AI researcher Prompting, creating, automating
- Tweets 914
- Following 154
- Followers 1,355
- Likes 2,606
Sam Altman (CEO of OpenAI):
"Dots can do what GrokBot and Muse can`t. I run 10+ Dots 24/7, even when I sleep"
he spent 20 minutes searching Slack for something, couldn't find it, decided he'd imagined it, and gave up
the next day his agent came back: "I kept thinking about it overnight and I found it for you"
"it just sat there very patiently looking for something to do to be helpful to me"
his line on it: I would have given up on that
watch now, then read how to build Dots team in article below
10
10
151
67,336
I gave Opus 5.5 a team of Dots..
They built a motion design studio inside ChatGPT
One prompt. Four agents. A complete motion design workflow
Full prompt below. Enjoy:
OpenAI just launched dots - agents that work for you 24/7
now you can close your laptop and your dots keep working without you
I prepared some useful prompts for your dot
save now, send to your Dot
3
3
21
1,148
OpenAI just launched dots - agents that work for you 24/7
now you can close your laptop and your dots keep working without you
I prepared some useful prompts for your dot
save now, send to your Dot
Introducing dots, powered by GPT-6 Astra.
Remarkably capable, always-on agents built to handle everything.
8
1
34
7,171
OpenAI just launched dots - agents that work for you 24/7
now you can close your laptop and your dots keep working without you
I prepared some useful prompts for your dot
save now, send to your Dot
Introducing dots, powered by GPT-6 Astra.
Remarkably capable, always-on agents built to handle everything.
8
1
34
7,171
1) Create a 24/7 coding team with four agents:
Architect plans the smallest reliable solution.
Builder implements scoped changes.
Reviewer checks security, quality, and edge cases.
Tester runs tests and verifies the final result.
Give every task an owner, clear acceptance criteria, and evidence of completion. Keep handoffs automatic, but ask me before deploying, merging, spending money, or making destructive changes.
1
391
2) Create a Motion Designer Dot agent that turns my ideas, scripts, or static designs into production-ready animations.
It should define the visual concept, storyboard, transitions, timing, typography, camera movement, and sound cues - then produce an implementation brief for After Effects, Rive, Framer Motion, or Remotion.
Show me a short creative direction first. Ask for approval before generating the final animation.
2
284
Anthropic just dropped Sonnet 5.5 and here is full breakdown
"Sonnet 5.5 is way faster and cheaper for coding. It will blow your mind"
Anthropic says Sonnet 5.5 can reach Opus-level performance on some evaluations while costing significantly less
early testers are already reporting fewer tool calls and faster iterations
Anthropic basically made Sonnet feel like Opus without the Opus price
5
2
20
990
Jev + Opus 5.5 is insane for autonomous coding team..
send this single prompt to Claude Opus 5.5 and connect Jev to turn it into a complete engineering team:
lead, architect, builder, reviewer and tester
then let Jev gate the big decisions - when to build, investigate, simplify, ask you, or stop.
5
7
27
1,115
Diogo Almeida just shared 30 Jev + Opus 5.5 prompts for coding agents
30 prompts. 3 roles. your agent stops burning Opus on work a classifier could decide in 0.1s
the loop:
task → Jev decides in ~0.1s → Opus 5.5 builds → Jev gates the result → ship
section A - the router. spend Opus only where it pays
→ Jev tags every task trivial / standard / hard. only hard wakes Opus
→ route by trust, not difficulty. secrets/ infra/ auth/ → Opus. docs → cheap model
→ one route per task. hopping Opus → cheap → Opus reloads the whole context
section B - the bouncer. nothing runs without a typed yes
→ Jev reads the script before bash runs. allow / ask / deny per line, not per command name
→ task spend over $2 → pause and show the plan first
section C - the shipper. review less, merge more
→ Jev scores risk per file. Opus deep-reviews only the top 20%
→ failures sorted: real bug / flaky / env. only bugs reach Opus
→ confidence ≥ 0.85 + green CI → auto-merge. anything else → human
Jev decides fast - Opus 5.5 builds it
all 30 prompts in the video ↓ send them to your Claude Code and let it rebuild your harness
Jev Engineers just dropped Jev and it splits agent architecture in half: one layer decides, another executes
What Jev Engineering is, how to set it up and where it prints money in my 12-page synthesis:
step 1 → stop using one model for everything. LLMs write. Jev decides. Code acts. Three layers, three failure modes, zero ambiguity on what broke.
step 2 → replace free-form outputs with three typed primitives. Choice picks one winner. Score ranks on a rubric. Noul returns probability of yes. Your code reads numbers, not paragraphs.
step 3 → state is evidence, not a prompt. Feed Jev structured snapshots: goal, verified facts, live options, previous actions. Rebuild after every external change.
step 4 → batch everything. 13 questions against one GDPR article: one call took 0.27s for $0.0005. Same questions sequentially: 2.71s for $0.006. That is 10x speed and 12x cost.
step 5 → five anti-patterns kill Jev deployments. Prompt-as-state. Choice-as-multi-label. DONE without proof. Stale action menus. Confidence treated as truth.
the result: every bounded fork in your agent stack becomes a millisecond decision with a probability you can audit, gate and replay.
Copy the complete 12-page Jev blueprint - then read full 10-step roadmap below ↓
4
2
23
1,162
TypeSafe just showed how to put Jev inside a Python app
"stop building a full AI agent when all you need is one decision"
the workflow is simple:
Python script
→ send the state
→ ask one focused question
→ get a typed result
→ let your code decide what happens next
you can use Jev for things like:
• bug triage
• routing
• classification
• review decisions
the crazy part:
you don't need to give Jev control of whole application
make the model answer one question your code can't answer with a simple rule
save now - then read full Jev playbook below
Jev Engineers just dropped Jev and it splits agent architecture in half: one layer decides, another executes
What Jev Engineering is, how to set it up and where it prints money in my 12-page synthesis:
step 1 → stop using one model for everything. LLMs write. Jev decides. Code acts. Three layers, three failure modes, zero ambiguity on what broke.
step 2 → replace free-form outputs with three typed primitives. Choice picks one winner. Score ranks on a rubric. Noul returns probability of yes. Your code reads numbers, not paragraphs.
step 3 → state is evidence, not a prompt. Feed Jev structured snapshots: goal, verified facts, live options, previous actions. Rebuild after every external change.
step 4 → batch everything. 13 questions against one GDPR article: one call took 0.27s for $0.0005. Same questions sequentially: 2.71s for $0.006. That is 10x speed and 12x cost.
step 5 → five anti-patterns kill Jev deployments. Prompt-as-state. Choice-as-multi-label. DONE without proof. Stale action menus. Confidence treated as truth.
the result: every bounded fork in your agent stack becomes a millisecond decision with a probability you can audit, gate and replay.
Copy the complete 12-page Jev blueprint - then read full 10-step roadmap below ↓
8
9
38
2,650