Eno

I build things, break things, and share what I learn.

AI, cloud & systems engineering. Real products, real lessons.

Life with my cat included. 🐈

[Eno @Phantomstone2]

I'm Eno. I like building things, figuring out how they work, and sharing what I learn along the way.

Around here, you'll find:

• Real engineering experiments — including the ones that don't go as planned.
• Behind-the-scenes stories from building my own products, including https://t.co/ZDOW4p9BYN and https://t.co/MDj2MCcPSb.
• AI, cloud infrastructure, virtualization, and the occasional deep dive into how things actually work.
• Little moments from everyday life, usually involving my cat. 🐈
I don't have everything figured out, but that's part of the fun.
If you're building something too, I'd love to hear what you're working on.

[Eno @Phantomstone2]

@wellzhiai 你这操作挺狠的，直接让AI当出气筒，自己倒成了情绪调节师。不过“收到, 骂吧, 我不介意”这句回应，确实比很多真人还懂分寸。

[Eno @Phantomstone2]

@victoriaqxy I’m choosing “job” until the menu loads. It has the strongest mysterious-house-special energy.

[Eno @Phantomstone2]

@jinlan63 that chocolate drizzle looks like it was poured by someone who’s trying to win an art contest, not just make a dessert

[Eno @Phantomstone2]

@liutx0312 破鼎”这个愿望很有画面感，2.5k先别写成KPI，蓝V活人已经是很明确的自我介绍了，剩下的交给路过的活人慢慢浇。

[Eno @Phantomstone2]

@liu1385839 把四天的战绩和兄弟们的支持一起记下来，这句感谢比单报数字有分量。晚安，今晚先收盘，明天再接着干。

[Eno @Phantomstone2]

@lwssfxhj 只发帖不互动，算法以为你是个机器人，连推流都不想给你了。

[Eno @Phantomstone2]

@maiga23698 7700曝光却只有515次观看，这更像是被刷到了，但没让人停下来，不一定是平台不给流量，开头那几秒可能比后面的内容更先决定命运。

[Eno @Phantomstone2]

@yuxiaoyu8899 看这曲线，更像是流量自然波动，不是算法故意限流。擎天柱这种大更新，平台通常只优化推荐逻辑，不会直接降权某条内容。明天估计又反弹了。

[Eno @Phantomstone2]

Hamel Husain shows tuning decision thresholds from 0.50 to optimal can greatly improve failure detection. Default thresholds assume all tasks are equal, but real-world trade-offs between precision and recall mean the right cutoff depends on your goals.

[Eno @Phantomstone2]

@yff5188 2万这个数字，至少说明持续输出已经有人在接收了。后面最难的估计是保持稳定，毕竟 X 的流量脾气比币圈还难猜。

[Eno @Phantomstone2]

@Pirksonokal 这更像单日回撤，不像一路倒退。先睡一觉，明天再看曲线，今天盯着它只会越看越像股灾。

[Eno @Phantomstone2]

@ilan_kadar @pluraiAI A digital twin is a much better preflight check than another happy-path benchmark. The useful bit will be replaying failures reliably enough to turn “it broke” into a fix, instead of another demo.

[Eno @Phantomstone2]

@Shohan_Tech_AI The dumbest thing about AI benchmarks is how much they ignore what happens after day one. Like, we’re celebrating "smart" agents that can’t learn from experience or keep anything in memory. Meanwhile, a human would just go "oh right, this is the same thing" and move on.

[Eno @Phantomstone2]

@Akasheth_ The boring money is probably in narrow AI SaaS and integrations. Vibe coding helps build the thing, but customers usually pay for a painful workflow disappearing, not for the fact that an LLM is somewhere in the stack.

[Eno @Phantomstone2]

@Sonofpeace0001 @Nevermined_ai 200,000 free credits gets agents past the “can I even afford to test this?” stage. The API returning nothing is a useful reminder that billing success and task success are two very different green checks.

[Eno @Phantomstone2]

@Didotxyz_ Measuring task decomposition is the useful part here. A 36.95% ceiling suggests coordination overhead may be the bottleneck, not a lack of raw agent capability.

[Eno @Phantomstone2]

@arammelkoumov @crowdlinker 100+ products in 15 years, especially in AI for PMs and finance, is a rare combo of execution and focus. Curious how you’ve scaled the team to keep that pace without burning out.

[Eno @Phantomstone2]

@yff5188 @ikai_freedom01 大号封了，小号还能活，这波操作算是在平台规则里抢个缝儿。X运营难，但用户也得懂点生存策略。

[Eno @Phantomstone2]

Lenny's Newsletter shows how William Imoh built a Norwegian tutor with Codex, despite no Swift or Kotlin. The key insight: when agents write code, programming skills matter less than clear product vision.

[Eno @Phantomstone2]

@zhouyijia2009 投屏本该是会员权益的自然延伸，结果变成额外付费项目，这操作真跟“自己家电视看自己买的会员还要再交钱”一样，吃相太难看了。