Middle-aged dad, late to the game, but so glad to be here. Banner art purchased with sats from @isolabellart
3378 sats received
Filtering by: codex · clear filter
Another wild Friday night with a new AI model. 😂 Anthropic Opus 5 came out today. We'll see how good it is.
The big unlock for me the past 10-14 days is using Codex for everything, but having it open up Claude Code in VS Code to do work I want it to do. Better than me going back and forth.
Browser control is still blowing my mind, especially for boring inbox tasks and admin tasks.
#ai #claude #codex #opus5
Two things.
My teenager hiked a 14,000 ft peak - 4,300m for my metric friends. They drove up the interstate and were at the trailhead at 5:45 am. Very cool. Also my first 14er when I was their age.
Her dad has Codex and Claude Code with Fable 5 open now.
The hack: ask Codex to open Fable 5 in VS Code and direct it from there, then have it come back with what it finds. Maybe this is obvious to everyone else, but wasn't to me. Trying to use up all my tokens on Fable 5 before it goes to API only on Monday. It's done a wild design project. It's so far ahead of Codex in that way.
Follow Nate B. Jones on YouTube for daily AI videos.
#ai #codex #claudecode
7/19/2026, 5 engagements
Hey
team.
I use Claude Code and Codex daily. I'm comfortable in the terminal with both. But what I want to do now is connect the GLM 5.2 Private model to either the terminal or perhaps use it in VS Code.
Simple put, if I want private model experiences like the ones I have in Claude Code and Codex how can I use your API.
Thanks!
#ai #claudecode #codex
6/21/2026, 4 engagements
I probably have a very niche use case, but for my new business Perplexity Computer is a linchpin. It's simply is the best research tool. Claude Code and Codex aren't close. And for what it does it's not that expensive.
Tonight it's Codex as the main worker, Claude Code offering feedback to Codes, Perplexity Computer to research.
Very much looking forward to some Friday nights when I'm not working. Gotta make this new business go.
#ai #claude #codex #perplexity #claudecode
6/20/2026, 2 engagements
If you have even a small budget for #ai - personally or with your business - check out Perplexity Computer.
It's an orchestrator of roughly 20 different AI models. It does such a good job of running agents in parallel if you give it a problem.
For my new business, I have to do some really boring research by looking at public databases. There are other tools you can use, such as one called Firecrawl, that theoretically should be able to do it better and cheaper, but it simply can't.
Perplexity Computer will use the Perplexity Sonar models to do the searching, but then it will use the Anthropic's Opus 4.8 model to do the thinking and the orchestration. It seems to oscillate between those two models, and then it will always use the cheaper Anthropic Sonnet model for basic things.
If you had just $20 a month, it's best to have Claude or OpenAI.
But let's say you were going to spend an additional $10, $15, or $20 a month, you can buy Perplexity Computer creidts. Perplexity Computer is going to speed up your learning curve if you watch what it does in real time and sit back with a cup of coffee or a beer and just watch it work. It basically teaches you how these things use different tools in different ways.
Anyway, just a thought.
...and the frontier models are called frontier models for a reason...
#ai #perplexity #perplexitycomputer #claude #codex
6/5/2026, 1 engagement
I'm using an AI tool I made that's specific to my business, for a task I've had to do for years. It's so helpful and will take a two hour task down to an hour.
I think that in a year this is going to be the norm for a lot of people: custom tools that are the perfect fit for your business, but by you, with iterations along the way to make it better.
It took a lot of iterations to build it, and it costs some tokens, but it's great.
#claude #ai #codex
Some thoughts on AI as I work with it all day, every day, as an educator who isn't technial.
You have to both (a) play with these tools and (b) use them to make your current work either go faster or be done at a higher quality. Speed and quality for work, fiddling with them in between.
Claude is nice to you, so it's easy to think it's better than Chat GPT. It's a better writer, and a very good thinking, but Chat GPT is better at most things.
Because Claude is nice to you prompting it with "Don't tell me what you think I want to hear - be honest with me" is crucial. ChatGPT doesn't need as much of this. It's always honest, and give sober outlooks on things.
People think "AI can't write" and that was true in August 2025. In May 2026 that's silly. I write a weekly newsletter to 20k people. I dictate it via Wisprflow, Claud writes the first draft and then I tweak it. I have fewer unsubscribes than I did before I started doing this. It still takes an hour, but it used to take two.
It's best to have both the $20 Claude and $20 Chat GPT subscription if you can swing it.
Claude is a great copywriter. ChatGPT keeps getting better.
https://wisprflow.ai/ is a must.
Kimi models with
are good when you want privacy, but I can't run a business with those models.
I think and Claude are the two must haves if you do knowledge work.
Knowing when to use High and Extra High and Medium with Codex is important.
Claude Code starts to do dumb sh*t past about 65% of the context window. When I ask it at that point if we should write a handoff and start a new conversation it almost always says yes.
I slept on Claude Cowork and that was a mistake. It saved me a ton of time a few weeks ago when I needed quotes from a dozen book publishers. It wrote the follow up emails too, and then compared the quotes. There was zero reason for me to do that work. I found a better company than my current printers.
Browser control in Claude and Cowork is voodoo.
"Early in their career" lawyers and accountants are in trouble. In a month I'll replace my bookkeeper. This tax season I created a simple tool that took all my tax info and was within 1k of what my accountant said I'd owe.
Again, Claude writes very well. You can prompt it to never use em dashes.
I should have more skill.md docs, but the key ones I have a so powerful.
90% of the time I either start a Claude Code session with /plan or /superpowers.
I spent a ton of time with Openclaw in Feb and it was a waste, other than it got me comfortable using the terminal, and then I dove into Claude Code.
I worry that token cost for plebs will quickly be really high in the next year or two.
What are you learning? #asknostr
#ai #claude #claudecode #chatgpt #codex #plebai
Wild Friday night: diet coke and AI.
Perplexity Computer is looking up places to move where we can have a bit of land.
Claude is brainstorming a dashboard/app that I can build with Codex or Claude Code that has all business inputs and will allow me to connect to bookkeeping software so I can stop paying the accounting firm for that. Such a waste. Though the accountant himself is good. I may net out that Cowork or Codex can do these without building something. Cowork is really powerful.
Codex is building an educational site for Gen X'ers who either have bitcoin or are "bitcoin curious" and wonder if they have enough. Follows up on this idea, but much better.
Back to real work tomorrow using Codex and Claude Code, but it's fun to fiddle and play.
#ai #claudecode #codex #plebai
5/23/2026, 4 engagements
#codex bowser control is wild. It's fast and it just saved me several hours of work, but I'm not sure I like it.
#ai #plebai
GM 😀 ☕
About to head out the door to take my mom to breakfast. She's 82 and runs 10k road races. She's such a wonderful role model for my kids.
I think what I'm going to do with this as a little project this summer is:
Similar to Wen Moon, create a calculator that give a guess as to when you can retire on bitcoin, but use the Catch-Up Model. https://observablehq.com/@jimbojw/wen-moon
Add 3-4 scenarios for inflation - similar to what's in this article - that walks us through the the importance of stacking now. https://primal.net/a/naddr1qvzqqqr4gupzqs2fpquh4unxgqj8qsk6072zljf3sd7gm8fjql38rje0k2aanf8dqq24wd34g36nq5zz899nyc640p6njmjnwskhyvukqwg
The Catch-Up Model has use getting back to Power Law trendline in June 2028. Just a guess, but it's reasonable.
Anyway, it's a fun project to hone my skills and I use Claude Code and Codex for my business.
Have a great Sunday!
#bitcoin #ai #codex #claudecode
At a minimum, this tool has helped me. I stacked a bit this weekend. I'll still touch grass this summer, but maybe a touch less than in the past.
Have a great Sunday. Headed to see my
Loading quoted post…
5/10/2026, 3 engagements
I built something I think is useful with Claude Code. It's called Stack Buddy. I don't think it's AI slop, at least I hope not. Link below if you want to check it out.
Here's the back story on it...
I'm starting a new business this year and working 6-7 days a week in what feels like two jobs. I feel like 2026 through 2028 will be an interesting window for two things: using AI to build leverage in my small business, and stacking sats while we're still well below the Power Law trendline.
Stack Buddy is the calculator I wanted for myself to see if I worked my face off, how much I could stack in this period. So I built in between sessions with Claude Code and Codex
The first thing it does is simple: You put in your income, your tax rate, your monthly expenses, and what you're trying to save in fiat, and it tells you how much BTC you can realistically buy each month. Rough numbers are fine.
The second thing is more interesting.
You put in a target stack and a deadline, and it shows you three different paths to get there.
For the third scenario, you might do 2x your normal DCA on January 1 and July 1 each year, plus a tax-refund buy each April.
The math anchor is what I'm calling the "Catch-Up Power Law." The price today is well below the Power Law trendline, and the question is when we get back to it.
For the price model I ran that question through several AIs looking for the most defensible price guess, and June 30, 2028 came back as the answer with the strongest justification. Obviously I'd hold that date loosely but the calculator needs a concrete anchor and that's the one I'm using.
The model lets you see what your plan looks like under that assumption and stress-test how sensitive your stack is to the timing.
A few things worth knowing before you click through...
It runs entirely in your browser, doesn't call any LLM, doesn't store your numbers anywhere, and there's no account system.
The whole thing is open source on GitHub, and there's a download button at the bottom of the app if you want to grab the code, fork it, or run it yourself in Claude Code or Codex.
And maybe the coolest part, is there's also a prompt you can download and paste into a privacy-respecting LLM like
, Venice.ai, or you can wire up
API if you want to talk through your plan with an AI rather than just clicking buttons.
I'd love to know what you think and if you have suggestions on improvements I'd love to hear them.
https://orangedaddocs.github.io/stack-buddy/
#bitcoin #stack #stacking #ai #claudecode #codex
If you're using Claude Code, and like me, you're not a developer, I humbly submit these two plugins as musts before you start building things.
I just typed /codex:adversarial-review and couldn't help but think it's a bit like "Don't Trust, Verify."
https://github.com/openai/codex-plugin-cc
Superpowers has been a huge help in planning things. My limited understanding is it's better than /plan mode that Claude Code can do.
https://github.com/obra/superpowers
Any other must have plugins?
#claudecode #codex #ai
For roughly a month I've been using Codex as an "auditor" for what I'm building in Claude Code. It's worked well, but it's a lot of back and forth.
I found the following plugin tonight from a YouTuber and it's perfect.
https://github.com/openai/codex-plugin-cc
Looking forward to using it this week to speed things up.
#ai #claude #codex