Sabitlenmiş Tweet
Charlie Deist
4.4K posts
Charlie Deist
@chdeist
Retired sailboat captain building a mini media empire from a northern California homestead. 20 acres and a cow. Father of 4. Finish your work and go outside.
Butte County, California Katılım Ekim 2021
736 Takip Edilen729 Takipçiler
English
I need this but with a @daylightco monitor.
ZARA@HeyZaraKhan
This dock turns a Mac mini into a classic Macintosh. wow
English
Charlie Deist retweetledi
Charlie Deist retweetledi
Vibe Coding is a Video Game with Real-World Rewards.
English
@trillhause_ This is exactly right.
It’s hard enough to design and impose a second brain on another person - let alone a company.
It has to emerge organically.
English
This is going to be a tarpit idea. It’s good in theory, but impossible to pull off unless it’s an internal company effort by a tyrant like CEO.
An external company will never be able to build a software that results in a company brain. It’s mostly because no tool will have perfect adoption from all employees and data will always be fragmented across new systems.
Chaotic systems are very hard to capture. It’s impossible to perfectly extract data from all sources as companies evolve and introduces new data sources.
You will spend all the time keeping track of the data instead of doing actual work. This is same trap that the second brain productivity folks fall for.
Y Combinator@ycombinator
Company Brain @t_blom Every company has critical know-how scattered across people's heads, old Slack threads, support tickets, and databases, and AI agents can't operate like that. We think every company in the world is going to need a new primitive: a living map of how the company works that turns its own artifacts into an executable skills file for AI.
English
@nbaschez Hey👋 my thing is - I'm doing 99% of the work in a monorepo that mixes code, markdown, and other media formats.
I default to @Zeddotdev raw speed, but I like the pretty markdown layer, fileviewer, etc. of @nimbalyst
Just food for thought.
English
Do you spend a lot of time reviewing markdown docs written by AI?
Wish it were a better experience?
Say hi if you wanna try a new (free, open source) thing
English
Voice vs. typing conceals a darker impending question/challenge for society:
Will we accept a brain-to-computer interface like Neuralink if it gives us an economic edge (and it almost certainly will).
For me, that is a bridge too far. And I'm even reconsidering voice.
LindyMan@PaulSkallas
Remember the Lindy rule. Humans only talk to things that are alive 1) Other humans 2) Themselves 3) animals 4) Things embodied with spiritual significance (gravestones, holy places, prayer) We could all just sit and talk to our computers now. We have the technology. But we don't. It feels weird. We sit and type in silence. Many people have tried so far to create a talking device. all have failed
English
Brand’s deeper problem - and he knows this already - is that he still seeks validation and approval from the chattering classes when he should probably retreat to private life or just focus on his own channels.
English
People dunking on Brand for this
But if you’ve ever had the experience of looking for that one verse across 65 chapters of Isaiah, this will just make you like him more.
Mukhtar@I_amMukhtar
Piers Morgan asked Russell Brand which passages were relevant to him when he brought a Bible into court.
English
Elon Musk says the all-AI corporation will outcompete the corporation with humans in the loop.
I think that's wrong.
My latest on:
– Jagged Intelligence
– Human/AI complementary
– Why UBI is a trap
And cultivating the ability to decide the exception.
x.com/chdeist/status…
Charlie Deist@chdeist
English
@alexhillman How all those bespoke GUIs tie together. How and when you spin them up or if they’re always on in some window in the background.
The full stack.
English
I think I wanna do a livestream tomorrow.
Happy to show off any parts of my Claude Code rig, specific examples of how I use it, maybe even add something to it.
What would you wanna see?
English
This is an example of a prompt with Claude Code for writing an article that is super helpful.
One of the most delightful things about using AI for me right now is when I have a general idea of the direction I want to take the article, but I don't quite know the history very well.
In this case, I was using general contractor as an analogy but it didn't quite land and so I gave it a brain dump of how it didn't quite fit and it came back with a really good summary of the industry and how it works for my analogy.
Just the removal of tedium (in this case I would have spent an hour googling around) like this is a source of cautious optimism for me.
On the whole, I spend way more of my working hours on interesting problems now vs. six months ago.
Taylor Pearson@TaylorPearsonMe
My current favorite metaphor I've found for working with AI is the general contractor. Architect doesn't quite fit because it's too hands off. (This is maybe not the case for software where the models seem better) A GC is on site. They know the dependency graph: the electrician can't go before the framing, the drywall can't go before the electrical inspection. They know which sub (agent/skill) to call for which job, and whether the work is good when it's done (QA/QC). That's what working with AI looks like for me right now. You just seem to get bad results from "Claude, rebuild my CRM, make no mistakes." That's like hiring a framing crew and telling them to build you a house. The failure modes for a GC are instructive. A bad GC fails in one of two directions. 1. Too hands-off: delegate everything, don't inspect, the plumbing fails behind the drywall. 2. Too hands-on: micromanage every nail and the project takes three times as long. Both failure modes show up constantly in how people use AI. Some people fire off a massive prompt and accept whatever comes back. Others approve every line and never let the system do what it's good at. I think it's also useful because what makes a good GC defensible is local knowledge. They know the subs, the building codes, the soil conditions, the inspector's quirks. And it doesn't transfer easily. Twenty years on commercial projects in Southern California doesn't translate to residential in Minnesota that well from what I can tell: Different climates, codes, subcontractors, client expectations. Broad skills carry over, but the deep local knowledge doesn't. Same with AI. Everyone gets the same models. The differentiator is the scaffolding you build around them: your files, your style guides, your workflow rules, your accumulated context. Two people with the same model can produce wildly different results depending on the 'local knowledge' they've encoded. And like a GC who (should) get better with each project, the scaffolding should compound. Each skill you build and piece of context you add should make every future session more capable.
English
@alexhillman everything started going haywire for me when I tried to switch to GWS. so much promise, but it forced me to re-authenticate every week it seemed...
English
back in October I spent a couple of weeks on a complex training approach to get Claude Code to triage my inbox.
by the end, I threw it away for something that's performed WAY better but I don't see people doing, so I thought I'd share: 🧵
English
I think the top tip I've picked up since I started using Claude Code: write a CLAUDE [dot] md at every level of your file structure and be intelligent about how you nest folders.
Claude Code automatically loads the Claudemd file from every directory level above whatever file you're working on. So if you organize it intelligently, it always has the right context at the right time.
Say you're running a business and doing some marketing work. You might have:
- A general CLAUDEmd on your computer — how you like Claude to work, voice, tools you have wired up
- A vault CLAUDEmd — how your files are organized
- A folder for your business with a CLAUDEmd — what the business does, who the clients are, how you price, etc.
- A folder for marketing inside the business with a CLAUDEmd — channels you run, what's working, brand voice, etc.
- The specific marketing project with its own CLAUDEmd
When you open anything in the marketing project folder, Claude Code walks up the tree and pulls all five into context automatically. Global → vault → business → marketing → project. That means the very first message in a chat about that marketing project is going to be like talking to someone that knows you, your business, your marketing history, etc.
You need to keep the CLAUDEmd at each level current for this to work
I have a "wrap" skill that I run at the end of every chat session. It updates the relevant CLAUDEmd(s) and makes a more detailed entry into a workbench file for whatever I was working on.
So if I make a decision about a project or learn something about a client, that context lives in the right file by the time I close the session. Next time I come back, Claude already has it.
Starter repo for this kind of setup: github.com/heyitsnoah/cla…
I built a little mini-extension for the workbench structure I use: github.com/ParrotDiceFoun…
English
The most optimistic scenario for the agentic coding era is that we can finally let computers do the computer stuff so that humans can do the human stuff.
English
@hasantoxr I remember using Oh My Open Code back in the day and it was really very impressive. That was back when you could still use your OAuth for Claude inside OpenCode.
I wonder if anyone remembers those days and whether this will get me back closer to that ideal.
English
🚨 Claude Code just got superpowers.
Someone built a multi-agent orchestration layer on top of Claude Code that gives it 5 execution modes, 32 specialized agents, and 3-5x faster output. Zero learning curve.
No new tools. No new subscriptions. Just Claude Code running like it was always meant to.
It's called oh-my-claudecode.
Here's what's inside:
→ Autopilot mode: fully autonomous execution, just describe the task and walk away
→ Ultrapilot mode: spins up parallel agents and runs 3-5x faster on multi-component builds
→ Swarm mode: coordinates multiple agents working independently toward the same goal
→ Pipeline mode: sequential chains for multi-stage processing tasks
→ Ecomode: token-efficient execution that saves 30-50% on costs without sacrificing quality
Here's what actually makes this different:
32 specialized agents for architecture, research, design, testing, and data science. Smart model routing uses Haiku for simple tasks and Opus for complex reasoning automatically. You never think about which model to use. The system decides.
One magic keyword and everything changes:
→ Type "autopilot" and it builds the whole thing autonomously
→ Type "ralph" and it goes into persistence mode, won't stop until the job is verified complete
→ Type "eco" and it switches to budget mode
→ Type "plan" and it runs a planning interview before touching a single file
The wildest part? It auto-resumes your Claude Code sessions when rate limits reset. No babysitting. No manually restarting. Just continuous execution.
3.6K GitHub stars. 100% Open Source.
Link in comments.
English