Josh Howard

I studied 10,000+ YouTube Community posts from channels with 200 to 17M subscribers. What makes a post reach a bigger audience? What’s the most engaging type of content? Do community posts even matter? What’s the point? Here’s my findings:

Google Opal + Skills in Google Docs to put in the chain. You're welcome.

@Daniel_Batal @SullyErna @rickbeato Your entire vibe with the lighting, gear, and setups I've seen, including this one, are really something I aim to have of my own in the future. You've got a great eye for these things. Very very cool stuff.

@RealJoshHoward @SullyErna @rickbeato Thanks, Josh! I've been meaning to add a more serious Mac station to my set for a long time and I finally figured out the perfect little nook to build it into.

My Mac Studio station is finally done! I’ll be putting @SullyErna at this station for tomorrow’s live stream with @rickbeato If something doesn’t work, I’m blaming Sully 😜 Be sure to set your reminder to join us LIVE! 👉 youtube.com/live/8OJX1t3-O…

@tylerfleisch I use voice dictation a LOT now for this, but I don't know if I'd feel comfortable doing it in public lol That said, now I feel inspired. Might as well put on a VR headset too.

Working in public, and I'm overhearing a dude prompting AI with his voice. He's whispering prompts under his breath. Practically casting spells.

Built an AI grocery buyer in @Claude Cowork. It knows your preferences, tracks order history, knows backup items, & self learns site changes on its own. No tech skills needed. Just log in & talk to it. Go to kitchen > voice dictation > done. What a time to be alive. @bcherny

@Lunayian A radial would help with accidental "clicks" when it loses tracking and you're moving through other buttons.

I've said this already, but they really need to make it into a radial or even a layered 'pill' along the lines of the vOS hand menu. There is genuinely too much going on here lol

@Liron_Segev @DougHewsonYT It was a decent chunk of time last night in primetime. For instance, I clicked on a Short on desktop, and you couldn't scroll or view next. Lots of things like that were broken and made YouTube unusable in a lot of ways for recommendations.

@RealJoshHoward @DougHewsonYT yeah - that's the bit I am concerned about. I am seeing so many channels with views down so much (including my own)

Apparently, the algorithm broke 😂

@Liron_Segev @DougHewsonYT Yah, I figured you trusted him! Mine was "in case you wanted to use his screenshot, I absolutely saw the exact same tweet" for confidence. 👊 Weird that they deleted it though.

@RealJoshHoward @DougHewsonYT Thanks, Josh! I just can't find it in their timeline. I see stuff about the outages, but not about the recommendation engine. That is what I am more interested in. To be clear: I wasn't obviously questioning Doug (he is a legend!). Only my searching ability lol🤦‍♂️

@Liron_Segev @DougHewsonYT Saw this exact same tweet last night during the outage in case you're looking for additional verification.

@DougHewsonYT hey - where did you see this tweet? I can't find it in their timeline (and thanks for flagging it)

@Liron_Segev @openclaw @AlexFinn Might need to go down this route. I think I have a few options where I can be operational.

@RealJoshHoward @openclaw @AlexFinn I know that feeling - bits and pieces everywhere. If you need something light, check this out: picoclaw.net

5am Setting up another computer with @openclaw which has evolved so much from the first version in such a short period. Kimi k2 and local llm is the winning combo I agree with @AlexFinn, the future is everything local. More control. More privacy. My advice is don't wait for a Mac mini. Get an old device, install @ZorinOS and install it today. Understand it. Make it do one thing. Build from there. (I need more RAM + more coffee)

@Liron_Segev @openclaw @AlexFinn All my old devices apparently don't want to work anymore without some TLC + parts. Trying to find a reasonable way to approach this.😅 Appreciate the feedback. If there's anyone I'll trust on this, it's you.

@RealJoshHoward @openclaw @AlexFinn Personally, I like a dedicated machine as it has access to everything. I know people who run it in a VM but its not ideal and loads to tinker with. I would prefer an older slower machine that is dedicated and this way you can really get stuck in.

@Liron_Segev @openclaw @AlexFinn Thoughts on safety of using something like Podman and running it in a container?

@dylan_bridger More isn’t always better. It’s more about what you’re having it analyze and what you’re trying to extract. Different mediums normally = different number of variables you want to focus on extracting and what. Are you trying to get tone and language or format or both?

Question for Claude power users: When using a Project to emulate a writing voice, are you better off uploading as many writing samples as possible or a curated set of ~12 examples? (Maybe this variable is low-impact or there's a better approach entirely.) What's the move??

@TheCraigHewitt It's unavoidable. I feel like AI-enabled people now will have a huge advantage in a lot of areas in the coming years.

My main priority every day is to get better at AI. To level my team up with their AI skills, to help/inspire other founders to do the same too. If I’m successful with this I’ll be successful in business and in life.

@jayclouse Opus has been incredible for inferring intent and logic. Been a big standout for me. Not surprised to hear this take for this use.

Spent hours trying to get ChatGPT to write its own Instructions propt for a consistent illustration style... Opus 4.6 succeeded in ~3 cycles. If it weren't for DALL-E, I wouldn't touch ChatGPT.

@bcherny Mine was visible, but transparent and then not usable. Then it just disappeared as a tab altogether. I'm sure the team is on it, but passing it along as someone who would love to stop remote desktoping into their laptop :)

Windows users, rejoice -- you can now try Cowork

@TheCraigHewitt I've used it more for playing around with things in ComfyUI for generation, but also experimented with some ML and random tools. Tested a bit when OSS came out in the summer, but just haven't found the thing to spend a lot of time on local models yet (until recently).

@RealJoshHoward no, from inital research it looks like 5090 gives a good sweet spot to that ability to run good models without breaking the bank. most of what i'd need for an agent could run there. mostly though, peace of midn

hyperfocused on getting my own hardware to run models locally. Looking like RTX 5090 32GB is the best bet for a mid-tier GPU. Anyone done this?

@RoxCodes If ever you do any type of roundtable discussion about use cases and people sharing insights, I'd be down to sit in. Not a dev background, but the last 12 months puts things into perspective on what's possible and the gap between prompting and coding continues to get more narrow.

damn this blew up if you want to work with me and figure out all this AI shit, please apply at startupdevs.com currently hiring for - senior founding engineer - junior/mid-level full-stack web dev - support engineer nextjs/ts/trpc/cloudflare workers/durable objects

my #1 problem with ai coding is I never trust it to actually test stuff but today I got codex to build something, then asked it to record a video testing the UI to prove it worked it built a whole playwright script, recorded the video, and attached it to the PR the game changes every month now crazy times