hyperfixation on AI | always dyor
Mascotte, FL
- Tweets 23,434
- Following 863
- Followers 8,824
- Likes 96,969
OpenAI’s most underrated GPT-6 builder feature: let the model write the tool loop
GPT-6 Astra can write JavaScript that calls your tools, joins their results and returns a compact output
it’s called Programmatic Tool Calling
read multiple sources → group → filter → rank → return the useful result
a concrete build:
feature requests + product usage + effort estimates → five feature candidates with the evidence attached
the repeatable data work runs in code
the model uses the result to explain what deserves attention
enable it in the Responses API:
{"type":"programmatic_tool_calling"}
opt eligible functions in:
"allowed_callers": ["programmatic"]
give every function a clear input schema and a structured output_schema
define the scoring rule, required evidence and stopping condition
> OpenAI runs the generated program
> your server still executes your function tools
> return each result with its original call_id and caller, then continue until the final assistant message
use this for predictable read-only joins, filtering and aggregation
keep adaptive judgment and approval-sensitive writes in direct calls
the goal: useful evidence reaches the model without every intermediate payload filling the context
start with one bounded stage. measure answer quality, tokens, latency and cost
save this, then build your solo company with Dots ⭣
7
8
45
1,823
holy sh*t, Sam Altman just shared his personal Dots setup
he kept feeding his dot rough notes about a feature
it built 5–6 versions. he gave feedback and kept iterating
the clever part:
you can start building before you know exactly which version you want
here’s the workflow I’d steal:
01 capture the raw idea
notes + sketches + voice memos → one clear problem
02 explore different implementations
same problem → different layouts, interactions and tradeoffs
03 make the options real
runnable prototypes + the core interaction working in each
04 pick a direction and keep going
your feedback → the next build
one messy idea → several working directions → one feature worth shipping
i mapped the full setup + article about Dots below ↓
10
7
75
5,109
holy sh*t, someone mapped an entire company’s workflows for OpenAI Dots
every solo builder can steal the playbook
100+ templates, plugin collections, workflow blueprints and custom rules across engineering, research, marketing, sales and operations
the workflows to steal:
→ turn customer feedback into proposed backlog issues
→ turn bug reports into reproducible problems and reviewable fixes
→ track competitor changes with a source-backed brief
→ turn meeting notes into decisions, owners and deadlines
→ turn product docs into launch copy and campaign assets
→ prepare sales calls using CRM and inbox context
→ surface release blockers from GitHub and your project tracker
→ triage your inbox into priorities and draft replies
the architecture:
templates define the job → plugins connect the tools → workflows connect the steps → rules guide approvals
the part solo builders should steal:
1. pick one recurring bottleneck
2. connect the apps it needs
3. define the deliverable
4. set the approval boundaries
5. schedule the next run
start with a weekly project pulse:
what shipped → what broke → what’s blocked → what needs your decision
give your Dot a clear job, the right context and a repeatable workflow
save this, then build your solo company with Dots ⭣
26
23
171
13,244
beamnxw ./ retweeted
Andrej Karpathy just dropped a useful upgrade for your next AI chat
ask the model to create something that helps you understand its answer
start with Standard ASD-STE100 (Simplified Technical English) for cleaner writing:
> short sentences, consistent terms, one instruction per sentence
then choose the output:
→ diagram for the relationships
→ interactive HTML for the variables
→ custom explainer video for the process
Karpathy is especially bullish on videos generated for the exact topic you want to understand
the interesting part: these explanations can be disposable
a tiny app for one question
a simulation for one confusing mechanism
a video you watch once
use it, understand the idea, move on
your next explanation could be something you can explore
send this to your AI, then build your company with Dots ⭣
18
14
98
7,285
Andrej Karpathy just dropped a useful upgrade for your next AI chat
ask the model to create something that helps you understand its answer
start with Standard ASD-STE100 (Simplified Technical English) for cleaner writing:
> short sentences, consistent terms, one instruction per sentence
then choose the output:
→ diagram for the relationships
→ interactive HTML for the variables
→ custom explainer video for the process
Karpathy is especially bullish on videos generated for the exact topic you want to understand
the interesting part: these explanations can be disposable
a tiny app for one question
a simulation for one confusing mechanism
a video you watch once
use it, understand the idea, move on
your next explanation could be something you can explore
send this to your AI, then build your company with Dots ⭣
18
14
98
7,285
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks:
Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better:
Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better:
Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better:
Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work!
In summary:
- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding.
- Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
1
5
372
Standard: asd-ste100.org
GitHub skill: github.com/danyuchn/asd-ste1…
1
6
204
beamnxw ./ retweeted
OpenAI Dots is f*cking brilliant for building a 24/7 AI company...
here’s the bigger architecture I’d build around it:
one founder. one dot powered by GPT-6 Astra. 15 specialist jobs connected by work packets
the system has four loops:
BUILD
feedback + support repros → verified evidence → scoped spec → code branch → tested PR
LAUNCH
approved changes → explainers + demo clips + documentation → launch pack
REVENUE
account context → working POC → proposal → follow-up draft
objections and missing features go back into research
OPERATIONS
support triage + invoice drafts + status tracking → one queue of decisions for the founder
the connections are where this gets useful:
a support ticket can become a repro, a patch, updated docs and an answer draft
a finished feature becomes both launch material and proof for the next proposal
a sales objection becomes evidence for the next product decision
give every handoff a file:
→ source references
→ the actual output
→ checks run + open blockers
→ the next job and its exact context
the dot routes the work. specialists return artifacts. you review the decisions and feed corrections into the next task
start with one loop. make it work. connect the next one
save this, then build your company with Dots ⭣
32
26
125
9,441
OpenAI Dots is f*cking brilliant for building a 24/7 AI company...
here’s the bigger architecture I’d build around it:
one founder. one dot powered by GPT-6 Astra. 15 specialist jobs connected by work packets
the system has four loops:
BUILD
feedback + support repros → verified evidence → scoped spec → code branch → tested PR
LAUNCH
approved changes → explainers + demo clips + documentation → launch pack
REVENUE
account context → working POC → proposal → follow-up draft
objections and missing features go back into research
OPERATIONS
support triage + invoice drafts + status tracking → one queue of decisions for the founder
the connections are where this gets useful:
a support ticket can become a repro, a patch, updated docs and an answer draft
a finished feature becomes both launch material and proof for the next proposal
a sales objection becomes evidence for the next product decision
give every handoff a file:
→ source references
→ the actual output
→ checks run + open blockers
→ the next job and its exact context
the dot routes the work. specialists return artifacts. you review the decisions and feed corrections into the next task
start with one loop. make it work. connect the next one
save this, then build your company with Dots ⭣
32
26
125
9,441
paste this into your dot ↓
Design this operating system for my project. Define the 15 jobs across research, product, content, sales and operations. Give each job an input, output and completion condition. Create a shared work folder and a handoff template. Connect support to product, product to content and sales, and sales feedback to research. Identify missing app access and propose schedules for recurring work. Start with one complete build loop. Keep publishing, sending, deploying and spending behind my approval. Return the plan and starter files first.
4
146
beamnxw ./ retweeted
Sam Altman (CEO of OpenAI) just revealed how he uses Dots:
at DevDay, he recounted how he spent 10–20 minutes searching Slack and couldn't find what he needed
Dotty kept searching overnight
the next day: it found the missing screenshot
that's one part of his setup:
→ protects his mornings by filtering noise and surfacing urgent issues
→ catches urgent decisions in the gaps between meetings
→ builds 5–6 feature versions from raw notes
→ keeps a business dashboard updated around his changing priorities
the move: give your dot a responsibility it can keep owning
paste this into your dot ↓
Protect my morning focus. Surface urgent decisions with context. Keep a live dashboard of my priorities. Turn rough feature ideas into local prototypes for review. Ask before sending messages or changing shared systems
capture the idea → let work continue → review what comes back
study the entire system, then build your company with Dots ⭣
28
16
141
9,556
Sam Altman (CEO of OpenAI) just revealed how he uses Dots:
at DevDay, he recounted how he spent 10–20 minutes searching Slack and couldn't find what he needed
Dotty kept searching overnight
the next day: it found the missing screenshot
that's one part of his setup:
→ protects his mornings by filtering noise and surfacing urgent issues
→ catches urgent decisions in the gaps between meetings
→ builds 5–6 feature versions from raw notes
→ keeps a business dashboard updated around his changing priorities
the move: give your dot a responsibility it can keep owning
paste this into your dot ↓
Protect my morning focus. Surface urgent decisions with context. Keep a live dashboard of my priorities. Turn rough feature ideas into local prototypes for review. Ask before sending messages or changing shared systems
capture the idea → let work continue → review what comes back
study the entire system, then build your company with Dots ⭣
28
16
141
9,556
beamnxw ./ retweeted
OpenAI Dots is f*cking insane for building a 24/7 AI company...
here's a one-person company architecture to build around it:
one founder. five workstreams. one connected operating loop
01 research
customer signals + documents → a source-backed brief with the evidence behind each decision
02 product
the brief + recurring feedback → scoped changes, tested code and PRs ready for review
03 content
product updates + your voice → launch materials, clips and social drafts
04 sales
account history + customer requirements → proposals, proof-of-concept work and a test plan
05 operations
business records + open commitments → prepared invoices, follow-up drafts and work ready for approval
the useful part is how the branches connect:
research shapes the product. product changes feed content. sales requirements feed testing. your corrections improve the next run
the loop:
set the goal → connect the context → collect evidence → prepare the work → review and approve → feed corrections back
the rollout starts with one primary dot. the five branches in this map are workstreams around it
set app permissions and approval rules upfront. proactive background research uses read-only tools
save this, then build your company with Dots ⭣
46
116
780
95,411
beamnxw ./ retweeted
Sam Altman (CEO of OpenAI) just showed his personal ChatGPT Dots setup
and this might be the best setup from DevDay worth copying:
every night, one Dot reviews what came in
email - calendar - Slack - docs + DMs
before his day starts, it sends one short brief with:
→ urgent items
→ draft replies
→ key updates
→ decisions that need him
then it learns from every review and approval
the loop:
overnight inputs → triage → morning brief → Sam reviews → better next brief
OpenAI also showed bigger jobs for Dots:
→ monitor bugs and test fixes
→ find what is slowing down your app
→ migrate off legacy APIs
i mapped the full setup below ↓
34
38
337
36,574
beamnxw ./ retweeted
steal this Official OpenAI setup for Dots: put Codex to work while your laptop sleeps
your dot can start a coding task in a prepared Codex cloud environment, check results and send follow-up instructions
set it up once ↓
→ create a cloud environment with your repo, dependencies and access settings
→ give your dot the environment + a clear feature brief
→ open Activity to inspect progress, files and results
paste this into your dot:
Use my [ENVIRONMENT] Codex cloud setup to add saved views to the dashboard. Bring back the changed files, diff and test results. Keep changes unmerged. Ask me if a decision blocks progress.
replace [ENVIRONMENT] with your prepared environment’s name
cloud tasks can run while your laptop is off
local tasks need your computer online with the ChatGPT app open
one message → a cloud build you can review
save this, then build your company with Dots ⭣
24
16
89
5,786