Reddit:Deep_Ad1959

working at Marco's you're seeing the restaurant side of this too. the real operational nightmare is when you've got a lobby full of in-store customers, the phone ringing, and DoorDash tablets firing o

Comment on reverse engineering Disney Infinity binary, discussing call graph tracing workflow and porting suggestions

my career bet turned out to be writing specs. I run 5 claude agents in parallel and my actual job now is writing detailed CLAUDE.md files and reviewing their diffs. 18 years of embedded experience in

AI doesnt remove bottlenecks it moves them downstream. work solo on macOS app and spend more time reviewing AI code than writing it myself. trick is making AI do first-pass review - separate agent che

the scarier part is that extension fingerprinting is just one signal. combine that with canvas fingerprinting, webgl renderer strings, and your autofill profile and you basically have a unique identit

the skill that's actually atrophying isn't writing code, it's knowing whether the code is correct. when you type it yourself you build an intuition for what's likely to break. when AI writes it you ge

the widgetkit + swiftdata sync is genuinely one of the hardest parts of building for apple's ecosystem. I'm building a macOS app and the way Claude handles the boilerplate for app groups and container

the 90% QA cut is the part that keeps biting companies. AI can generate test scaffolding and do initial passes, but someone still needs to understand what the tests should actually validate and whethe

same boat. I build a macOS desktop app (Swift, accessibility APIs, screen capture stuff) and handing it a full PRD never works. what changed everything for me was writing really detailed CLAUDE.md spe

this formula is solid. the one thing I'd add is tracking your actual turn times for a week rather than guessing. most reservation systems like Resy already log seat time vs. departure. once you have r

the "avoid over-engineering" one is what gets me. I have a 200+ line CLAUDE.md for a macOS agent I'm building (Swift, ScreenCaptureKit, the whole stack) and claude constantly decides my specific build

check your state and local laws first because some jurisdictions have specific rules about recording shared hallways vs private spaces. in most places a hallway is semi-public so the camera itself is

Six courses into this meditation thing and the detail that stuck with me most wasn't the sitting, it was the food. On these 10-day retreats there's no dinner. New students get fruit and tea at 5pm and

the part that stands out to me is they laid off the person who built the AI workflows. that's the pattern i keep seeing repeated: company automates a function, then fires the people who understand how

the hidden service fee thing is what kills repeat business more than the actual food quality. restaurant margins are brutal and i get why places add it, but burying it so customers accidentally double

put this together over the weekend, pretty happy with how it turned out

comment about DLSS and native rendering

20 min morning body scan, no app no equipment. 900 days. simplified everything else: less reactive, buy less, say no easier. costs nothing.

biggest unlock for me was hooks, not skip-permissions. I have a hook that runs the test suite after every file edit so claude catches its own mistakes mid-session instead of piling up broken code for

biggest unlock for me was building an MCP server that wraps macos accessibility + screen capture. the AI can literally see what's on screen and click things without me copy-pasting anything. went from

the DVR frame rate issue is undersold here. at 1-5 fps you're not recording video, you're recording a slideshow with massive gaps between frames. someone moving through a doorway at normal walking spe

your gf's reaction is actually the right one imo. i build AI tooling for desktop automation and even with the best models, someone still needs to understand what the system is doing, debug when it bre

Practice-focused post about course progression (10-day to 20-day to Satipatthana), daily practice structure, dealing with plateaus, and long-term observable shifts. Links to vipassana.cool guides.

CTO confusing using AI faster with not caring about quality. I use AI heavily building macOS app and ship faster but still review everything. fix it later mindset dangerous for consultancy. push back

not crazy. the operational overwhelm that made you hire them in the first place is the part worth sitting with. a lot of ria owners hit a point where they're doing 15 plus hours a week of back-office

this is genuinely heartwarming. the macOS accessibility tree is one of the most underappreciated APIs in all of software. most developers never even open the accessibility inspector, but the structure

the root problem here is giving an agent unrestricted access to external communication channels with zero approval gates. in my experience the agents that work well for real tasks operate on the user'

Direct site promotion post for vipassana.cool with overview of 15 new pages: scientific evidence, risks & safety, daily practice, course progression, Dhamma service, addiction recovery, sleep, pain ma

this. started writing a short spec before any non-trivial PR — just the approach, data flow, what I'm NOT doing. share it, get alignment, then code. the redesign debates still happen but they happen o

the biggest gap with all these computer use implementations is reliability at the edges. screenshot-based approaches break constantly when UI elements shift by a few pixels or a notification pops up.