Replying to @AndrewYNg
Software Engineering (#2) and Product Engineering (#4) have always been skills that set the best engineers apart and will continue to.
In the AI era, the 2 new skills to focus are:
1/ AI Systems Engineering (#1) - this is the “real engineering” that is closest to the traditional software & systems engineering
2/ Agentic Coding (#3) - this is the closest to “programming” where it’s the base skill that you then still need to apply towards the rest #1, #2, #4
Replying to @petergyang
“Running local models on Mac Minis and Mac Studios is looking more appealing now as a safety net” - However, we can’t just drop any model into a harness and expect it to be the same harness. It’s not really a safety net. We need a harness that’s optimized for these local models.
Replying to @flaviocopes
I built myself a workflow that helps me stay on top of AI development across X, Web and YouTube.
I personally track, curate and put together these into threads like this (because that really makes these concepts stick for me:
xitter.cf/goldenaidev/status/200…
📡 AI Agent Builder Tips: Dec 21 - 28
This week, top AI agent builders emphasized “smarter planning” to cut wasteful loops.
Tips from the builders:
• Write a spec first to mimic plan mode - from @simonw
• Use agents as main or subagents for modular builds - from @hwchase17
• Brainstorm with subagents to test new ideas fast - from @omarsar0
🧵
Replying to @zarazhangrui
I’ve been using these 2 ways to “claude on the go”:
🧠 Hacker mode: VPS + Termius + Happy
💬 Chatter mode: Claude iOS + GitHub + Vercel
I pick the right mode based on the project I’m working on.
Hacker mode for server scripts, when firing off commands is easier, more YOLO. Chatter mode when I’m working on web apps, everything is a PR.
Replying to @anshnanda
claude code in the browser - sandbox.dev is 🔥
Replying to @DanielleMorrill
I actually need Claude Code to run my home - manage school schedules for kids, order grocery, plan my travels, pay my bills etc. I’m ready to hand over the keys for Claude to take over 🔑
Replying to @Ghodakiya_Ankur
> keep it focused and don’t overbuild
That’s the crux of not going overboard with building with Claude Code.
Reminds me of this apple design principle:
“A designer knows he has achieved perfection not when there is nothing more to add, but when there is nothing left to take away”
Replying to @shiri_shh
I’m vibecoding all the personal software I need; like this fitness dash 👇
xitter.cf/goldenaidev/status/200…
Homer retweeted
Replying to @GergelyOrosz @leerob
Fullstack devs are going to be the first and fastest to adapt the AI dev mindset in 2026.
They already enjoy the breadth of engineering skills and have a product-first mindset.
2026 will be the year of the Fullstack dev rebranded as an AI dev 😊
Homer retweeted
Replying to @petergyang
Here are 2 setups I use daily to vibe-code with Claude:
🧠 Hacker mode: VPS + Termius + Happy
💬 Chatter mode: Claude Code on iOS + GitHub
Shipping > scrolling.
Homer retweeted
Replying to @dominikkoch
I was in the same boat. Watch this talk by @dexhorthy for an explanation on how MCP tools push agents into the “dumb zone”
youtu.be/rmvDxxNubIg?si=tY_g…
Homer retweeted
Replying to @ishaansehgal
💯being able to prompt is no longer a differentiator, literally anyone can do that now.
That’s why my bet is that when the dust settles the pure vibe coders won’t have much moat as they’re not building real “skills” 👇
xitter.cf/goldenai_/status/20054…
Vibe Coders are splitting into 2 camps 🔥
1/ Pure Vibe Coders: Ship fast without touching code. Pure builder energy – vibes only, no code review drama.
2/ Hybrid Vibe Coders: Seasoned engineers + AI agents = superpowers. They dive in, steer precisely, scale efficiently. True architect mode.
Hot take: Hybrids win long-term. They move faster, break less, and scale better.
Which camp are you in?
Pure, hybrid, or something else? 👇
Homer retweeted
Replying to @jefftangx
Claude on Terminus is 🔥
Happy takes it to the next level 🚀
github.com/slopus/happy
Homer retweeted
Replying to @Saboo_Shubham_
Vibe Coders are here - seeing 2 camps emerge:
1. Pure Vibe Coders - ship without seeing code. Builder-first energy.
2. Hybrid Vibe Coders - experienced engineers supercharged by agents. They dive deep, steer better, operate efficiently. Architect energy.
Hybrid wins long-term. Thoughts?
Homer retweeted
Replying to @vasuman
claude code with aisdk v6 + grok 4.
you can collapse many multi-agent systems into way simpler systems with these building blocks.
xitter.cf/goldenai_/status/20053…
Replying to @jayendra_ram
Homer retweeted
Replying to @jayendra_ram
Homer retweeted
Replying to @petergyang
re: personal OS in Claude on iPhone
If you’re running claude on your Laptop (via tailscale) or VPS (say digital ocean), then you could use Terminus or Happy to connect to it from your iPhone
These steps might help:
xitter.cf/goldenai_/status/20046…
🦀 Claude in the Cloud from iPhone
1. Get a VPS for $6/mo
2. Login & Install Claude
3. Use Termius from iPhone to SSH in
You now have your own Claude in the Cloud to code on the go, push, deploy, build agents - you don’t need vibe coding apps, you got Claude!
Homer retweeted
PSA Grok Live Search for web search is being deprecated on Dec 15 - Move to Agentic Search
docs.x.ai/docs/guides/live-s…
Homer retweeted
Replying to @simonw
Mobile vibecoding lets me do the thing vs planning/thinking/dreaming of doing the thing.
I’ve used it to learn - by building something small and then studying/tweaking the implementation like this 👇
xitter.cf/goldenai_/status/19944…








