Olivia
Online
DmitrAI
@Dmitr_AI
"AI-powered content & creator tools
Scripts, voice, metadata — what I actually use
Sharing the stack, not the hype"
Joined May 2021
530 Following
233 Followers
379 Posts
Pinned Tweet
Nine days of silence in the chat. Nine days of it building a business. $11,000 waiting when he finally looked.
Nine out of ten prompt for five minutes and close the tab. He built the one that runs for days — and every run leaves the next one sharper.
Fable 5 — Mythos-class, one rung above Opus. Not a five-minute chat. It plans across days, hands tasks to sub-agents, and checks its own work with its eyes.
One agent writes — a separate one grades it. The maker is never the judge. It doesn't hide a failure: fail → work out why → verify → distill a rule → next time it reads the rule instead of guessing again.
A business runs on top. A Routine fires at 7am on a trigger — laptop shut, and it's already re-running yesterday's data, distilling a new rule into a Skill, dropping the digest. It gets sharper while he sleeps.
And every morning the number on his phone is bigger than yesterday's.
$10 per million input tokens, $50 per output — you pay for work, not a subscription ↔ $11,000 in nine days he never touched it.
Build the system this weekend. STATE.md written before bed, read at start — the next session continues instead of restarting from zero. After that it compounds while you sleep.
Most people see a chat that answers. He sees a worker that gets smarter every night, never forgets, and never bills you for learning. Anyone can build it.
You just read this — and you'll still just save the post.
https://t.co/hmpJtXuXCq
💳 You pay $20 a month to keep doing the work yourself. He pays per token — to not do it at all.
Your subscription buys you the right to sit in the chat. His token bill buys him the result while he sleeps.
At the end of the month the number can be the same. Only one thing differs: who did the work — you or the system.
He doesn't spend more than you. He just stopped being the one the subscription pays for.
📌 You just read this — and your subscription just renewed so you can sit down and do it yourself again tomorrow.
Nine days of silence in the chat. Nine days of it building a business. $11,000 waiting when he finally looked.
Nine out of ten prompt for five minutes and close the tab. He built the one that runs for days — and every run leaves the next one sharper.
Fable 5 — Mythos-class, one rung above Opus. Not a five-minute chat. It plans across days, hands tasks to sub-agents, and checks its own work with its eyes.
One agent writes — a separate one grades it. The maker is never the judge. It doesn't hide a failure: fail → work out why → verify → distill a rule → next time it reads the rule instead of guessing again.
A business runs on top. A Routine fires at 7am on a trigger — laptop shut, and it's already re-running yesterday's data, distilling a new rule into a Skill, dropping the digest. It gets sharper while he sleeps.
And every morning the number on his phone is bigger than yesterday's.
$10 per million input tokens, $50 per output — you pay for work, not a subscription ↔ $11,000 in nine days he never touched it.
Build the system this weekend. STATE.md written before bed, read at start — the next session continues instead of restarting from zero. After that it compounds while you sleep.
Most people see a chat that answers. He sees a worker that gets smarter every night, never forgets, and never bills you for learning. Anyone can build it.
You just read this — and you'll still just save the post.
🧬 He's no longer afraid the system will fail. Every failure makes it smarter — permanently.
💸 Over a month, the system that learns from its own mess-ups pulled $19,300 — and not one bug repeated twice.
A normal agent fails → you fix it → tomorrow it fails the same way. Fable 5 fails → works out why → verifies the cause → distills a rule → next time it reads the rule instead of hitting the same wall. 🔁
⚖️ The one who writes is never the judge: a separate agent confirms the lesson is true before it becomes a rule. So no false conclusion ever enters memory. ✅
📈 That's why it doesn't just run — it gets cheaper and sharper every week. For it, mistakes are an asset, not a cost.
🌙 He sleeps while the system turns yesterday's failure into tomorrow's rule. Most people fix the same bug for the hundredth time. He built the thing that fixes it once — and remembers forever.
📌 You just read this — and tomorrow you'll fix what you already fixed.
⚔️ He doesn't check the work anymore. The system argues with itself until only the right answer survives.
💸 $14,000 last month — and nothing shipped that couldn't withstand its own attack.
Nine out of ten run Fable 5 like a chatbot: one prompt, one answer, close the tab. He runs it as a workflow — one goal fans out into a swarm, each agent doing a piece and checking the others. 🔁
🧠 Fable 5 — Mythos-class, built for days, not minutes. It plans the work, splits it across sub-agents, and turns them on each other on purpose: one writes → another breaks it → a third rewrites what breaks.
⚖️ The maker is never the judge. Nothing passes because it sounds confident — it passes because a separate agent tried to tear it apart and couldn't. Weak answers die before morning. ✅
🌙 A business runs on top. A Routine fires overnight, runs the whole adversarial loop, and files what survived into a Skill. He wakes to results that already fought for their place — and every morning they land cleaner than the day before.
💰 $10 per million input, $50 per output — you pay for work, not a subscription ↔ $14,000 off a machine that refuses to ship anything it can't defend.
🛠️ Build the system this weekend: one goal, a swarm, a verifier that attacks. After that it argues, decides, and ships while you sleep.
Most people see a chat that agrees with them. He sees a room full of workers that fight until only the truth is left standing.
📌 You just read this — and you'll still just save the post.
https://t.co/hmpJtXuXCq
Fable 5 was built to run for days. You're using it for 5 minutes.
9 out of 10 people run it like a chatbot with a bigger context window — prompt, wait, close the tab. Nothing carries over. Tomorrow starts from zero.
The 1 who wins does one thing: they make it compound. Three parts turn a chat into a system that gets smarter every run 👇
1. A state file. The model forgets every session; a STATE.md doesn't. It records what's done and what's next, so the next run resumes instead of restarting.
2. Skills. Write a lesson into a SKILL.md, not just the chat. Next run reads it automatically. Intent stops evaporating — it stacks.
3. A verifier. A second pass — a separate agent — grades the work. The one who wrote it is never the one who approves it. That's how bad output never becomes a "lesson."
The loop: it fails → works out why → writes the rule → and never makes that mistake again. That's what "self-improving" actually means.
This week: add a STATE.md to one repetitive task and make the agent read it first. That single file is 80% of the effect.
Save this — it's the difference between a tool you restart and a system that compounds. 🔖
https://t.co/hmpJtXuXCq
🚀 Month one: $4,200. Month six: $38,000. Month twelve: $121,000. Same Fable 5, same laptop — it just got smarter every single night.
🌙 While you slept, the system didn't just work — it improved itself. Every run left the next one smarter, every state file grew, every skill sharpened. And the money didn't climb in a line — it went exponential.
Most people run Fable 5 like Sonnet with a bigger context window: prompt, five minutes, close the tab. He gave it a goal — and it started to compound. 🔁
⚖️ The one who writes is never the judge: a separate agent validates, so only truth enters memory. Failures become rules, rules become skills, skills become money. ✅
📈 By year's end he doesn't "work" anymore. The system runs four businesses, hires its own sub-agents, and he just reads the morning number — bigger than yesterday's, every time.
Most people see a five-minute chat. He saw a machine that grows exponentially while he sleeps.
📌 You just read this — and tomorrow you'll open the same Fable 5 for five minutes.
https://t.co/hmpJtXuXCq
This got 8.6M views — and 9 out of 10 who read it are still the loop.
You type a prompt, read the diff, type again. You're the one deciding every step. That's the job a loop takes over.
Here's the smallest loop that replaces you — 4 parts, no swarm 👇
1. One automation. /goal runs until a condition you wrote is true; /loop re-runs on a rhythm. This is the heartbeat — everything else hangs off it.
2. One gate. A test, a type check, or a build that can reject bad output. The one who wrote the code is never the one who grades it — no gate means the agent grading its own homework.
3. One state file. STATE.md. The agent forgets every run; the file doesn't. Tomorrow it resumes instead of restarting from zero.
4. One skill. SKILL.md — write the project context once, read it every run. Without it, the loop re-derives everything from scratch each cycle.
The honest part most threads skip: most people don't need a loop yet. Build one only when the task repeats weekly, a test can verify it, and your budget absorbs the retries. Otherwise a single good prompt still wins.
This weekend: take one task you do by hand every week. Wrap it in /goal, add one gate. Run it once while you watch every step.
Tomorrow I'll drop the exact charter you paste into Claude Code — fill in the brackets and go.
Save this — it's your checklist the day you build your first loop. 🔖
https://t.co/eJB3TgyJwV
📉 Every night his system gets more valuable. Every night your edge gets cheaper.
He sleeps — his worker gets better. You sleep — your lead melts, because someone else is building while you don't.
It's not a race of talent. It's a race of directions: one compounds on its own, the other decays on its own.
In a year the gap isn't about who's smarter. It's about whose curve was climbing while you both looked away.
📌 You just read this — and your curve just ticked down one more day.
The scary part of this post isn't the $11,000.
It's that his system got smarter over those nine days — and you changed nothing in the same nine.
He designed a worker that learns every night. You still open the chat the way you did last year.
The difference isn't the model anymore. It's who built the system and who's still prompting by hand.
You'll save this one too — and tomorrow, nothing again.
👔 You're still an employee in your own business. He promoted himself — now he's the manager and Claude does the work.
💸 Over a month that "department of one manager" pulled $15,600 — and he didn't write a single prompt, only approved the output.
Prompting is doing the work yourself. Loop engineering is managing a worker: you define what "done" means, and it finds the work, does it, checks it, remembers, and repeats. 🔁
⚖️ A second Claude quietly checks the first after every turn — the one who did the work never grades it. LOOP-STATE.md remembers where it stopped. ✅
🌙 He handed off the job overnight and went to sleep. Woke up to work that was done, not "in progress." Boris Cherny stopped prompting back in June; he just did the same.
Most people are looking for how to type prompts faster. He stopped being the one who types.
📌 You just read this — and tomorrow you'll clock in at your own company again.
https://t.co/eAsr7cajuP
🧬 He's no longer afraid the system will fail. Every failure makes it smarter — permanently.
💸 Over a month, the system that learns from its own mess-ups pulled $19,300 — and not one bug repeated twice.
A normal agent fails → you fix it → tomorrow it fails the same way. Fable 5 fails → works out why �� verifies the cause → distills a rule → next time it reads the rule instead of hitting the same wall. 🔁
⚖️ The one who writes is never the judge: a separate agent confirms the lesson is true before it becomes a rule. So no false conclusion ever enters memory. ✅
📈 That's why it doesn't just run — it gets cheaper and sharper every week. For it, mistakes are an asset, not a cost.
🌙 He sleeps while the system turns yesterday's failure into tomorrow's rule. Most people fix the same bug for the hundredth time. He built the thing that fixes it once — and remembers forever.
📌 You just read this — and tomorrow you'll fix what you already fixed.
https://t.co/hmpJtXuXCq
👁️ He doesn't check the system's work. It looks at it with its own eyes — and throws out the broken parts before he even wakes up.
💸 Over a month the system that sees itself pulled $17,800 — and not one broken result reached a client.
Most AI writes and hopes it guessed right. Fable 5 generates → opens the result → looks at it with vision → compares it to what it was supposed to be → discards the junk and redoes it. 🔁
⚖️ The one who made it is never the judge: a separate agent grades it, and nothing ships until the picture matches the goal. ✅
🌙 He sleeps while the system inspects and corrects itself. He wakes to "done," not "almost done."
📈 Most people spend the morning redoing what the AI botched overnight. He has nothing to redo — it checked itself.
Most people see a chat that lies with confidence. He sees a worker that looks at its own work before it calls it finished.
📌 You just read this — and tomorrow you'll check every output by hand again.
https://t.co/hmpJtXuXCq
🌳 You do one task at a time. He runs six agents at once — each in its own tree, and none of them collide.
💸 Over a month that parallel line pulled $16,200 — because six hands do in the same hour what your one hand does.
git worktree gives every agent its own working directory on its own branch. One writes the feature, one refactors, one verifies — and none of them overwrites another's code. 🔁
⚖️ The one who writes is never the judge: a separate agent checks while the rest keep working. STATE.md holds the place for every thread. ✅
🌙 He launched six overnight and went to sleep. By morning — six finished branches and a queue waiting for his sign-off, and he hadn't written a line.
📈 Your ceiling is how much you can review yourself. His is review too — but six threads are prepping the work for him, not one tab.
Most people scale by working longer. He scales by cloning the worker.
📌 You just read this — and you're going back to your single tab.
https://t.co/eJB3TgyJwV
@milesdeutscher Good news we have been waiting for too long 🙌
@starmexxx Good news we have been waiting for too long 🙌
@0xwhrrari Good news we have been waiting for too long 🙌
🌑 Your most valuable shift is the one you sleep through. His is worked by a machine. Yours by no one.
At 3:14 his system closes tasks. At 3:14 you're just asleep — and that's fine. What isn't fine is that your business sleeps with you.
Eight hours × 30 nights = 240 hours a month where his gets work done and yours gets silence and zero.
He doesn't work harder than you by day. He just stopped giving the nights away.
📌 You just read this — and tonight you'll hand those eight hours to no one again.
💓 At 3am his loop wakes itself up. At 3:14 it closes three tasks. At 3:15 it goes back to sleep. He doesn't know — he's asleep too.
💸 Over a month that night rhythm pulled $8,900 — off tasks he never touched by hand.
An automation is the heartbeat of a loop. It beats on a schedule: nightly CI triage, weekly dependency-bump PRs, lint-fix on every commit. You sleep — it works.
🌙⚖️ /goal runs until the condition is true, and a separate model checks it — the one who wrote the code never grades its own homework. STATE.md remembers where to start tomorrow.
✅📈 $12.37 per accepted change · 128 of them while he slept. Multiply that across months and you'll see why his margin drops while his volume climbs.Most people check every diff by hand till midnight. He built a heart that beats without him — and went to sleep.
📌 You just read this — and tomorrow you'll be the one hitting Enter by hand again.
💼 He hired a worker that gets better every night — and will never ask for a raise.
A human in that role would cost thousands a month, take vacations, and quit within a year — taking everything it learned with them.
His worker gets cheaper every month, never sleeps, never forgets, and never leaves.
🌙 While you're deciding who to hire, his team is already fully staffed — and it fits in a single file.
The difference isn't who's smarter. It's who stopped clocking in himself.
📌 You just read this — and tomorrow you'll clock in again instead of your system.
The scary part of this post isn't the $11,000.
It's that his system got smarter over those nine days — and you changed nothing in the same nine.
He designed a worker that learns every night. You still open the chat the way you did last year.
The difference isn't the model anymore. It's who built the system and who's still prompting by hand.
You'll save this one too — and tomorrow, nothing again.
🌅 You both wake up at seven. His system resumes from yesterday. You start from zero.
He wrote one file before bed — STATE.md. It remembers what's done and what's next. Your progress from yesterday lives only in your head, and it already forgot half of it.
🔁 So for him, every day stacks on top of the last. For you, every day is a blank page.
Over a year that isn't "a little more." It's a gulf: he's on step 365. You're on step one — 365 times in a row.
📌 You just read this — and tomorrow you'll open the chat like yesterday never happened.
Nine days of silence in the chat. Nine days of it building a business. $11,000 waiting when he finally looked.
Nine out of ten prompt for five minutes and close the tab. He built the one that runs for days — and every run leaves the next one sharper.
Fable 5 — Mythos-class, one rung above Opus. Not a five-minute chat. It plans across days, hands tasks to sub-agents, and checks its own work with its eyes.
One agent writes — a separate one grades it. The maker is never the judge. It doesn't hide a failure: fail → work out why → verify → distill a rule → next time it reads the rule instead of guessing again.
A business runs on top. A Routine fires at 7am on a trigger — laptop shut, and it's already re-running yesterday's data, distilling a new rule into a Skill, dropping the digest. It gets sharper while he sleeps.
And every morning the number on his phone is bigger than yesterday's.
$10 per million input tokens, $50 per output — you pay for work, not a subscription ↔ $11,000 in nine days he never touched it.
Build the system this weekend. STATE.md written before bed, read at start — the next session continues instead of restarting from zero. After that it compounds while you sleep.
Most people see a chat that answers. He sees a worker that gets smarter every night, never forgets, and never bills you for learning. Anyone can build it.
You just read this — and you'll still just save the post.
📁 This .claude breakdown hit 12M. Nine out of ten bookmarked it and stayed exactly where they were. He configured three files — and now the folder pays him while he sleeps.
🧠 Not "prompt hacks." Three files that turn a chat into a worker 👇
1. CLAUDE.md — loaded first, every session. Whatever you write, it follows — no reminders. He wrote his business process here once and stopped re-explaining it every day.
2. Stop-hook (exit 2) — Claude can't say "done" until the tests are green. ⚖️ That's his night-shift QA: while he slept, nothing broken reached a client — nothing to redo in the morning.
3. Path-scoped rule — a rule that loads only for its own files. Right context, zero bloat, zero drift. ✅
🌙 He configured the folder on Friday. Skills invoked themselves, hooks held the bar, agents reviewed each other. By Monday — finished tasks and $2,900, and zero "check my work" by hand.
📈 In a month, three files most people just saved became a system pulling $17,400 — because he configured a worker, not a tab.
Most people bookmark the post about the folder. He configured the folder — and went to sleep.
📌 You just read this — and your .claude/ is still empty.
https://t.co/SSSIK3BX4z
📎 Claude just lied to you with a straight face — and no single prompt will catch it. A loop will.
🧠 The guy who built Claude Code said it out loud in June: "I don't prompt Claude anymore." Loops prompt Claude for him.
Ask Claude for a one-page brief. It writes something clean and confident, with sources at the bottom. Some of those sources are invented. It has no idea it lied. The links go nowhere — or somewhere that says the opposite.
⚠️ A single prompt will never see it: Claude stays confident it's right until someone opens the link. That's the quiet way it burns you.
🔁 Now run the same thing as a loop. Add a bar you can measure: every claim needs at least three sources, every link must open to a real page that backs up the claim. Claude writes the brief, then goes link by link — throws out the dead ones, throws out the fakes, finds real replacements — and doesn't stop until every source opens. ✅
That's the whole difference: a prompt does the work. A loop manages the worker.
🌙 He set the bar once and went to sleep. LOOP-STATE.md remembered where it stopped. By morning — three clean client briefs and $3,400, and he'd checked zero links by hand.
📈 In a month, "I don't prompt anymore" became a system pulling $18,600 — off the work he used to babysit twenty minutes at a time.
Most people are still babysitting the thing they wanted to hand off. He stopped being the loop — and became the one who designs it.
📌 You just read this — and tomorrow you'll be pasting the error back into the chat again.
https://t.co/eAsr7cajuP
💓 At 3am his loop wakes itself up. At 3:14 it closes three tasks. At 3:15 it goes back to sleep. He doesn't know — he's asleep too.
💸 Over a month that night rhythm pulled $8,900 — off tasks he never touched by hand.
An automation is the heartbeat of a loop. It beats on a schedule: nightly CI triage, weekly dependency-bump PRs, lint-fix on every commit. You sleep — it works.
🌙⚖️ /goal runs until the condition is true, and a separate model checks it — the one who wrote the code never grades its own homework. STATE.md remembers where to start tomorrow.
✅📈 $12.37 per accepted change · 128 of them while he slept. Multiply that across months and you'll see why his margin drops while his volume climbs.Most people check every diff by hand till midnight. He built a heart that beats without him — and went to sleep.
📌 You just read this — and tomorrow you'll be the one hitting Enter by hand again.
https://t.co/eJB3TgyJwV
Last Seen Users on Sotwe
Trends for you
Most Popular Users
Elon Musk
@elonmusk
241M followers
Barack Obama
@barackobama
119.2M followers
Cristiano Ronaldo
@cristiano
111.8M followers
Donald J. Trump
@realdonaldtrump
111.8M followers
Narendra Modi
@narendramodi
107M followers
Rihanna
@rihanna
98M followers
NASA
@nasa
92.2M followers
Justin Bieber
@justinbieber
91.2M followers
KATY PERRY
@katyperry
88.3M followers
Taylor Swift
@taylorswift13
82.2M followers
Lady Gaga
@ladygaga
73.7M followers
Virat Kohli
@imvkohli
70.9M followers
Kim Kardashian
@kimkardashian
70.1M followers
YouTube
@youtube
68.7M followers
Bill Gates
@billgates
64.3M followers
Neymar Jr
@neymarjr
63.8M followers
The Ellen Show
@theellenshow
62.4M followers
CNN
@cnn
61.8M followers
Selena Gomez
@selenagomez
61.4M followers
X
@x
60.8M followers