Tinfoil Shiestyt retweeted
$ZEC
"update"
The micro roadmap has been updated. wave (X) is in, looking for a move into the red box to complete wave (Y) of red wave (B) before the primary breakdown takes over.
Macro plan remains unchanged.
nitter.cf/sarper_onder/status/21…
$ZEC
"update"
Plan unchanged. Expecting a rise to the red box before the sharp correction down to the blue box.
Opened a scalp long here to ride the move up.
nitter.cf/sarper_onder/status/21…
Highly recommend building 𝗮 𝘀𝗺𝗮𝗹𝗹 𝗳𝗿𝗲𝗲 𝘁𝗼𝗼𝗹 for your product to bring in users from Google search.
Say you make a fitness app. Build a free protein calculator. Someone searches "how much protein do vegetarians need a day" and lands on it. Once they get their number, 𝘆𝗼𝘂𝗿 𝗮𝗽𝗽'𝘀 𝗱𝗼𝘄𝗻𝗹𝗼𝗮𝗱 𝗯𝘂𝘁𝘁𝗼𝗻 𝗶𝘀 𝗿𝗶𝗴𝗵𝘁 𝗯𝗲𝗹𝗼𝘄.
Send this prompt to Opus 5.5 in Claude Code 👇
"Read this project and come up with 3 free tools for me to pick from. Each one should relate to my product and be something people search for. Give me the search term for each. Skip broad terms and pick specific ones with little competition. Search each term online and check whether big sites hold the top results. If you can't find the search volume, say it's a guess. Don't make up numbers.
Once I pick one, build it as its own page on my website. If my website isn't in this project, ask me first. It needs no signup, works the moment the page opens, and works well on phones. Under the result, put just one button that takes people to my product. Write the SEO title, meta description and H1, answer a few common questions on the page, and link to it from the homepage or footer. Add it to the sitemap, make sure Google can index it, and make sure the title and main text show up without JS.
When it's done, run it locally so I can see it. List the URL, title and meta description, and tell me what else could be improved. Change whatever I point out, round after round, until I'm happy. Then walk me through launching it step by step."
Tinfoil Shiestyt retweeted
You’re not lazy. You’re dopamine depleted.
Scientists proved it by DEPLETING a healthy man’s dopamine in a lab.
They blocked ONE enzyme. Within hours he couldn’t function.
Sound familiar?
• staring at tasks unable to start
• rereading the same paragraph 3 times
• needing coffee just to feel normal
• no motivation by evening
• scrolling your phone because nothing else feels worth doing
That’s your Tuesday.
The enzyme is tyrosine hydroxylase. It controls how much dopamine your brain produces.
The Soviets built a compound that activates it by 65%. I take it every morning: 🧵
Tinfoil Shiestyt retweeted
I don't know how, but Bitcoin will reach 0.786 fib level in a month.
As it always does in bear markets.
Tinfoil Shiestyt retweeted
the best GPT-6 Astra coding advice I've found in one prompt:
"You are the accountable engineering owner for this repository:
1) Simplify how humans and coding agents work here so goals become complete, verified, mergeable changes with minimal coordination.
2) Start by inspecting the actual product, architecture, agent instructions, skills, development setup, tests, and CI.
3) Identify the specific things that cause unfinished work, repeated confusion, or unnecessary effort.
4) Make a small, reviewable set of the highest-value improvements.
5) Prefer consolidating and clarifying existing mechanisms.
6) Add new process only when it solves a demonstrated problem.
7) Keep the main agent instructions short and accurate: what the product does, where important code lives, the critical boundaries, and how to verify work.
8) Link to deeper guidance.
9) Give skills clear purposes and load them when relevant.
10) Avoid duplicated sources of truth.
11) Define completion through observable user outcomes.
12) For the work being changed, trace the necessary path through the interface, backend, persistence, workers, and external services.
13) State which parts were exercised and which remain unverified.
14) Keep one owner responsible for integration and completion.
15) Delegate only bounded, independent work when it will reduce total effort.
16) Avoid recursive delegation and repeated review exchanges that add no new evidence.
17) Match planning, testing, and review to the change’s risk.
18) Distinguish blocking defects from optional improvements.
19) When an approach repeatedly fails, investigate the root cause and change direction.
20) Make essential verification easy to run locally and in CI.
21) Preserve useful regression coverage and report failures honestly.
22) Remove tests or checks only with evidence that their protection is obsolete, redundant, or ineffective.
23) Inspect existing code and suitable services before building custom infrastructure or integrations.
24) Keep cleanup within a justified scope.
25) Measure performance before and after making performance claims.
26) Respect existing permissions, security boundaries, user work, and release approvals.
27) Do not merge or deploy without authorization.
28) Implement the improvements and run the relevant checks.
29) Finish with a concise account of what changed, why it helps, what passed or failed, and any remaining blocker.
30) Stop when the agreed scope is complete and sufficiently verified.
31) Start with one real feature and see how much intervention it still needs. Use that experience to decide what to improve next."
Tinfoil Shiestyt retweeted
how to become a marketing engineer
the role is changing, people call it marketing engineer, gtm engineer, growth engineer, same idea.
instead of relying on tools that are prompted one chat message at a time, we are moving to building custom agents and orchestration
I made this shift over the last year, here's the path:
1. build, don't just use
right now you open a chat and ask for one output. the move is building the thing that produces that output on its own, again and again
2. build your company brain first
this is the moat, a centralized place for everything you and your company know:
> transcripts and call notes
> previous campaigns and what they returned
> client learnings and strategy docs
> brand rules, positioning, ICP, offers
> internal workflows and examples of what good looks like
every agent reads this before it works. aim an agent at good inputs and weeks of work turn into hours. point it at a mess and it just makes slop, faster
3. learn agent profiles
a profile is a dedicated agent with its own config, skills, memory and model. build open ones that need you in the loop for research and one-offs, and closed ones that own an outcome end to end and run on their own. promote the open ones into closed loops over time
4. find your agents by watching your own work
dont jump to building a new agent. prototype in your main agent first, do the work yourself, run the same shape of task a few times, notice what you keep repeating and what context it needs. that pattern becomes your next specialist
5. design the loop before you run it
a poorly designed loop burns tokens and hands you slop fast. design the steps, the checks, the budget and the done state, then run it. a loose loop wanders, a designed loop runs the same every time
6. orchestrate agents, not tools
one agent per vertical, SEO, PR, paid, CRO, each with its own skills and scoped sub-agents. an orchestrator routes the work and a human champion keeps the brain clean. now its a team running your marketing, with you directing it
7. close the eval loop
slop is a systems problem, judge the output against what good looks like, feed the misses back into the brain, and the system gets better
you really don't need a CS degree for any of this, you need a builders mindset
Tinfoil Shiestyt retweeted
I also think we hit peak “terminally online” too, sometime in 2024-25. The culture of onlineness itself is starting to feel terminally Millennial-Facebookian
Google engineer explained how to fine-tune a tiny LLM from 46% to 90% accuracy on your phone in 21 minutes - better than $1500 on-device AI bootcamps.
pick Gemma 270M -> generate synthetic task data -> fine-tune with LoRA -> quantize to int4 -> deploy to Pixel and hit 2000 tokens per second.
That loop is how a 270M model beats a 70B one on your task, running fully offline in your pocket.
Gemma 270M + synthetic data + LoRA + int4 quantization + on-device runtime - that's the stack.
Watch and save it, then fine-tune your own tiny agent tonight.
4,400 bookmarks....
I'm surprised that people didn't know about this.
If you dont know, you can use your Codex subscription inside Claude Code.
That means Fable 5 can call GPT 5.5 + 10 subagents to work together.
This will save you atleast 60% of your Fable 5 token consumption.
You just need to setup once. Here's how:
Step 1: Open Claude Code and install the Codex plugin
/plugin marketplace add openai/codex-plugin-cc
/plugin install codex@openai-codex
/reload-plugins
Step 2: Tell Fable 5 to finish the setup
Paste this prompt:
"Set up Codex inside this Claude Code environment. Use the official OpenAI Codex plugin that was just installed. Run /codex:setup. If Codex CLI is missing, install it. If Codex is installed but not authenticated, ask me to authenticate with my ChatGPT account. After auth is complete, verify that Codex works from inside Claude Code. Then confirm that the codex:codex-rescue sub-agent is available. Do not change any project code during setup."
Step 3: Authentication
Fable 5 will trigger the Codex setup.
You authenticate your ChatGPT/Codex account once.
After that, Codex runs from inside Claude Code using your Codex subscription.
Step 4: Tell Fable 5 how to delegate work
Paste this prompt:
"From now on, use this workflow:
You are the orchestrator.
> Use Fable 5 for planning, repo understanding, architecture decisions, task decomposition, and final review.
> Use codex-rescue as the executor when a task needs heavy implementation, debugging, test fixing, refactoring, or multi-file code edits.
When delegating to Codex, use /codex:rescue.
Prefer GPT 5.5 (xtra high) as the go to Codex model.
Keep Codex tasks focused and specific.
After Codex finishes, inspect the result yourself before accepting it.
Do not blindly trust Codex output."
Pro Tips:
1. Turn this into a skill (i named it Fable-GPT) and call that skill at the start of the session.
2. Use skill + goal to get the heavy tasks done. Goals are best for long horizon tasks.
3. If you're on Codex 20x pro plan, you can use subagents. I use 5-7 agents at one time and never hit 5-hour limit.
4. Context rot is real so clear the conversation after 4 compactions. Use /handoff skill to preserve context.
Tinfoil Shiestyt retweeted
Anthropic’s official Claude Certified Architect Foundations badge is free until August 31, and it covers the real stuff: Claude Code, API, MCP, RAG, agents, tool use, security, and business automation.
Tinfoil Shiestyt retweeted
If you’re interested in the ideas of Ray Peat—the real ideas, not the stupid meme ideas—follow this account, they post a lot of helpful material like these charts
a good system for locking in and compounding in your career, with Claude as a growth coach:
> init Claude in a fresh folder, dump your baseline into career.md — where you're at, where you want to go, goals, old docs, journal entries, job descriptions, etc
> have it interview you one question at a time to find your gaps, appending to career.md each turn
> after a few turns have it draft your horizonX.md — a 6mo plan with a single time-bound mission, and 3-5 supporting themes that ladder up to it
> then break each theme into units of work + target outcomes
> execute daily: feed it journals, proof of work, PRs, meeting transcripts, whatever you have to show it objective progress
> have it continuously challenge and refocus you against horizonX.md
> weekly: a 15min review in wXX.md — what moved, what stalled, what's next. this is your clock speed, run it tighter to compound faster
> close outcomes, set new ones, repeat every 6mo







































