Reading markets, AI. Ignoring the noise
Joined July 2026
- Tweets147
- Following37
- Followers39
- Likes117
Pinned Tweet
Gemini fans, look away🤓
Same prompt: "3D human anatomy atlas".
Gemini 3.8 Flash: floating dots.
GPT-6 Astra: a full human body. Muscles, bones, organs. Click any part and it tells you what it is.
Med students would love this. It runs in a browser tab.
Prompt in the first reply 👇
Gemini or Astra? Reply 1 or 2
Aimore retweeted
🚨 OpenAI is reportedly preparing to buy custom AI chips directly from its fiercest rival Google
The Information reports that OpenAI is breaking its near exclusive dependence on Nvidia GPUs by purchasing Google TPUs to power its next compute cluster
OpenAI has already begun hiring engineers to optimize its model architectures for Google hardware, while Google prepares to offer TPUs on Oracle Cloud
Paying your biggest competitor billions for silicon proves how desperate frontier labs are to escape Nvidia eighty percent gross margins and compute bottlenecks
The AI chip wars just turned into strange bedfellows
Source: The Information, Oct 28, 2026 (reported, not confirmed by the company)
Would you buy hardware from your top competitor to escape the Nvidia monopoly?
Aimore retweeted
Everyone is bragging about launching 300 AI agents👀
A new security report looked at what happens next.
76% of enabled agents hadn't run in 60 days.
44% hadn't run in 180.
70% touch a sensitive system.
99.4% of those have no verified DLP.
"When an agent stops being used, nobody removes anything, because in most companies nobody owns it."
How many agents have you launched and never switched off?
Everyone is bragging about launching 300 AI agents👀
A new security report looked at what happens next.
76% of enabled agents hadn't run in 60 days.
44% hadn't run in 180.
70% touch a sensitive system.
99.4% of those have no verified DLP.
"When an agent stops being used, nobody removes anything, because in most companies nobody owns it."
How many agents have you launched and never switched off?
Source: Arrakis Security, 2026 Autonomous Agents Threat Report (vendor data from customer environments, 12 weeks, "thousands" of agents, exact count not disclosed). Coverage: helpnetsecurity.com/2026/10/…
Aimore retweeted
I'M F*CKING LOSING MY MIND OVER GROK BOT × BUZZCORE
I gave Grok Bot $5.38 and four ants. 30 seconds later the balance read $14,592.
every trade ran on pumpfun, and every one went through the same steps: scout finds it, sniper gets in, curve rides it, grad takes it out. no ant skips its job.
40 trades. 33 wins, 7 losses.
the big closes:
> 00:05, on PYTH, +$96
> 00:11, on BONK, +$392
> 00:16, on W, +$804
> 00:23, on MEW, +$612
> 00:27, on WIF, +$10,486
look at that last one.
the first four big closes made $1,904. WIF alone made $10,486. that's 72% of the whole session from one trade.
right before WIF the desk sat at 763x. one close later it ended at 2,712x.
most people would've closed WIF at +100% and called it a good day. the ants held it.
so be honest: at what % would you have sold WIF? 👇
send this to the friend who always sells too early.
Aimore retweeted
🦾 Tesla has officially kicked off early production on its high volume Optimus line at the Fremont factory
These are the first production quality Gen 3 humanoids built on a commercial manufacturing layout rather than handcrafted in a lab
Tesla is targeting external customer deliveries by late 2027, powered by their upcoming custom AI5 silicon architecture
The biggest engineering hurdle remains the hands, where dozens of degrees of freedom suffer extreme mechanical wear and tear under continuous factory shifts
Scaling embodied AI is no longer a software demo, it is a brutal industrial manufacturing race
Source: Tesla Q3 Earnings Call, Oct 2026
Will Tesla solve commercial humanoid mass production before dedicated robotics startups?
Aimore retweeted
Same prompt. DeepSeek V4.1 Flash: 19 seconds. Claude Opus 5.5: 5m 23s💤
The task: a 15-second Apple-style launch animation for a flying Tesla Roadster.
DeepSeek: a car sweeps in over a glowing title.
Opus 5.5: 5 labeled scenes, with a scrubber and a loop.
My take: for a first draft I’d take the 19 seconds. For something I’d actually post, I’d wait the 5 minutes.
Caveat: one prompt, one run each, different settings. The waiting part is sped up 3x, and the timers are the original creator’s overlays.
Poll: What do you want first?
The 19-second draft / The 5-minute film
Same prompt. DeepSeek V4.1 Flash: 19 seconds. Claude Opus 5.5: 5m 23s💤
The task: a 15-second Apple-style launch animation for a flying Tesla Roadster.
DeepSeek: a car sweeps in over a glowing title.
Opus 5.5: 5 labeled scenes, with a scrubber and a loop.
My take: for a first draft I’d take the 19 seconds. For something I’d actually post, I’d wait the 5 minutes.
Caveat: one prompt, one run each, different settings. The waiting part is sped up 3x, and the timers are the original creator’s overlays.
Poll: What do you want first?
The 19-second draft / The 5-minute film
Aimore retweeted
it is becoming f*cking dangerous
Official OpenAI documentation for Codex: "dots" agents are here, and they write code autonomously in the cloud while your laptop is completely powered off
The idea: your personal agent runs 24/7 on GPT-6 Astra in its own cloud environment with a dedicated browser, delegating builds directly to Codex CLI
Next-level autonomy with zero waiting: you assign a task, close your laptop, and the agent continues shipping code while pinging your Slack only when approval is needed
You launch it directly from your terminal like this:
codex --dot @alfred --autonomous
here's the split:
→ your dot runs 24/7 inside a secure cloud container powered by GPT-6 Astra
→ operates its own browser and task memory, carrying out research even while you are offline
→ autonomously schedules and executes development workflows in your repository via Codex Micro
→ alerts you in Slack or Teams only when an architectural decision strictly requires human judgment
The dot only reaches out at three critical moments:
→ when an external portal requires private credential input via the Take over action
→ when an ongoing run risks exceeding allocated token budgets or deadlines
→ before pushing final production diffs to upstream branches to request human sign-off
Astra reasons · Dot coordinates · Codex builds
Process isolation executes one layer down: task state and execution memory sync across continuous cloud containers, so background workers run without human prompt latency and yield control in milliseconds when you hit "Take over"
the full tree
> persistent cloud dot agent on GPT-6 Astra maintains long-term memory across sessions
> direct integration with Codex CLI enables remote delegation of complex engineering tasks
> Tasks & Memory engine anchors project context so architectural decisions survive reboots
> cross-platform bridges route routine development output to logs and critical gates directly to Slack
> Take over controls let developers seize command of the cloud browser or terminal in one click
paste the tree and this prompt into Codex ↓
"Rebuild my Codex setup around this tree:
• Configure persistent dot agent on GPT-6 Astra with cloud container and headless browser access
• Set up workspace bridges to route routine dev logs to files and critical architectural gates to Slack
• Delegate multi-file coding tickets to local Codex CLI environments with automatic task locking
• Add one rule to AGENTS.md: pause autonomous execution and request manual take-over before modifying external infrastructure or security configs
Show every change as a diff first. No edits until I say go"
developers.openai.com/codex/…
save it, it'll come in handy
Aimore retweeted
For every hour Anthropic’s staff works, its AI agents work six. 🫥
That’s by August, per Anthropic. Over the same period the junior share of its technical departments fell from 21% to 14%, and the junior share of technical hires from 27% to 9%.
Caveats from the report itself: human hours are assumed (headcount x 40 h), agents run in parallel, and the authors say this is descriptive, not proof of cause.
My take: the number to watch isn’t 6:1. It’s who gets hired when one engineer can direct a team of agents.
Video: my simulation of the report’s numbers, not live telemetry.
What happens to junior developers?
For every hour Anthropic’s staff works, its AI agents work six. 🫥
That’s by August, per Anthropic. Over the same period the junior share of its technical departments fell from 21% to 14%, and the junior share of technical hires from 27% to 9%.
Caveats from the report itself: human hours are assumed (headcount x 40 h), agents run in parallel, and the authors say this is descriptive, not proof of cause.
My take: the number to watch isn’t 6:1. It’s who gets hired when one engineer can direct a team of agents.
Video: my simulation of the report’s numbers, not live telemetry.
What happens to junior developers?
Aimore retweeted
THIS IS A PURE F*CKING GOLDMINE
stop running every Claude Code task on one model.
The idea: Opus thinks, Sonnet types, Haiku reads. Each model only touches the part it's actually good at.
One feature, start to commit: Sonnet burned 356k tokens doing the typing. Opus showed up 4 times for 60k. Three Haiku agents read the codebase in parallel. And the bug that would've shipped got caught before the commit.
You just launch Claude Code like this:
claude --model opusplan
here's the split:
→ plan mode runs on Opus 5.5: architecture, approach, nothing gets touched
→ you approve the plan, execution switches to Sonnet 5.5: writes code, runs tests
→ read/grep goes to Haiku explorers in parallel, they return file:line, not essays
→ same error twice? an Opus advisor subagent reads the trace and sends back one line of guidance
→ diff ready → an Opus judge subagent reviews it with one rule: "assume it's wrong"
→ judge finds something → back to Sonnet. judge passes → commit
Opus only shows up at four moments:
→ the plan
→ when the loop is stuck
→ the review
→ the re-review
Haiku reads. Sonnet types. Opus judges.
In my run the judge caught "refund > charge accepted" at refunds.ts:57. Tests were 62/62 green. Green tests don't mean correct code.
the full tree
> /model opusplan → Opus in plan mode, Sonnet after you approve
> Explore subagent → fast read-only search
> .claude/agents/advisor.md → model: opus, read-only tools
> .claude/agents/judge.md → model: opus, read-only tools, "assume it's wrong"
> /compact between big phases so context stays clean
paste this prompt into Claude Code ↓
"Set up my routing:
Create two subagents in .claude/agents/.
advisor: model opus, tools Read, Grep, Glob only. Use it when the same error happens twice or when choosing between approaches. Return one short recommendation with file:line.
judge: model opus, tools Read, Grep, Glob only. Review every diff before commit. Assume it's wrong. One line per finding with file:line. Pass only if there are zero findings.
Use Explore for searching the codebase.
Show me both files before saving."
↳
code.claude.com/docs/en/mode…
code.claude.com/docs/en/sub-…
save it, your next feature ships with a reviewer built in
Aimore retweeted
Gemini 4 Argon is 5x cheaper than GPT-6 Astra per token. From Dec 1 it costs more per task.
The number to use instead, plus a 30 minute test you can run for about $8.
What do you check first: price per token or cost per task?
Gemini 4 Argon is 5x cheaper than GPT-6 Astra per token. From Dec 1 it costs more per task.
The number to use instead, plus a 30 minute test you can run for about $8.
What do you check first: price per token or cost per task?
Aimore retweeted
A planet may have been hiding in NASA data for 7 years
One developer with Claude Code may have just found it
Pavel Rabtsevich isn't an astronomer. He ran public TESS data through Claude Code and spotted a star 116 light-years away that dims by 0.05% for two hours every 3.18 days.
If a planet causes it, it's about 1.4x Earth's size. The same dip shows up in 2018, 2020 and 2025 data.
Still a candidate, not a confirmed planet. TESS will watch the star again from October 31.
My take: the data was public all along. What changed is how fast one person can test an idea. AI won't replace astronomers, but it lets outsiders ask better questions.
Would you trust this discovery once NASA confirms it?
Aimore retweeted
SHE KNOWS EXACTLY WHAT SHE'S DOING WITH THAT POPSICLE
Girl in a white tank and denim shorts leans on the kitchen counter, red popsicle in hand.
She takes her time.
A woman walks in. A man follows right behind her.
He sees the girl. Eyes wide. Jaw on the floor.
She glances sideways, takes one more slow lick, and smiles. Zero guilt.
None of it is real.
Not the girl. Not the kitchen. Not the sitcom.
The hard part isn't her. It's his face.
Be honest: who did you watch first, her or him?
Aimore retweeted
"AI is not a model. AI is a whole stack of technologies." - Jensen Huang, in Taiwan last week with Lex Fridman.
the sentence is the bottom-up version of the thesis in my land grab article.
Jensen frames the stack from the compute end: a few trillion dollars of compute, hundreds of billions per company, Taiwan as the physical bottleneck.
the Grok Bot Template Marketplace is the top of that same stack. role-based bots that package a workflow and sell it ready-to-use.
compute at the bottom. templates at the top. everything else is the layers nobody fully owns yet.
if Jensen is right that AI is a whole stack and leadership is happening in different places up and down it, the template layer is the one that's still unclaimed by a top 100.
that's the land grab window. 69 bots deep today.
the full math and the 5 templates to ship this week:
👇
teapawt.org/roxitsu/status/2109599…
A planet may have been hiding in NASA data for 7 years
One developer with Claude Code may have just found it
Pavel Rabtsevich isn't an astronomer. He ran public TESS data through Claude Code and spotted a star 116 light-years away that dims by 0.05% for two hours every 3.18 days.
If a planet causes it, it's about 1.4x Earth's size. The same dip shows up in 2018, 2020 and 2025 data.
Still a candidate, not a confirmed planet. TESS will watch the star again from October 31.
My take: the data was public all along. What changed is how fast one person can test an idea. AI won't replace astronomers, but it lets outsiders ask better questions.
Would you trust this discovery once NASA confirms it?
Aimore retweeted
ELON MUSK MADE $22 MILLION AT 27. HIS FIRST BIG PURCHASE WAS A $1 MILLION McLAREN. HIS BIGGEST FEAR? BECOMING SOMEONE WHO DIDN’T DESERVE IT
In 1999, a truck pulled up outside a modest apartment at 7 AM. Inside was a McLaren F1, one of just 62 ever built. Musk had just made a fortune selling Zip2 to Compaq for roughly $307 million.
His personal share was around $22 million.
For most people, that would have been the finish line. For Musk, it was the beginning of a different game.
He remembered the days when he showered at the YMCA and slept on his office floor. Now, standing beside his fiancée Justine and a million-dollar supercar, he was worried about what wealth might do to him.
“I’m afraid we’ll become spoiled brats,” was the concern behind his words. He understood that money could buy private islands, luxury, and a life without limits. But that wasn’t what he wanted to build.
Instead, he committed the vast majority of his fortune to X.com, an online banking startup that few people believed would work.
He described it like a poker game: rather than taking his winnings off the table, he kept putting almost everything back into the next hand.
X.com eventually became part of PayPal. When PayPal was sold, Musk walked away with roughly $180 million and directed much of that money into SpaceX and Tesla.
The McLaren, meanwhile, had its own legendary moment. In 2000, Musk was driving with Peter Thiel on the way to pitch Sequoia Capital when the car crashed and went airborne. Neither man was seriously hurt, and the uninsured supercar was badly damaged.
The numbers tell the story. That $1 million car would be worth tens of millions today. The roughly $22 million Musk earned from Zip2 became the foundation for companies that reshaped entire industries.
He could have bought an island and spent the rest of his life enjoying the winnings. Instead, he kept betting on ideas most people thought were too ambitious.
The real difference wasn’t how much money Musk made. It was what he chose to do with it.
The McLaren was a luxury purchase. Everything that followed was a wager on the future.
Aimore retweeted
Anthropic just let Claude build its own team of 1,000 agents.
One prompt in. A lead agent wrote the plan, split it into phases and spawned the rest. 85 hours later it's still running.
Visualized in the video. The prompt to run it on your own project is pinned, full workshop in the quote.
Anthropic engineers just dropped a 2-hour masterclass on making Claude run its own team of agents:
"Feel free to be brave, and use up to 1,000 agents." – still running 85 hours later.
6:44 → Flows: Claude designs its own team of agents, up to 1,000
39:01 → why they don't tell Claude its budget: "it gets anxious"
46:17 → one prompt, 1,000 agents, an operating system, 85 hours and counting
1:12:02 → the 3 ways a single agent fails: laziness, self-bias, goal drift
1:22:24 → the sandbox only holds fake keys, real ones get swapped in on the way out
1:27:07 → humans click approve 97% of the time, so turn on auto mode
your weekend watch. then learn how to architect your own team of agents in the quoted article↓