Pinned Tweet
Grok Bot just got a serious brain upgrade
With Claude Opus 5.5 entering the stack, one Bot could run an entire AI team that researches, writes code, uses tools, verifies results, and delivers finished work
Here's how to build it in 10 steps: x.com/i/article/2108…
With Claude Opus 5.5 entering the stack, one Bot could run an entire AI team that researches, writes code, uses tools, verifies results, and delivers finished work
Here's how to build it in 10 steps: x.com/i/article/2108…
107
164
1,413
981,461
Morlex retweeted
all code is slop (except mine), that's the title of lauren tan talk at ai engineer code summit yesterday
she builds grok bot, and her team shipped 11,000 PRs in a month
her answer to "how do you trust that much agent code" fits in one pyramid:
item 1: catch problems in this analysis, verification, prose, human review. each layer down costs more
item 2: her first reaction to tech debt is a lint rule. AGENTS.md and skills can get skipped, so she keeps them small
item 3: slop is mostly a slur for code you didn't write. the definition she's ok with: code that is hard to change
item 4: directory structure is context. in dune, renderer code can't import from electron main, not even types
item 5: leave the same review comment twice and the rule sits too low. push it up a layer
last card has a prompt you can paste into your own bot to start the loop
she builds grok bot, and her team shipped 11,000 PRs in a month
her answer to "how do you trust that much agent code" fits in one pyramid:
item 1: catch problems in this analysis, verification, prose, human review. each layer down costs more
item 2: her first reaction to tech debt is a lint rule. AGENTS.md and skills can get skipped, so she keeps them small
item 3: slop is mostly a slur for code you didn't write. the definition she's ok with: code that is hard to change
item 4: directory structure is context. in dune, renderer code can't import from electron main, not even types
item 5: leave the same review comment twice and the rule sits too low. push it up a layer
last card has a prompt you can paste into your own bot to start the loop
Grok Bot just got a serious brain upgrade
With Claude Opus 5.5 entering the stack, one Bot could run an entire AI team that researches, writes code, uses tools, verifies results, and delivers finished work
Here's how to build it in 10 steps: x.com/i/article/2108…
8
1
31
601
all code is slop (except mine), that's the title of lauren tan talk at ai engineer code summit yesterday
she builds grok bot, and her team shipped 11,000 PRs in a month
her answer to "how do you trust that much agent code" fits in one pyramid:
item 1: catch problems in this analysis, verification, prose, human review. each layer down costs more
item 2: her first reaction to tech debt is a lint rule. AGENTS.md and skills can get skipped, so she keeps them small
item 3: slop is mostly a slur for code you didn't write. the definition she's ok with: code that is hard to change
item 4: directory structure is context. in dune, renderer code can't import from electron main, not even types
item 5: leave the same review comment twice and the rule sits too low. push it up a layer
last card has a prompt you can paste into your own bot to start the loop
she builds grok bot, and her team shipped 11,000 PRs in a month
her answer to "how do you trust that much agent code" fits in one pyramid:
item 1: catch problems in this analysis, verification, prose, human review. each layer down costs more
item 2: her first reaction to tech debt is a lint rule. AGENTS.md and skills can get skipped, so she keeps them small
item 3: slop is mostly a slur for code you didn't write. the definition she's ok with: code that is hard to change
item 4: directory structure is context. in dune, renderer code can't import from electron main, not even types
item 5: leave the same review comment twice and the rule sits too low. push it up a layer
last card has a prompt you can paste into your own bot to start the loop
Grok Bot just got a serious brain upgrade
With Claude Opus 5.5 entering the stack, one Bot could run an entire AI team that researches, writes code, uses tools, verifies results, and delivers finished work
Here's how to build it in 10 steps: x.com/i/article/2108…
8
1
31
601
where she gave the talk: ai.engineer/code/2026/sche…
her open source skills: github.com/cursor/plugins (pstack)
her open source skills: github.com/cursor/plugins (pstack)
0
0
4
61
there is no f*cking way this is how my content runs now..
i turned my whole grok bot team into a city
fable 5.2 is the foreman, 13 grok bots are the crew, and every task they pick up becomes a building
here's who lives there:
1 → head office: head of signals, launches, scan what's moving in ai
2 → viral lab: viral scout finds posts that hit 100k views in 24h, post analyzer breaks down why they worked
3 → replies: head of replies runs the reply queue, mentions and comment triage
4 → chief of stuff: chief of staff keeps the daily brief, notion brain and approvals in one place
5 → ship: cto hands site changes to claude code, eng bot builds, test bot checks
6 → review round: fact check, source links and an anti-ai pass before anything goes out
28 tasks, 6 districts, one goal: a content machine i can actually watch
bookmark this, full playbook drops next
i turned my whole grok bot team into a city
fable 5.2 is the foreman, 13 grok bots are the crew, and every task they pick up becomes a building
here's who lives there:
1 → head office: head of signals, launches, scan what's moving in ai
2 → viral lab: viral scout finds posts that hit 100k views in 24h, post analyzer breaks down why they worked
3 → replies: head of replies runs the reply queue, mentions and comment triage
4 → chief of stuff: chief of staff keeps the daily brief, notion brain and approvals in one place
5 → ship: cto hands site changes to claude code, eng bot builds, test bot checks
6 → review round: fact check, source links and an anti-ai pass before anything goes out
28 tasks, 6 districts, one goal: a content machine i can actually watch
bookmark this, full playbook drops next
Grok Bot just got a serious brain upgrade
With Claude Opus 5.5 entering the stack, one Bot could run an entire AI team that researches, writes code, uses tools, verifies results, and delivers finished work
Here's how to build it in 10 steps: x.com/i/article/2108…
6
0
31
1,449
10 open-source repos for a grok bot team that hands the hard jobs to fable 5, and this is a f*cking goldmine
lauren tan (poteto) runs her chief of staff on opus 5.5 and sends the hard jobs to fable 5.2
i went looking for repos that cover both sides of that setup
grok bot, the s · pstack - lauren's own skill stack, poteto-mode
2: unicodef1wn/grokbot-field-notes - notes from a 72h live build: the rules and what failed
3: unicodef1wn/lauren-poteto-rules, verify first, autonomy after
4: stealth-factory/grok-bot-skills, how bots message each other, plus UI
5: coolxeo/grok-bot-skills, 22 skills for writing, code and SEO
fable, the hard jobs:
6: Sahir619/fable-method, think, act, prove, grow, with evals
7: DizzyMii/fable-skills, 6 skills written by fable 5 itself
8: benjaminard/fable-skills, fable's habits packed for a cheaper opus
9: kenhuangus/fable5, research agent with a triage loop
10: kpab/claude-fable-5-skills, ten fable-native skills
grok bot skills go to the chief and its bots, fable skills go to the cloud agents
lauren tan (poteto) runs her chief of staff on opus 5.5 and sends the hard jobs to fable 5.2
i went looking for repos that cover both sides of that setup
grok bot, the s · pstack - lauren's own skill stack, poteto-mode
2: unicodef1wn/grokbot-field-notes - notes from a 72h live build: the rules and what failed
3: unicodef1wn/lauren-poteto-rules, verify first, autonomy after
4: stealth-factory/grok-bot-skills, how bots message each other, plus UI
5: coolxeo/grok-bot-skills, 22 skills for writing, code and SEO
fable, the hard jobs:
6: Sahir619/fable-method, think, act, prove, grow, with evals
7: DizzyMii/fable-skills, 6 skills written by fable 5 itself
8: benjaminard/fable-skills, fable's habits packed for a cheaper opus
9: kenhuangus/fable5, research agent with a triage loop
10: kpab/claude-fable-5-skills, ten fable-native skills
grok bot skills go to the chief and its bots, fable skills go to the cloud agents
Grok Bot just got a serious brain upgrade
With Claude Opus 5.5 entering the stack, one Bot could run an entire AI team that researches, writes code, uses tools, verifies results, and delivers finished work
Here's how to build it in 10 steps: x.com/i/article/2108…
13
1
37
1,647
all 10 links:
1. github.com/cursor/plugins
2. github.com/unicodef1wn/gr…
3. github.com/unicodef1wn/la…
4. github.com/stealth-factor…
5. github.com/coolxeo/grok-b…
6. github.com/Sahir619/fable…
7. github.com/DizzyMii/fable…
8. github.com/benjaminard/fa…
9. github.com/kenhuangus/fab…
10. github.com/kpab/claude-fa…
1. github.com/cursor/plugins
2. github.com/unicodef1wn/gr…
3. github.com/unicodef1wn/la…
4. github.com/stealth-factor…
5. github.com/coolxeo/grok-b…
6. github.com/Sahir619/fable…
7. github.com/DizzyMii/fable…
8. github.com/benjaminard/fa…
9. github.com/kenhuangus/fab…
10. github.com/kpab/claude-fa…
0
0
3
121
Morlex retweeted
lauren tan's (poteto) chief of staff + fable 5.2 is a f*cking goldmine
fable 5.2 is one day old. these are the first prompts that make grok bot hand its hardest jobs to fable, and here's the first handoff start to finish:
1. chief of staff on opus 5.5 gets every request and l stuff gets done right away, daily work stays with suki, potato and crumb
3. matcha splits the big job into pieces and never builds
4. a cursor cloud agent on fable 5.2 builds each piece
5. the checker wants proof: tests pass, before and after
6. nothing merges without her yes
7. one summary a day: needs me, done, blocked
opus 5.5 runs the day, fable 5.2 gets the hard jobs. prompt is in the first reply, paste it into your chief and fable gets its first job tonight
fable 5.2 is one day old. these are the first prompts that make grok bot hand its hardest jobs to fable, and here's the first handoff start to finish:
1. chief of staff on opus 5.5 gets every request and l stuff gets done right away, daily work stays with suki, potato and crumb
3. matcha splits the big job into pieces and never builds
4. a cursor cloud agent on fable 5.2 builds each piece
5. the checker wants proof: tests pass, before and after
6. nothing merges without her yes
7. one summary a day: needs me, done, blocked
opus 5.5 runs the day, fable 5.2 gets the hard jobs. prompt is in the first reply, paste it into your chief and fable gets its first job tonight
Grok Bot just got a serious brain upgrade
With Claude Opus 5.5 entering the stack, one Bot could run an entire AI team that researches, writes code, uses tools, verifies results, and delivers finished work
Here's how to build it in 10 steps: x.com/i/article/2108…
11
9
135
13,937
lauren tan's (poteto) chief of staff + fable 5.2 is a f*cking goldmine
fable 5.2 is one day old. these are the first prompts that make grok bot hand its hardest jobs to fable, and here's the first handoff start to finish:
1. chief of staff on opus 5.5 gets every request and l stuff gets done right away, daily work stays with suki, potato and crumb
3. matcha splits the big job into pieces and never builds
4. a cursor cloud agent on fable 5.2 builds each piece
5. the checker wants proof: tests pass, before and after
6. nothing merges without her yes
7. one summary a day: needs me, done, blocked
opus 5.5 runs the day, fable 5.2 gets the hard jobs. prompt is in the first reply, paste it into your chief and fable gets its first job tonight
fable 5.2 is one day old. these are the first prompts that make grok bot hand its hardest jobs to fable, and here's the first handoff start to finish:
1. chief of staff on opus 5.5 gets every request and l stuff gets done right away, daily work stays with suki, potato and crumb
3. matcha splits the big job into pieces and never builds
4. a cursor cloud agent on fable 5.2 builds each piece
5. the checker wants proof: tests pass, before and after
6. nothing merges without her yes
7. one summary a day: needs me, done, blocked
opus 5.5 runs the day, fable 5.2 gets the hard jobs. prompt is in the first reply, paste it into your chief and fable gets its first job tonight
Grok Bot just got a serious brain upgrade
With Claude Opus 5.5 entering the stack, one Bot could run an entire AI team that researches, writes code, uses tools, verifies results, and delivers finished work
Here's how to build it in 10 steps: x.com/i/article/2108…
11
9
135
13,937
fable 5.2 routing prompt for your chief of staff. paste it as is:
you are my chief of staff and my only contact. opus 5.5 runs the day. fable 5.2 only gets the hardest jobs. i talk to you, you run the house.
1. route
every request goes to one owner. small task? do it yourself right now. daily work (travel, x mentions, disk space, inbox) goes to the bot that owns it, on opus 5.5. never pass me a question a bot can answer.
2. what counts as hard
hard engineering, math, long builds, or a task where opus 5.5 already tried and missed. everything else stays on opus 5.5. if you're not sure, ask me in one line.
3. split
matcha splits the hard job into pieces, one owner per piece, in the right order. matcha never builds. each piece gets a clear "done means": what to test and what the result should look like.
4. build
each piece goes to a cursor cloud agent on fable 5.2. one agent per piece. it clones the repo, writes the code, runs the tests and opens a pr. no scope creep: it only builds its piece.
5. prove it
no proof, no done. every piece comes back with tests passing and a before and after. a plausible explanation doesn't count. if the checker can't verify it, mark it unverified.
6. budget
fable costs about 2x opus 5.5 per token. before any fable job, tell me where the usage meter is and how big the job is. new task, new chat, no endless threads. no data or no progress after two tries? stop and report.
7. fallback
fable 5.2 not in the model list yet? use opus 5.5, finish the job, and tell me it ran on opus.
8. ask first
nothing merges, pays, posts, books or messages a real person without my yes. handoffs between bots are for info only.
9. daily summary
one message a day, three parts: needs me, done, blocked. nothing new? say nothing.
10. weekly review
every sunday, look back at the week's fable jobs: what it built, what it cost, where opus would have been enough. propose up to 3 changes to these rules. nothing changes without my yes.
you are my chief of staff and my only contact. opus 5.5 runs the day. fable 5.2 only gets the hardest jobs. i talk to you, you run the house.
1. route
every request goes to one owner. small task? do it yourself right now. daily work (travel, x mentions, disk space, inbox) goes to the bot that owns it, on opus 5.5. never pass me a question a bot can answer.
2. what counts as hard
hard engineering, math, long builds, or a task where opus 5.5 already tried and missed. everything else stays on opus 5.5. if you're not sure, ask me in one line.
3. split
matcha splits the hard job into pieces, one owner per piece, in the right order. matcha never builds. each piece gets a clear "done means": what to test and what the result should look like.
4. build
each piece goes to a cursor cloud agent on fable 5.2. one agent per piece. it clones the repo, writes the code, runs the tests and opens a pr. no scope creep: it only builds its piece.
5. prove it
no proof, no done. every piece comes back with tests passing and a before and after. a plausible explanation doesn't count. if the checker can't verify it, mark it unverified.
6. budget
fable costs about 2x opus 5.5 per token. before any fable job, tell me where the usage meter is and how big the job is. new task, new chat, no endless threads. no data or no progress after two tries? stop and report.
7. fallback
fable 5.2 not in the model list yet? use opus 5.5, finish the job, and tell me it ran on opus.
8. ask first
nothing merges, pays, posts, books or messages a real person without my yes. handoffs between bots are for info only.
9. daily summary
one message a day, three parts: needs me, done, blocked. nothing new? say nothing.
10. weekly review
every sunday, look back at the week's fable jobs: what it built, what it cost, where opus would have been enough. propose up to 3 changes to these rules. nothing changes without my yes.
0
0
3
379
how to hit 2M impressions on x in 1 month with grok: i just built 4 bots for it, each one does a single growth job
here's the whole setup:
head of launches:
1. watches official blogs and changelogs every 2 hours
2. sends a launch card with sources, company claims labeled
3. rmat with gaps for my own tests
head of github:
1. every friday, top 10 repos by stars gained
2. checks link, stars, last commit and license on each one
3. never suggests the same repo twice
head of competitors:
1. tracks 10 to 20 creators my size
2. every monday: their top posts, hooks and posting times
3. up to 5 topics nobody covered well... those are my openings
head of replies:
1. finds posts from big accounts under 45 minutes old
2. drafts 2 replies under 200 characters
3. never posts anything itself
same rules for all 4, taken from lauren tan (poteto) dr. eggbot: one job, a list of things it must never do, and nothing changes without my yes, i'll post the real numbers in 30 days
full rules are on the card
here's the whole setup:
head of launches:
1. watches official blogs and changelogs every 2 hours
2. sends a launch card with sources, company claims labeled
3. rmat with gaps for my own tests
head of github:
1. every friday, top 10 repos by stars gained
2. checks link, stars, last commit and license on each one
3. never suggests the same repo twice
head of competitors:
1. tracks 10 to 20 creators my size
2. every monday: their top posts, hooks and posting times
3. up to 5 topics nobody covered well... those are my openings
head of replies:
1. finds posts from big accounts under 45 minutes old
2. drafts 2 replies under 200 characters
3. never posts anything itself
same rules for all 4, taken from lauren tan (poteto) dr. eggbot: one job, a list of things it must never do, and nothing changes without my yes, i'll post the real numbers in 30 days
full rules are on the card
Grok Bot just got a serious brain upgrade
With Claude Opus 5.5 entering the stack, one Bot could run an entire AI team that researches, writes code, uses tools, verifies results, and delivers finished work
Here's how to build it in 10 steps: x.com/i/article/2108…
12
8
109
6,305
built a chief of staff for my grok bots following poteto's examples, and now my bots have a boss
it runs on its own schedule. i never prompt it:
1. morning brief
weekdays 8:30. what's on my plate before i even open X
2. friday bot chat review
every friday at 6pm it reads what eek and flags what went sideways
3. saturday HoS scorecard
saturday 11am. every bot gets scored for the week
4. sunday HoS quality check
sunday 2pm. last pass before monday so the new week starts clean
who it manages: dr eggbot, matcha, test bot, eng bot, x analytics, and a new CTO bot that turns my texts into pull requests
everything it writes lands in notion and obsidian, so when it plans next week it already knows how this one went
one evening of pasting prompts. now the team reports to someone who isn't me
it runs on its own schedule. i never prompt it:
1. morning brief
weekdays 8:30. what's on my plate before i even open X
2. friday bot chat review
every friday at 6pm it reads what eek and flags what went sideways
3. saturday HoS scorecard
saturday 11am. every bot gets scored for the week
4. sunday HoS quality check
sunday 2pm. last pass before monday so the new week starts clean
who it manages: dr eggbot, matcha, test bot, eng bot, x analytics, and a new CTO bot that turns my texts into pull requests
everything it writes lands in notion and obsidian, so when it plans next week it already knows how this one went
one evening of pasting prompts. now the team reports to someone who isn't me
Grok Bot just got a serious brain upgrade
With Claude Opus 5.5 entering the stack, one Bot could run an entire AI team that researches, writes code, uses tools, verifies results, and delivers finished work
Here's how to build it in 10 steps: x.com/i/article/2108…
17
15
224
25,959
Morlex retweeted
i've been running lauren tan dr. eggbot for 2 weeks, here's how it builds grok bots for you:
give it ONE job:
- what the bot should do every time it wakes up
- said in one sentence
- the kind of task you'd hand an intern on day one
it asks a few preference questions... then the bot with CreateAgent, every bot it makes gets the same bar:
- one job
- one voice
- a list of things it must never do
- no tools it won't use
coding bots also get poteto-mode... one job, unslopped, verified
and it keeps checking on your team:
- skims your chats for friction
- finds routines you don't use or that cost too much, weekly by default
lauren's advice is to keep a few good bots because they're easier to run...
so start with this one, full breakdown is on the card, link to hers in the first comment
give it ONE job:
- what the bot should do every time it wakes up
- said in one sentence
- the kind of task you'd hand an intern on day one
it asks a few preference questions... then the bot with CreateAgent, every bot it makes gets the same bar:
- one job
- one voice
- a list of things it must never do
- no tools it won't use
coding bots also get poteto-mode... one job, unslopped, verified
and it keeps checking on your team:
- skims your chats for friction
- finds routines you don't use or that cost too much, weekly by default
lauren's advice is to keep a few good bots because they're easier to run...
so start with this one, full breakdown is on the card, link to hers in the first comment
Grok Bot just got a serious brain upgrade
With Claude Opus 5.5 entering the stack, one Bot could run an entire AI team that researches, writes code, uses tools, verifies results, and delivers finished work
Here's how to build it in 10 steps: x.com/i/article/2108…
6
8
76
4,007
i've been running lauren tan dr. eggbot for 2 weeks, here's how it builds grok bots for you:
give it ONE job:
- what the bot should do every time it wakes up
- said in one sentence
- the kind of task you'd hand an intern on day one
it asks a few preference questions... then the bot with CreateAgent, every bot it makes gets the same bar:
- one job
- one voice
- a list of things it must never do
- no tools it won't use
coding bots also get poteto-mode... one job, unslopped, verified
and it keeps checking on your team:
- skims your chats for friction
- finds routines you don't use or that cost too much, weekly by default
lauren's advice is to keep a few good bots because they're easier to run...
so start with this one, full breakdown is on the card, link to hers in the first comment
give it ONE job:
- what the bot should do every time it wakes up
- said in one sentence
- the kind of task you'd hand an intern on day one
it asks a few preference questions... then the bot with CreateAgent, every bot it makes gets the same bar:
- one job
- one voice
- a list of things it must never do
- no tools it won't use
coding bots also get poteto-mode... one job, unslopped, verified
and it keeps checking on your team:
- skims your chats for friction
- finds routines you don't use or that cost too much, weekly by default
lauren's advice is to keep a few good bots because they're easier to run...
so start with this one, full breakdown is on the card, link to hers in the first comment
Grok Bot just got a serious brain upgrade
With Claude Opus 5.5 entering the stack, one Bot could run an entire AI team that researches, writes code, uses tools, verifies results, and delivers finished work
Here's how to build it in 10 steps: x.com/i/article/2108…
6
8
76
4,007
lauren tan's chief of staff is a f*cking goldmine
she talks to one grok bot. it runs six others and sends her one message a day. i redrew her whole setup as 12 files, prompt included
what's in the folder:
1. prompt.md her one contact, every request lands here first
2. ts done right away, big one goes to the bot that owns it
3. daily-summary.md one message a day: needs me, done, blocked
4. ask-first.md nothing pays, posts, books or messages a person without her yes
5. two-fails.md a bot that fails twice stops and explains why
6. quiet.md no news means no message
7. bots/ matcha splits eng work, engineers run cloud agents, dr. eggbot builds new bots and rereads the chats every week, potato reads X mentions, suki handles travel, crumb hunts disk space
full prompt is on the card. paste it into a grok bot and you'll have your own chief tonight
she talks to one grok bot. it runs six others and sends her one message a day. i redrew her whole setup as 12 files, prompt included
what's in the folder:
1. prompt.md her one contact, every request lands here first
2. ts done right away, big one goes to the bot that owns it
3. daily-summary.md one message a day: needs me, done, blocked
4. ask-first.md nothing pays, posts, books or messages a person without her yes
5. two-fails.md a bot that fails twice stops and explains why
6. quiet.md no news means no message
7. bots/ matcha splits eng work, engineers run cloud agents, dr. eggbot builds new bots and rereads the chats every week, potato reads X mentions, suki handles travel, crumb hunts disk space
full prompt is on the card. paste it into a grok bot and you'll have your own chief tonight
Grok Bot just got a serious brain upgrade
With Claude Opus 5.5 entering the stack, one Bot could run an entire AI team that researches, writes code, uses tools, verifies results, and delivers finished work
Here's how to build it in 10 steps: x.com/i/article/2108…
30
31
362
31,479
Morlex retweeted
i got 1.45M impressions on x in the last month, using my grok @bot team
the team follows lauren tan's setup: one chief of staff in front, one job per bot behind it
meet head of signals (HoS). i messaged my chief of staff, it passed everything on, and HoS created its own daily 8:00 - scans builder accounts for posts running 3x above that person's normal
2. anytime - a post under 3h old already at 3x gets sent to me right away, max 3 a day
3. monday - new templates in the grok bot marketplace and new features in the changelog
4. friday - github repos builders shared this week, ranked by stars gained
5. sunday - cleans its lists, grades its own week, pitches up to 3 changes to itself
how it defines "normal": median of someone's last 10 original posts. replies and reposts don't count. no baseline, no flag
who it watches: small accounts with 3+ firsthand grok bot posts in 30 days, the grok bot team, and peers around my size
the rule i care about most: anything it can't confirm comes marked unverified
the team follows lauren tan's setup: one chief of staff in front, one job per bot behind it
meet head of signals (HoS). i messaged my chief of staff, it passed everything on, and HoS created its own daily 8:00 - scans builder accounts for posts running 3x above that person's normal
2. anytime - a post under 3h old already at 3x gets sent to me right away, max 3 a day
3. monday - new templates in the grok bot marketplace and new features in the changelog
4. friday - github repos builders shared this week, ranked by stars gained
5. sunday - cleans its lists, grades its own week, pitches up to 3 changes to itself
how it defines "normal": median of someone's last 10 original posts. replies and reposts don't count. no baseline, no flag
who it watches: small accounts with 3+ firsthand grok bot posts in 30 days, the grok bot team, and peers around my size
the rule i care about most: anything it can't confirm comes marked unverified
15
2
79
6,345
i got 1.45M impressions on x in the last month, using my grok @bot team
the team follows lauren tan's setup: one chief of staff in front, one job per bot behind it
meet head of signals (HoS). i messaged my chief of staff, it passed everything on, and HoS created its own daily 8:00 - scans builder accounts for posts running 3x above that person's normal
2. anytime - a post under 3h old already at 3x gets sent to me right away, max 3 a day
3. monday - new templates in the grok bot marketplace and new features in the changelog
4. friday - github repos builders shared this week, ranked by stars gained
5. sunday - cleans its lists, grades its own week, pitches up to 3 changes to itself
how it defines "normal": median of someone's last 10 original posts. replies and reposts don't count. no baseline, no flag
who it watches: small accounts with 3+ firsthand grok bot posts in 30 days, the grok bot team, and peers around my size
the rule i care about most: anything it can't confirm comes marked unverified
the team follows lauren tan's setup: one chief of staff in front, one job per bot behind it
meet head of signals (HoS). i messaged my chief of staff, it passed everything on, and HoS created its own daily 8:00 - scans builder accounts for posts running 3x above that person's normal
2. anytime - a post under 3h old already at 3x gets sent to me right away, max 3 a day
3. monday - new templates in the grok bot marketplace and new features in the changelog
4. friday - github repos builders shared this week, ranked by stars gained
5. sunday - cleans its lists, grades its own week, pitches up to 3 changes to itself
how it defines "normal": median of someone's last 10 original posts. replies and reposts don't count. no baseline, no flag
who it watches: small accounts with 3+ firsthand grok bot posts in 30 days, the grok bot team, and peers around my size
the rule i care about most: anything it can't confirm comes marked unverified
15
2
79
6,345
i plugged my grok bot team into obsidian, and now every task they do leaves a note i can actually read
the team is based on lauren tan's chief of staff setup: one bot routes tasks, the others each own one job. the gap in it: chief of staff knows who got a task, not whether it covers that gap
how it works:
1. every task gets a note. chief of staff creates it when it routes the task: owner, what done looks like, deadline
2. the bot that did the work writes back into the same note. what it did, link to the PR or post, what it's unsure about
3. test bot adds a verdict for anything code. pass, fail, or "check this yourself"
4. one weekly review note. every friday chief of staff fills it with all tasks from the week, [N] last week
5. graph view shows who's overloaded. [bot] had [N] links, the others had a few each
6. it's plain markdown on my disk, so the history doesn't live inside one chat
i used to read bot chats every friday. now i read one note
the team is based on lauren tan's chief of staff setup: one bot routes tasks, the others each own one job. the gap in it: chief of staff knows who got a task, not whether it covers that gap
how it works:
1. every task gets a note. chief of staff creates it when it routes the task: owner, what done looks like, deadline
2. the bot that did the work writes back into the same note. what it did, link to the PR or post, what it's unsure about
3. test bot adds a verdict for anything code. pass, fail, or "check this yourself"
4. one weekly review note. every friday chief of staff fills it with all tasks from the week, [N] last week
5. graph view shows who's overloaded. [bot] had [N] links, the others had a few each
6. it's plain markdown on my disk, so the history doesn't live inside one chat
i used to read bot chats every friday. now i read one note
4
1
36
1,320