 14 layers out
- 5 parts: legs 96, body 1,576, arms 32, eyes 12, lid 144
- the arms clip on at layer 7
- 2 models: Opus 5.5 plans and checks, Sonnet 5.5 prints the layers
The trick is the split. One model decides, the other one stacks. Swap either and the crab stays the same
Watch second 11 again. Nothing was ever glued
Honestly, "never smooth a pixel" is the whole look in 4 words. I keep rereading that line
Bookmark the full breakdown ↓
This is getting a little too easy: 1 message, 30 days of email and calendar, and a Grok Bot ranks 5 things it could take off your plate before you've set anything up
The board:
- 15 minutes to set up
- 5 things, ranked by hours saved per week
- 1 Chief of Staff, 6 specialists, one domain each
- 7 days to find out if it paid for itself
The trick is the order. Ask first, hire second. Most setups do it the other way round
Watch second 1. That's the message
Honestly, the day 7 question is the one I'd actually answer out loud
Bookmark the full breakdown ↓
This is getting a little too easy: 1 message, 30 days of email and calendar, and a Grok Bot ranks 5 things it could take off your plate before you've set anything up
The board:
- 15 minutes to set up
- 5 things, ranked by hours saved per week
- 1 Chief of Staff, 6 specialists, one domain each
- 7 days to find out if it paid for itself
The trick is the order. Ask first, hire second. Most setups do it the other way round
Watch second 1. That's the message
Honestly, the day 7 question is the one I'd actually answer out loud
Bookmark the full breakdown ↓
you show a grok bot a task once, and on the next run it adds a step you never showed it. that's card f of 10, and the other nine are just as strange
it's the grok bot a-to-z breakdown - one card for each thing a bot can do, from its own machine to its own memory
its own computer, watch its screen, take over to log in, create it by talking, connectors, teach a task, routines, triggers, a team in roles, three layers of memory
grok bot now comes with the $30 x premium plan
you hit plus and describe the job. "plan a vacation, watch the fares" turned into a bot called atlas, travel scout
the routines, the triggers and the hand-offs between bots - all handed off
my favourite card is a. 16 gb, 8 cores, and the line under it says "laptop closed · still running"
how to use the video:
> pause on card a and read the specs
> watch card f for the step it adds on its own
> stop at card h and set one trigger from slack
full breakdown below ↓
can't code? 10 trading repos, 528k github stars between them, one pasted prompt each. the card i stopped on was microsoft's - an agent that does a quant researcher's job overnight
it's the setup edition of an ai fund build. one line on each:
tradingagents - a trading firm of ai agents, you load your real book
ai-hedge-fund - pick strategies and tickers, backtest, then paper trade
daily_stock_analysis - fork it once, it runs itself on github actions
kronos - reads candles like words and predicts the next one
rd-agent - the microsoft one, factors and models overnight
openbb - the free data layer, you ask it in english
ccxt - one api, 100+ exchanges, switch with one string
freqtrade - a crypto bot that tunes and retrains itself
hummingbot - quotes both sides of the book
nautilustrader - the same strategy class in backtest and live
the repos are free. claude code comes with the $20 claude pro plan
honestly the list is the easy part. picking one desk and leaving the other three alone is the hard part.
how to use the video:
> pause on each card and read its 3-line task log
> start with one desk, not all four
> paper first, the way hummingbot's card says: no keys, no money
full prompts below ↓
can't code? 10 trading repos, 528k github stars between them, one pasted prompt each. the card i stopped on was microsoft's - an agent that does a quant researcher's job overnight
it's the setup edition of an ai fund build. one line on each:
tradingagents - a trading firm of ai agents, you load your real book
ai-hedge-fund - pick strategies and tickers, backtest, then paper trade
daily_stock_analysis - fork it once, it runs itself on github actions
kronos - reads candles like words and predicts the next one
rd-agent - the microsoft one, factors and models overnight
openbb - the free data layer, you ask it in english
ccxt - one api, 100+ exchanges, switch with one string
freqtrade - a crypto bot that tunes and retrains itself
hummingbot - quotes both sides of the book
nautilustrader - the same strategy class in backtest and live
the repos are free. claude code comes with the $20 claude pro plan
honestly the list is the easy part. picking one desk and leaving the other three alone is the hard part.
how to use the video:
> pause on each card and read its 3-line task log
> start with one desk, not all four
> paper first, the way hummingbot's card says: no keys, no money
full prompts below ↓
these 10 GitHub repos are the most DANGEROUS thing in trading right now
together they give you a full AI hedge fund: a research desk, an investment committee and an execution desk
in this article I've shown EXACTLY how to build it from scratch https://t.co/17pf1e63VG
40 seconds. that's how long you show a grok bot a task instead of writing a paragraph about it, and it never once asks for your password
it's part 1 of a 10-step tutorial - stop prompting, start delegating
install, meet the chief, job title, charter, plugins, login handoff, show it once
grok bot now comes with the $30 x premium plan
when it hits a login wall it hands you the screen. you sign in, click done, it keeps the session. a session, not a secret
the triage, the drafts and the weekly report - all handed off
my favourite part is the charter's third line, "where it stops." everyone skips it. it's the only one that matters at 3am
how to use the video:
> follow the bot card by card
> pause on "show it once"
> pick one weekly task with two or more tools and show it
full tutorial below ↓
40 seconds. that's how long you show a grok bot a task instead of writing a paragraph about it, and it never once asks for your password
it's part 1 of a 10-step tutorial - stop prompting, start delegating
install, meet the chief, job title, charter, plugins, login handoff, show it once
grok bot now comes with the $30 x premium plan
when it hits a login wall it hands you the screen. you sign in, click done, it keeps the session. a session, not a secret
the triage, the drafts and the weekly report - all handed off
my favourite part is the charter's third line, "where it stops." everyone skips it. it's the only one that matters at 3am
how to use the video:
> follow the bot card by card
> pause on "show it once"
> pick one weekly task with two or more tools and show it
full tutorial below ↓
12 grok bots run a whole company-in-a-box, and one judge on top decides every next step. the counter says 0 sent without you
it's the "self-managing ai company" roadmap - grok bot does the work, jev does the judgment
projects manager, cooper, writing bot, seo desk, stills and clips, outbound, engineer, figma bro, people ops, haggle bot
grok bot now comes with the $30 x premium plan
you give it one objective and jev runs four checks: who goes next, is the research good enough, is it really done, does it need you
the briefing, the drafts, the clips and the vendor counteroffers - all handed off
my favourite part: cooper brings back "company x launched y yesterday" with no source, and jev sends it back with two words: verify more. i would have posted it
how to use the video:
> follow the little bot from desk 1 to desk 10
> pause on the two desks marked "parked · you"
> start with one bot and one check, not twelve
full roadmap below ↓
12 grok bots run a whole company-in-a-box, and one judge on top decides every next step. the counter says 0 sent without you
it's the "self-managing ai company" roadmap - grok bot does the work, jev does the judgment
projects manager, cooper, writing bot, seo desk, stills and clips, outbound, engineer, figma bro, people ops, haggle bot
grok bot now comes with the $30 x premium plan
you give it one objective and jev runs four checks: who goes next, is the research good enough, is it really done, does it need you
the briefing, the drafts, the clips and the vendor counteroffers - all handed off
my favourite part: cooper brings back "company x launched y yesterday" with no source, and jev sends it back with two words: verify more. i would have posted it
how to use the video:
> follow the little bot from desk 1 to desk 10
> pause on the two desks marked "parked · you"
> start with one bot and one check, not twelve
full roadmap below ↓
This is f*cking gold..
9 grok bots go from a trend list to a launch plan in one weekly loop, and the counter at the top of the board says 0 published without you
it's the "build anything with a grok bot team" course - every bot writes one file and passes it to the next
scout, pain miner, competitor mapper, strategy lead, spec writer, builder, video director and two more
grok bot now comes with the $30 x premium plan
you start the week and the relay runs: trending.md → pain.md → strategy.md → SPEC.md → build/ → video.md
the trend scan, the 150 public quotes, the spec and the launch plan - all handed off
my favourite part: the builder's tests read 24/24 on mac, windows and linux, and it still stops and waits for you. i would have shipped at 24/24
how to use the video:
> follow the little bot as it walks from card 1 to card 10
> pause on card 10, the only one marked "your call"
> start your own week with one bot, not nine
full course below ↓
This is f*cking gold..
9 grok bots go from a trend list to a launch plan in one weekly loop, and the counter at the top of the board says 0 published without you
it's the "build anything with a grok bot team" course - every bot writes one file and passes it to the next
scout, pain miner, competitor mapper, strategy lead, spec writer, builder, video director and two more
grok bot now comes with the $30 x premium plan
you start the week and the relay runs: trending.md → pain.md → strategy.md → SPEC.md → build/ → video.md
the trend scan, the 150 public quotes, the spec and the launch plan - all handed off
my favourite part: the builder's tests read 24/24 on mac, windows and linux, and it still stops and waits for you. i would have shipped at 24/24
how to use the video:
> follow the little bot as it walks from card 1 to card 10
> pause on card 10, the only one marked "your call"
> start your own week with one bot, not nine
full course below ↓
claude code has 4 ways to keep an agent working while you sleep, and in every one a second agent gets to say no
it's the full loop engineering breakdown - each loop type with a one-line note on what starts it and what ends it
verify skill, /goal, /loop and /schedule, plus all of them stacked in auto mode
every loop is the same 5 parts: trigger, work, gate, state, stop
you write one skill file, one state file and one gate, then hand it the schedule
the 3 a.m. babysitting, the copy-pasted prompts and the "looks fine to me" reviews - all handed off
my favourite part: the judge rejects a lighthouse score of 88. honestly, i would have shipped that 88
how to use the video:
> watch the /goal card go from 84 to 88, both rejected
> pause on the four bills at the bottom
> build your first loop in the order it shows
full breakdown below ↓
claude code has 4 ways to keep an agent working while you sleep, and in every one a second agent gets to say no
it's the full loop engineering breakdown - each loop type with a one-line note on what starts it and what ends it
verify skill, /goal, /loop and /schedule, plus all of them stacked in auto mode
every loop is the same 5 parts: trigger, work, gate, state, stop
you write one skill file, one state file and one gate, then hand it the schedule
the 3 a.m. babysitting, the copy-pasted prompts and the "looks fine to me" reviews - all handed off
my favourite part: the judge rejects a lighthouse score of 88. honestly, i would have shipped that 88
how to use the video:
> watch the /goal card go from 84 to 88, both rejected
> pause on the four bills at the bottom
> build your first loop in the order it shows
full breakdown below ↓
A bot went through a computer and found 470 GB it could delete. It deleted nothing, and handed back three buckets instead:
"It deletes nothing. It sorts - you say yes or no."
That's the cleanup bot from a breakdown of Grok Bot, xAI's agent that runs on its own cloud computer. One bot, one job, nine of them.
scan → sort → low risk → medium risk → you → delete
Low risk was 90 GB of caches and old installers. Medium was 380 GB of old projects and raw video. High risk was left empty on purpose, for the owner to decide.
The other eight jobs work the same way. Mail sorted before 9. School emails turned into calendar events. A coding bot that runs until CI is green. A Telegram chief of staff set up in about 2 minutes.
Honestly, I go back and forth on this one. 380 GB of "medium" is a lot of folders to say yes to, and most people will just click yes on all of it.
But I keep thinking about the empty high-risk row. The bot knew which part wasn't its call.
Would you let a bot clean your computer if it had to ask first?
a grok bot watched someone do a 4-step report for 40 seconds and never needed the paragraph explaining it again. the catch is what it does when it hits a sign-in page.
it's part 1 of a 10-step grok bot breakdown: 5 bots, each on its own cloud computer, 8 plugins wired once for all of them.
errand → charter → plugins → login → show it once → routine → back to the errand
"40 s to show it once instead of writing it twice." - the breakdown
the four steps are grafana, sheets, notion, slack. read the numbers, find what dropped, paste it under the right heading, ping the lead if anything fell more than 15%.
the first errand is tiny on purpose: find a $84.20 dinner receipt, file it under october, 30 seconds for you to check.
the part that got me is the login wall. the bot stops, you sign in on its browser, click done, and it carries on in the same session. the password never touches the chat.
and the honest part: plugins connect at the account level, so the blast radius of one connection is every bot.
bookmark & watch today ↓