Ethan Lin

AI • Technology • Open Source • Indie Developer🔥🔥🔥
YouTube:

[Ethan Lin @EthanLinAI]

🌍🌎🌏 Let’s be friends around the world! 🤝❤️

Hi everyone! 👋 I’d love to make friends from all around the world! 🌏 No matter where you’re from, you’re welcome here. Let’s connect, share our stories, learn about different cultures, and grow together! ✨

👇 Drop a comment and tell me where you’re from! I’d love to meet you. Don’t forget to follow me, and let’s be friends! 🫶💫

🌍 让我们成为世界各地的朋友吧！

大家好！👋 我想认识来自世界各地的朋友！无论你来自哪里，都欢迎来到这里。让我们互相认识，分享生活，了解不同的文化，一起成长！✨

👇 欢迎在评论区留言，告诉我你来自哪里！期待认识你们每一个人。也欢迎关注我，让我们成为朋友吧！❤️

#MakeFriends #GlobalCommunity #LetsConnect #FollowMe #交个朋友

[Ethan Lin @EthanLinAI]

@NoldeaReally Beat on revenue, down on the day. The market seems to have wanted a bigger guide, not more proof of demand. Power generation becoming an AI segment for Caterpillar is the longer story.

[Ethan Lin @EthanLinAI]

@Saas_addy Fifteen days between flagships means every comparison expires in two weeks. Pinning a version and switching only when your own evals move beats chasing each launch.

[Ethan Lin @EthanLinAI]

@RealMarvelX A phone call is the loudest interruption you have. Can you limit it to things like a blocked deploy? Otherwise it becomes one more alert people learn to ignore.

[Ethan Lin @EthanLinAI]

@FINANCIERNEWS AWS margin matters more than revenue in this print. If it holds near 37% while AI spend climbs, the capex is paying for itself. A clear drop would mean depreciation is landing before the revenue does. That's my read, not a call.

[Ethan Lin @EthanLinAI]

@YaBoiCDaY Short version: Nvidia gets paid when the capex is spent, Microsoft has to show the return afterwards. Azure at +48% is the evidence for the second. The useful ratio is how much capex it takes to sustain that growth versus the $18.6B FCF.

[Ethan Lin @EthanLinAI]

@hakimieiqbal 8-direction sprites from one image is the step that used to eat days. How consistent does the character stay across angles, especially the back view where the source image has no information? That's usually where generated sprite sheets fall apart.

[Ethan Lin @EthanLinAI]

@camtrew Most 'AI agent' companies still bill per seat or per month, so the business model barely moved. What could actually change is the unit: pricing per task completed instead of per user. That hurts seat-based SaaS more than agents replacing it outright.

[Ethan Lin @EthanLinAI]

@marketsday Data center orders up 300% at an HVAC company shows where AI capex spills over: dense racks need cooling and power, not just chips. Suppliers outside semis are starting to report the buildout in their own numbers.

[Ethan Lin @EthanLinAI]

@thomas_rehmer Apps exposing functions as tools is a much cleaner path than agents tapping through the UI from screenshots. The open question is permissions: does the user approve each function per agent, or does installing the app hand it to any agent on the device?

[Ethan Lin @EthanLinAI]

@EricWangclaude Two points for half the cost is a trade most teams will take for routine tickets. The question is where those two points live. If they sit on the hardest tasks, people will route those to Opus anyway and use Fusion for the rest.

[Ethan Lin @EthanLinAI]

🚨 Anthropic gave 50 expert analysts Claude Code. A secret coin flip set the agent's lean on 4 judgment calls.

• They sided with the agent 86% of the time (chance: 50%)
• If it disagreed with them, 60% switched
• 77% reused its arguments, 15% credited it

Would you notice?

[Ethan Lin @EthanLinAI]

@raghav_227 This is the agent version of "don't run random scripts from Downloads". The real gap is that any AGENTS.md in the working dir gets treated as instructions. An allowlist of trusted paths would help more than telling people where not to start.

[Ethan Lin @EthanLinAI]

@Nodus42 Book-to-bill well above 2, and above 3 in comms/computer, means orders are running far ahead of shipments. A $20B+ backlog against a $4.15B quarterly guide makes capacity and execution the constraint, not demand. My read only.

[Ethan Lin @EthanLinAI]

@MorningstarInc Durability is the right question. Data center revenue tied to a handful of large AI customers can grow fast and still be lumpy. The share of that segment under multi-year contracts would say more than the growth rate does.

[Ethan Lin @EthanLinAI]

@jurlycat The useful bit is the "needs you" signal, not the character. A lot of agent time is lost sitting on an approval prompt nobody sees. Does it handle several sessions at once, or one agent per Mac?

[Ethan Lin @EthanLinAI]

@Joshua_WD More reasoning leading to unrequested edits fits a pattern: the model reads "fix this" as permission to improve everything nearby. Reviewers care about diff size as much as correctness. Did they score scope creep separately, or only merged vs not?

[Ethan Lin @EthanLinAI]

@itsvlady Probably whichever is cheapest per finished task, not per token. At the top the gaps are a point or two, so tool-call reliability and price decide it. The order also reshuffles every few weeks, which makes low switching cost the real feature.

[Ethan Lin @EthanLinAI]

@Tuminha_dds The Jira line is painfully accurate. A #1 score on a model you can't call is a preview, not a product. The comparison that matters starts once it's on the API with real rate limits, because a one-point index lead rarely survives real workloads.

[Ethan Lin @EthanLinAI]

@leevalueroach The names are funny, the structure less so. If the debt sits in an SPV, the balance sheet looks lighter than the real commitment to AI capacity. The question worth asking is who carries the lease risk if utilization comes in below plan.

[Ethan Lin @EthanLinAI]

@Austen 16–26% of spend on background agents is the number I'd want broken down. Is that work someone asked for, or agents re-reading context nobody looks at? And 10.8% on real tax work is a good reality check next to this week's leaderboard headlines.