比 “加油” 更重要:抑郁症患者家属必知的三个被忽略的真相

比 “加油” 更重要:抑郁症患者家属必知的三个被忽略的真相 Mans International

你身边有没有那种变化大到让人心疼的人?我有个朋友,曾经是圈里的『小太阳』周末探店打卡,工作日晚上学唱歌,练跳舞,到各地旅行,浑身散发着对生活的热忱。

直到半年前的一个活动,她来办理退款。那天她站在我对面,眼神空洞得像蒙了层灰,和从前那个神采飞扬的模样判若两人。后来才知道,她女儿的抑郁症加重了,这大半年,她像被抽走了所有力气,几近崩溃。

如果你也是抑郁症患者的家属,或许能懂那种无力。今天想分享三个最容易被抑郁症患者家属忽略的真相:

一、“想开点” 救不了 TA,这不是你的错

朋友刚知道女儿生病时,做过很多现在想来 “徒劳” 的努力。她觉得 “行动能打败情绪”,特意规划了三亚的旅行,想着阳光沙滩总能让女儿开心起来。

结果呢?到了三亚,孩子把自己反锁在酒店房间,一天没出来。她坐在海边的礁石上,眼泪混着海水,默默抽泣。

《三联周刊》曾经报道过一位抑郁症的男朋友的亲身经历,我把链接放在这里,供您参考。陪伴女友治疗抑郁症近3年后,我们还是分手了-三联生活网

一位男友得知女友患抑郁症后,查遍了 “抗抑郁指南”,坚信 “运动和晒太阳能治愈一切”。他从一开始的 “求着出门”,到后来的 “命令下楼”,直到某天女友突然捂着脸大哭,他才发现 —— 那些他以为 “有益” 的事,对她而言竟是翻不过去的高山。

有位精神科医生的比喻很贴切:“你不会跟哮喘病人说‘别喘气’,对吧?” 抑郁症同样是大脑的生理病变 —— 就像免疫系统出问题会感冒,神经递质失衡也会让情绪 “生病”。它不是 “想不开”,更不是 “不够坚强”。

家属最该做的第一步,不是急着推 TA “好起来”,而是先接纳这个事实:这需要时间,更需要专业的帮助。你的 “努力” 没能立刻见效,从来不是因为你做得不够好。

二、你不用做 “拯救者”,“陪着” 就够了

去朋友家时,我看到她家冰箱上贴着的 “康复计划”——A4 纸打印,密密麻麻写满了时间节点:“7:30 必须吃含 Omega-3 的早餐”;“15:00 晒太阳不少于 30 分钟”;“20:00 亲子阅读(禁止负面话题)”……

她指着计划说:“我怕漏了什么,万一因为我没做好,孩子更严重了怎么办?”

这大概是很多家属的误区:把自己当成了 “拯救者”,觉得肩上扛着 “治好 TA” 的重任。可事实上,家属的陪伴,更像黑夜里陪对方坐一会儿。你不用点燃火把照亮前路,只要让 TA 知道 “你不是一个人” 就够了。

就像医生说的:“骨折的人需要拐杖,但拐杖从不会自责‘为什么不能代替腿走路’。你现在做的,就是让 TA 靠着你时,伤处能少受点力。”

TA 不想说话时,不用逼 TA 开口;TA 反复焦虑同一件事时,不用急着劝 TA “别多想”;TA 今天比昨天多吃了一口饭,哪怕只是对着你笑了一下,都是值得被看见的微光。

你在,本身就是一种力量。

三、别忽略自己,你的疲惫也该被看见

有次在超市,朋友拿着两瓶酸奶站了很久 —— 一瓶是黄桃味的,她自己爱喝;另一瓶是无糖款,是患抑郁症的女儿常喝的。最后她放下了黄桃味,攥着无糖款转身时,突然蹲在货架旁哭了。

“女儿都这样了,我哪还有资格想自己喜欢什么?连选瓶自己喜欢的酸奶都觉得是错的。” 她的话像根针,扎得人心里发疼。

她的手机相册,最新的 100 张照片全是女儿的:吃药的时间、今天吃的饭、坐在窗边发呆的侧影…… 朋友像把自己的生活彻底 “清空” 了,只装下了女儿的喜怒哀乐。

可你有没有想过:如果连你自己都在枯萎,又怎么能给TA浇水?

每天留哪怕 10 分钟发会儿呆,每周和我们这些朋友打个电话说说近况,甚至允许自己躲在卫生间哭一场 —— 这些 “微小的自私”,从来都不是对 TA 的辜负。

照顾好自己,才能更久地站在 TA 身边。你值得被看见,也需要被关怀。

最后想说的话

抑郁症患者的家属,就像走在一条没有路灯的路上。一边要扶着身边的人慢慢走,一边要给自己找光。这条路很难,你或许会疲惫、会自责、会无数次想放弃,但请一定记得:你不是孤军奋战。

那些你悄悄抹掉的眼泪,那些你强撑着的微笑,那些你深夜查过的资料,都在悄悄支撑着你们往前走。

愿每一个在黑暗中默默支撑的人,都能被这个世界温柔地接住。

如果你身边也有正在经历抑郁阴霾的人,请把这篇文章转发给TA。

把这句话轻轻说给TA听:

“再坚持一会儿,你不是一个人在战斗。”

Three Essential Truths for Supporting Someone with Depression

Love in the Shadows: Three Essential Truths for Supporting Someone with Depression KellyOnTech

Depression doesn’t just affect one person—it casts a shadow over everyone who loves them. If you’re walking alongside someone in this struggle, here are three often-overlooked truths that may help guide the way.

1. “Just Cheer Up” Won’t Help Them—And It’s Not Your Fault

When my friend first learned of her daughter’s illness, she thought sunshine and the ocean might help lift her daughter’s spirits. She carefully planned a trip to Hawaii. But when they arrived, her daughter locked herself in the hotel room and didn’t come out all day. My friend sat alone on the rocks by the sea, tears mixing with the saltwater as she quietly sobbed.

Depression isn’t something that sunshine or sheer willpower can fix. Just as you wouldn’t tell someone with asthma to “just breathe,” you can’t expect someone with major depressive disorder to “just think happy thoughts.” Depression is a medical condition—a neurochemical imbalance—not simply a matter of mindset.

The first and most important step for family members isn’t to rush their loved one into “getting better,” but to first accept this simple truth: healing takes time—and professional help. If your efforts haven’t brought immediate results, it’s not because you haven’t done enough or haven’t tried hard enough.

2. You Don’t Need to Be a “Saviour” — Just Being There Is Enough

When I visited my friend’s home, I noticed a printed “Recovery Plan” taped to the fridge — an A4 sheet filled with tightly packed schedules and rules:

“7:30 AM — Must eat a breakfast rich in Omega-3s”

“3:00 PM — Minimum 30 minutes of sunlight”

“8:00 PM — Parent-child reading (no negative topics allowed)”

Pointing to the plan, she said, “I’m scared I’ll miss something. What if things get worse because I didn’t do enough?”

This reflects a common misconception among families: thinking they have to be the “rescuer,” carrying the burden of “fixing” their loved one.

But the truth is, being there for someone with depression is less about lighting the path and more about sitting with them in the dark.

You don’t need to carry a torch — just let them know they’re not alone. That’s enough.

As one doctor wisely put it: “A person with a broken leg needs crutches, but the crutches don’t blame themselves for not being the leg.” What you’re doing now is offering something to lean on, to take the pressure off the pain.

If they don’t want to talk, don’t force it.

If they repeat the same worries over and over, don’t rush to say, “Stop overthinking.”

If they eat one more bite than yesterday, or even just manage a small smile in your presence, that’s a flicker of light worth noticing.

Your presence alone is a quiet, powerful form of support!

3. Don’t Forget Yourself — Your Exhaustion Deserves To Be Seen Too

Another time, at the grocery store, my friend stood frozen, holding two yogurt: her favourite peach, and the sugar-free one her daughter preferred. She put back the peach. Then, she crouched down by the shelf and cried. “With my daughter like this, how could I possibly care about what I like? Even picking a yogurt feels selfish.”

Her words cut deep. Her phone’s album told the story: the last 100 photos, all of her daughter — medication schedules, meals, a profile staring out the window. It was as if she’d emptied her own life just to make room for her daughter’s every emotion.

But have you considered: if you’re withering inside, how can you truly water someone else?

Ten minutes a day to just zone out. One call a week to chat with friends. Even allowing yourself to cry behind a bathroom door. These small acts of “selfishness” are not betrayals. These are ways to sustain yourself, so you can keep standing beside your loved one for longer.

You deserve to be seen. You need care, too.

Final Words

Family members supporting someone with depression are like travellers on a road without streetlights. You have to gently help your loved one take each step, while also finding your own source of light. This path is incredibly difficult — there will be exhaustion, guilt, and moments when you feel like giving up. But please, always remember: you are not fighting alone.

The tears you wipe away in secret, the smiles you force when you feel broken, the late nights spent researching — these quiet acts of love carry both of you forward.

May everyone silently supporting someone in the darkness be gently embraced by this world.

If you know someone currently experiencing the shadows of depression, please share this message with them. And quietly, softly, whisper these words:

“Hold on a little longer — you’re not fighting this battle alone.”

A New Chapter in Depression Care: An Innovative Therapy KellyOnTech

A New Chapter in Depression Care: An Innovative Therapy KellyOnTech

In past articles, we’ve focused mainly on innovative treatments for mild depression. Today, we turn to a breakthrough in the treatment of major depressive disorder (MDD) — a condition that affects millions worldwide. For the first time, patients with major depression can access a prescription digital therapeutic via their smartphone: Rejoyn.

What is Rejoyn?

Rejoyn is currently the only mobile app approved by the U.S. FDA as an adjunctive treatment for adults (22+) diagnosed with MDD who are already on antidepressants. Available by prescription only, Rejoyn offers a 6-week course that combines proprietary brain training exercises with medication to reduce depressive symptoms, with no significant side effects reported.

Unlike typical wellness apps, Rejoyn is classified as a medical device. It earned FDA approval in 2024 after successful Phase 3 clinical trials, with strong scientific evidence supporting its safety and efficacy.

How Does Rejoyn Work?

In many patients with depression, communication between the amygdala (emotion centre) and prefrontal cortex (thinking centre) becomes impaired, making emotional regulation increasingly difficult over time.

Rejoyn’s exercises are based on neuroplasticity, the brain’s natural ability to rewire itself. By simultaneously activating emotional and cognitive regions, Rejoyn strengthens these connections, helping the brain better manage emotional responses.

EFMT: The Emotional Faces Memory Task

One of Rejoyn’s key technologies is the Emotional Face Memory Task (EFMT), a scientifically validated exercise. Users are asked to recognize facial expressions (e.g., happy, sad, angry) and recall matching ones from earlier. This trains both working memory and emotional processing, gradually increasing in difficulty to challenge the brain to stay focused even in the face of emotional distraction.

In addition to EFMT, Rejoyn offers short skill-building lessons that teach users how to manage strong emotions, combat negative thinking, and take positive actions. Each lesson is around 5 minutes, and it is recommended to do it three times a week. Motivational reminders are sent to help users stay on track.

Importantly, Rejoyn is not a replacement for medication, but a complementary tool prescribed alongside existing antidepressant treatments. For patients seeking accessible, structured support in their daily lives, Rejoyn is a promising new option.

Advances in medical science have led to FDA-approved devices like Rejoyn, opening new pathways for treating depression. Still, even the most advanced technologies must be supported by compassionate, human-centred care to have a meaningful impact.

In our next issue, I want to speak directly to the families and loved ones standing by someone with depression. There are important truths you may not yet recognize, but understanding them can make all the difference in your journey together.

FDA 权威认证!Rejoyn 为重度抑郁症治疗注入新活力 KellyOnTech

重度抑郁症治疗新进展 KellyOnTech

在过往的分享中,我们更多聚焦于轻度抑郁症的创新疗法。而今天,我们将目光投向重度抑郁症领域,为关注这一领域的读者介绍一项重要进展:重度抑郁症患者如今可借助手机应用进行辅助治疗。

这款名为 Rejoyn 的应用,是目前唯一通过美国食品药品监督管理局(FDA)认证的重度抑郁症辅助工具,其适用人群为 22 岁及以上、正在接受抗抑郁药物治疗的成年重度抑郁症(MDD)患者。

与普通健康类应用不同,Rejoyn 被归类为医疗器械,并于 2024 年在完成三期临床试验后获得了 FDA 的批准,其安全性和有效性具有坚实的科学依据。

值得注意的是,该应用需凭处方获取。Rejoyn 在抑郁症症状治疗方面有着独特的机制。对于部分抑郁症患者而言,大脑中负责情感处理的杏仁核与负责思维活动的前额叶皮质之间存在沟通障碍。这导致大脑难以正常处理情绪,随着负面情绪的不断累积,情绪处理的难度会呈螺旋式上升。

Rejoyn 的核心原理是激发患者大脑自身的潜能来对抗抑郁症。它运用了神经可塑性这一科学原理 —— 即大脑天生具备改变和重塑的能力。

在为期 6 周的疗程中,它通过专利性的大脑练习与抗抑郁药物协同作用,旨在同时激活大脑的思考区域和情感区域,以此加强两者之间的联系,促进它们的有效沟通,进而帮助大脑更好地处理情绪,改善抑郁症状,且未发现明显副作用。

Rejoyn 的核心原理是激发患者大脑自身的潜能来对抗抑郁症。它运用了神经可塑性这一科学原理 —— 即大脑天生具备改变和重塑的能力。Rejoyn 设计的大脑训练练习,旨在同时激活大脑的思考区域和情感区域,以此加强两者之间的联系,促进它们的有效沟通,进而帮助大脑更好地处理情绪,改善抑郁症状。

情绪面孔记忆任务(EFMT)

Rejoyn 在重度抑郁症治疗方面有着独特的机制,技术之一便是情绪面孔记忆任务(EFMT)。

在 EFMT 任务中,用户需要识别一系列面部表情所传递的情绪,并回忆之前看到过的相同情绪。这项练习巧妙地结合了工作记忆训练与情绪处理两方面内容:工作记忆训练要求用户记住各种模式或序列,情绪处理则通过展示快乐、悲伤、愤怒等不同情绪的面孔来实现。随着用户操作的深入,任务难度会逐渐增加,以此挑战大脑在应对情绪干扰的同时保持专注。

除了这些核心的大脑训练练习,Rejoyn 还提供简短的技能治疗课程,教授用户一些有助于管理抑郁症状的实用技能,包括应对强烈情绪、抑制消极思维以及采取积极行动等。每节课程时长约 5 分钟,建议每周进行 3 次。改变大脑的运作模式需要持续的努力,为此,平台会发送激励短信,以强化用户对练习和课程的坚持。

需要明确的是,Rejoyn 并非一种独立的治疗方法,不能替代患者目前正在使用的药物,患者必须遵照医嘱将其作为辅助治疗手段。对于正在接受药物治疗的重度抑郁症患者来说,Rejoyn 无疑为他们提供了一种可随时使用的 “大脑锻炼工具”,是该领域一个值得关注的新选择。

医学的进步带来了像 Rejoyn 这样的 FDA 认证医疗器械,为抑郁症治疗开辟了新路径。然而,再先进的技术也需要温暖的人文关怀来支撑。

下期我想和所有陪伴在抑郁症患者身边的无名英雄 – 家属们聊聊:有些容易被忽视的真相,也许你还没意识到,但它真的很重要。

视频版

Decoding China’s AI Advantage: Insights from Mary Meeker’s Report

China is rapidly emerging as a global front-runner in the race to integrate AI with consumer hardware.

On June 26, 2025, Xiaomi officially launched its first pair of AI glasses, positioned as “the personal intelligent device of the next era and your portable AI gateway,” directly challenging the Ray-Ban Meta AI glasses.

Image source: Xiaomi, Xiaomi AI glass

Core Feature Comparison: Xiaomi AI Glasses vs. Meta AI Glasses

Here’s the comparison table with prices converted from CNY to USD using the latest exchange rate (¥1 = $0.1396 as of early July 2025):

FeatureXiaomi AI GlassesMeta AI GlassesAdvantage
Price¥1,999 ≈ $279¥2,400 ≈ $335Xiaomi (cheaper)
Weight40g (without lenses)49g (frame)Xiaomi (lighter)
AudioDual speakers + 5 mics with sound leakage control2 speakers + 5 micsXiaomi (better quality & privacy)
Camera12MP (Sony IMX861), 2K 30fps video, phone integration12MP ultrawideXiaomi (video calls, scanning, livestreaming)
AI FeaturesReal-time multilingual translation (10 languages), LLM Q&A, smart home controlMeta AI Q&A onlyXiaomi (richer AI capabilities)
Battery Life8.6 hours (typical use)8 hoursXiaomi (longer life)
Charging45 min (USB-C)75 minXiaomi (faster)
EcosystemDeep integration with HyperOS, Mi AI assistant, and smart home devicesMeta ecosystemXiaomi (stronger interconnectivity)
Style3 colors (black, tortoise brown, parrot green)20+ Ray-Ban stylesMeta (more fashion-forward)

While Xiaomi lags slightly behind Meta in terms of fashion appeal and brand recognition, it demonstrates clear advantages in hardware performance, AI interaction, ecosystem integration, and pricing, especially within the Chinese market.

With more Chinese companies entering the space, leveraging lightweight design, intelligent features, high cost efficiency, and tightly integrated ecosystems, China is poised to lead global consumer AI hardware-software innovation.

China and the U.S. in Foundational AI Research: Each With Its Own Edge

According to Trends-Artificial Intelligence, the BOND AI trends report released by “Internet Queen” Mary Meeker, the U.S. and China are now in an intense phase of competition in AI. In foundational research, each country holds distinct strategic advantages.

🇨🇳 China’s Strength: Dual Engines of Open-Source Ecosystems and Industrial Intelligence

  1. China is rewriting the global playbook for AI through the explosive growth of its open-source ecosystem and rapid AI industrialization:
  • Scale and Quality of Open Models: As of Q2 2025, China has released several benchmark open-source models, including DeepSeek-R1 (trained at just 1/10 the cost of OpenAI’s), Alibaba’s Qwen-32B, and Baidu’s Ernie 4.5, covering a wide range of use cases from language to multimodal and code generation.
  • World’s Largest Open Model Hub: Alibaba Cloud’s ModelScope now hosts over 70,000 open models and has a developer base of 16 million, making it one of the largest open-source AI communities globally.
  • Affordable, High-Performance AI: Chinese models like DeepSeek V3 are optimized for cost-efficiency, offering industry-grade performance at a fraction of traditional training costs (as low as 1/10). This affordability accelerates the democratization of AI in SMEs and developing nations, enabling broader global access to advanced technology.
Image source: Mary Meeker BOND Trends-Artificial Intelligence Page 265

2. China’s Lead in Industrial Robotics: A Result of Policy Support and Supply Chain Strength

  • Global #1 in Installations: In 2023, China accounted for over 50% of the world’s industrial robot installations. International manufacturers such as BMW and Tesla are already incorporating Chinese robotic solutions into their production lines.
  • Challenges Remain: Despite rapid progress, China still relies heavily on imported high-end sensors and real-time control systems. However, domestic substitution is accelerating as local innovation and investment ramp up.
Image source: Mary Meeker BOND Trends-Artificial Intelligence Page 288

🇺🇸 U.S. Strength: Dual Advantage in Foundation Models and Chip Dominance

  1. The U.S. continues to hold a clear edge in the development of core AI foundation models:
  • Closed-Source Superiority: Proprietary models like OpenAI’s GPT‑4.5 and Anthropic’s Claude consistently lead on benchmarks such as MMLU and HumanEval, particularly in complex reasoning and long-context understanding.
  • Open-Source Closing In: While Chinese models like DeepSeek-R1 have rapidly narrowed the performance gap (from 15.9% in 2023 to just 1.7% by 2025), closed-source U.S. models still dominate high-end commercial ecosystems, especially in enterprise and advanced research use cases.
Image source: Mary Meeker BOND Trends-Artificial Intelligence Page 142

2. Chip and Compute Supremacy: From Hardware Monopoly to Global Ecosystem Control

The U.S. has secured a dominant position in global AI infrastructure through its leadership in chip technology:

  • NVIDIA’s Absolute Dominance: NVIDIA GPUs power over 90% of AI model training worldwide. Its hardware, particularly the H100 and H200 series, remains the gold standard for large model training.
  • Infrastructure Leverage: NVIDIA-related compute accounts for an estimated 25% of global data centre capital expenditure (CapEx), underscoring its strategic role in the AI economy, from cloud providers to model developers.
Image source: Mary Meeker BOND Trends-Artificial Intelligence Page 109

Export Controls Spark a New Paradigm of “Accelerated Evolution” in China’s Chip Industry

Amid the growing U.S.–China tech decoupling, particularly around semiconductors, U.S. export restrictions on AI chips to China have triggered a powerful “crisis-response mechanism” across China’s chip sector:

  • Technological Pressure: Real-world deployment demands are driving architectural innovation and improved chip yields.
  • Market Pull: Rising substitution demand for domestic chips is fueling greater R&D investment.
  • Strategic Loop: A dual-cycle strategy is emerging — import substitution at home, technology export abroad.

Beyond Huawei, Xiaomi has also entered the self-developed chip race — primarily to reduce material costs (mirroring how Apple’s M1 chip boosted Mac profit margins) and to mitigate geopolitical risk, such as potential sanctions or supply disruptions similar to Huawei’s.

The launch of Xiaomi’s Surge O1 chip marks the company’s first fully self-developed System-on-Chip (SoC) for smartphones, giving it core control from architecture design to functional integration. This shift reduces dependency on Qualcomm, lowers licensing fees, and improves gross margins, similar to how Apple used in-house chips to reduce reliance on Intel and unify its mobile and PC ecosystems.

In 2024, even NVIDIA CEO Jensen Huang acknowledged the shift, stating: “Export controls on China have failed. Our market share in China has dropped from 95% to 50% in just four years.”

Strategic Battleground

According to a study by MacroPolo, the think tank under the Paulson Institute, nearly half of the world’s top AI researchers received their undergraduate degrees from Chinese universities, compared to just 18% from U.S. universities. In 2019, that figure for China was only 29%, highlighting the country’s remarkable progress in cultivating world-class AI talent.

Take DeepSeek, for example — a team composed largely of graduates from top Chinese universities such as Tsinghua and Peking University. The team exemplifies a new wave of AI talent: highly educated, younger, open-source driven, and innovation-focused.

Similarly, Chinese talent is playing a critical role in Elon Musk’s Tesla Robotaxi project. Notably:

  • Pengfei Duan, Tesla’s Chief Software Engineer for AI, earned his undergraduate degree at Wuhan University of Technology.
  • Charles Qi, the machine learning engineer behind Tesla’s Full Self-Driving (FSD) system, graduated from Tsinghua University.

These cases clearly demonstrate that China’s homegrown AI talent is now shaping global innovation at the highest levels.

Summary

In the global AI race, China is swiftly narrowing the lead, driven by two key strengths: seamless hardware–software integration and a rising wave of homegrown AI talent. The launch of Xiaomi’s AI glasses marks not only a milestone for China’s consumer AI hardware but also highlights its unique strengths in ecosystem integration and cost-effectiveness. Meanwhile, China’s momentum in open-source models and industrial robotics is actively reshaping the global AI landscape.

At the same time, the United States maintains a firm grip on the foundations of AI — core algorithms and semiconductor supremacy, particularly in high-end compute infrastructure and closed-source model ecosystems. The core of this technological rivalry has shifted from pure performance benchmarks to a full-scale competition of ecosystems and innovation models.

Looking ahead, as China continues to export top AI talent and deepen its industrial chain integration, the global AI landscape may be on the brink of a major realignment. As NVIDIA CEO Jensen Huang put it, “Technology blockades only accelerate innovation.” In this silent war of systems and scale, China is advancing its own AI era, driven by openness and grounded in real-world applications.

AI 创新浪潮下的中国路径:软硬件一体化引领未来 —— 解读 “互联网女皇” Mary Meeker 趋势报告

中国引领AI技术与硬件一体化

在AI软硬件一体化的赛道上,中国厂商正以惊人的速度赶超全球竞争对手。

2025年6月26日,小米正式发布首款小米AI眼镜,定位“面向下一个时代的个人智能设备, 随身的AI入口”,直接对标Ray-Ban Meta AI 眼镜。

图片来源:雷科技

核心功能对比:小米AI眼镜 vs. Meta AI眼镜

功能类别小米 AI 眼镜Meta AI 眼镜优势对比
价格1999 元+2400 元+小米更便宜
重量40g(无镜片)49g (镜框)小米更轻
音频功能双扬声器 + 五麦克风,漏音控制优秀2 扬声器 + 5 麦克风小米音质更好,漏音控制更优
摄像功能1200 万像素(索尼 IMX 861),2K 30fps 视频1200 万像素超广角小米支持手机联动(视频通话、直播、扫码支付)
AI 功能支持 10 种语言同声传译、大模型实时问答、智能家居控制不支持翻译,仅支持 Meta AI 问答小米 AI 生态更强
续航(单次充电)8.6 小时 (典型使用)8 小时 (典型使用)小米续航更长
充电速度45 分钟(Type-C)75 分钟小米充电更快
互联生态深度集成小米澎湃 OS、小爱同学、米家智能家居依赖 Meta AI 生态小米互联能力更强
时尚款式3 种配色 (黑、玳瑁棕、鹦鹉绿)20+ 种款式(Ray-Ban 合作)

小米 AI 眼镜虽然在时尚性和品牌影响力上略逊于 Meta,但在硬件性能、AI交互、生态协同、性价比等方面展现出强大竞争力,尤其是在中国市场,其软硬一体化能力和澎湃 OS 生态联动使其具备独特的不可替代性。

在随着越来越多的中国厂商介入,凭借轻量化、智能化、高性价比和生态整合能力,中国正在引领全球AI消费级软硬件产业。

中美在人工智能基础研究领域各有什么领先优势

根据被誉为“互联网女皇”的Mary Meeker发布的BOND AI趋势报告《Trends-Artificial Intelligence》,中美人工智能竞争已进入白热化阶段,其中在基础研究领域各具领先优势。

(一)中国的领先领域:开源生态与工业智能化的“双轮驱动”

  1. 中国AI开源生态的爆发式增长,正在改写全球技术共享的规则:
  • 模型数量与质量并进:截至2025Q2,中国已推出DeepSeek-R1(训练成本仅为OpenAI的1/10)、阿里Qwen-32B、百度Ernie 4.5等标杆级开源模型,覆盖语言、多模态、代码生成等场景。阿里云推出的AI大模型开源社区–魔搭社区更以7万个开源模型、1600万开发者的规模,成为全球最大AI开源社区之一。
  • 低成本普惠化:中国AI大模型以“高性能+低成本”见长,例如DeepSeek V3的训练成本优化至行业1/10,推动全球AI技术下沉至中小企业和发展中国家。
图片来源:Mary Meeker BOND AI趋势报告《Trends-Artificial Intelligence》265 页
  1. 中国在工业机器人领域的领先,是政策红利+产业链优势的必然结果:
  • 安装量全球第一:2023年中国工业机器人装机量占全球50%以上,宝马、特斯拉等国际大厂均采用中国机器人解决方案。
  • 挑战:中国工业AI的短板在于高端传感器和实时控制系统仍依赖进口,但“国产替代”正在加速
图片来源:Mary Meeker BOND AI趋势报告《Trends-Artificial Intelligence》288 页

(二)美国领先领域:基础模型与芯片霸权的双重优势

1. 基础模型研发:技术代差与开源追赶

美国在AI基础模型领域仍保持显著优势:

  • 闭源模型领先:OpenAI的GPT-4.5、Anthropic的Claude等模型在MMLU、HumanEval等基准测试中综合性能领先,尤其在复杂推理和长文本理解上存在技术代差。
  • 开源生态竞争加剧:尽管Meta的Llama、中国的DeepSeek-R1等开源模型性能快速逼近(差距从2023年的15.9%缩小至2025年的1.7%),但产业界主导的闭源模型仍控制着高端应用生态。
图片来源:Mary Meeker BOND AI趋势报告《Trends-Artificial Intelligence》142 页

2. 芯片与算力霸权:从硬件垄断到全球生态

美国通过芯片技术卡位全球AI基础设施:

NVIDIA的绝对统治:其GPU占据全球AI训练芯片90%以上份额,数据中心资本支出(CapEx)占比达25%,H100/H200系列仍是大模型训练的“黄金标准”。

图片来源:Mary Meeker BOND AI 趋势报告《Trends-Artificial Intelligence》109 页

封锁催生中国芯片“加速进化”新范式

在中美芯片技术脱钩背景下,美国对华AI芯片出口限制。然而祸兮福所倚,美国的限制政策却激活了中国芯片产业的“危机响应机制”:

  • 技术层面通过场景倒逼良率提升和架构创新;
  • 市场层面刚需替代率攀升反哺研发投入;
  • 战略层面形成“国产替代-技术输出”双循环。

除了华为,小米也推出了自研芯片,主要目的是降低物料成本(参考苹果M1芯片对Mac产品线利润率的提升),并规避美国制裁风险(如华为遭遇的断供危机)。玄戒O1的推出意味着小米首次实现手机核心系统级芯片(System on Chip,简称 SoC)自主化,掌握芯片核心技术,实现从芯片架构设计到功能集成等关键环节的自主可控。未来可减少向高通支付专利费,提升毛利率。类似苹果通过自研芯片摆脱Intel依赖,实现从移动端到PC端的全生态整合。

2024年,英伟达CEO黄仁勋直言:“对华芯片管制是失败的,四年间其在中国市场份额从95%跌至50%”。

(三)关键博弈点

根据美国保尔森基金会旗下麦克罗波洛智库(MacroPolo)的研究,从出身的本科院校来看,中国高校几乎培养了全球一半的顶尖 AI 研究人员,相比之下,仅有约 18% 研究人员来自美国大学。在2019 年,本科毕业于中国高校的顶尖 AI 研究人员占全球的比例还只有 29%,这一数据的大幅跃升体现了中国在 AI 人才培养方面的卓越成效。

以DeepSeek团队为例,其成员主要来自清华、北大等国内顶尖高校,团队呈现出”高学历年轻化、开源导向、创新驱动”的鲜明特质。

同样,在马斯克旗下特斯拉自动驾驶 Robotaxi 项目中,华人团队构成了核心技术力量:武汉理工大学本科毕业的段鹏飞担任特斯拉人工智能首席软件工程师;负责 FSD(全自动驾驶)开发的机器学习工程师祁芮中台(Charles Qi),本科毕业于清华大学。

这些案例清晰表明,在全球人工智能产业加速发展的浪潮中,中国本土培养的人才已展现出强劲的技术实力与创新潜力。

在全球AI竞赛中,中国正以软硬件一体化和人才红利两大优势加速追赶。小米AI眼镜的发布,不仅是中国消费级AI硬件的里程碑,更展现了生态整合与性价比的独特竞争力。与此同时,中国在开源模型和工业机器人领域的快速崛起,正在重塑全球AI技术格局。

然而,美国仍牢牢掌控基础算法与芯片霸权,尤其在高端算力和闭源模型上占据主导。这场科技博弈的核心,已从单纯的技术比拼,升级为生态体系与创新模式的全面竞争。

未来,随着中国AI人才的持续输出和产业链的深度整合,全球AI版图或将迎来新一轮洗牌。正如黄仁勋所言:“技术封锁只会加速创新”——在这场没有硝烟的战争中,中国正用开放生态和场景落地,书写属于自己的AI时代。

Your Ticket to the AI-Native Future: Core AI Subscriptions KellyOnTech

Have you got your ticket to the AI-native world — the “Core AI Subscription”?

This concept was introduced by Sam Altman, CEO of OpenAI, at the 2025 Sequoia AI Summit. More than just a new term, it could become the central theme of AI commercialization in the years to come.

For investors, this signals a shift: the future winners won’t be the ones merely selling models, but those building long-term user relationships and platform-level capabilities. 

For entrepreneurs, it offers clear direction: those who can create truly personalized AI assistants, and a sustainable business model around them, will be the ones to rise above the rest.

The “Core AI Subscription” is more than just a new business model; it may signal that humanity is accelerating toward a new era: the AI-native world.

What Is the AI-Native World?

The AI-native world refers to a future paradigm where artificial intelligence is not just a tool, but the foundational driving force deeply embedded in every aspect of society. It reshapes how we live, work, produce, and interact, much like electricity and the internet redefined previous eras.

In this world, AI becomes core infrastructure, pervasive and indispensable. It transforms human cognition, economic models, technological ecosystems, and social structures.

Imagine life in an AI-native world:

  • You no longer use AI just to complete isolated tasks.
  • AI understands your intentions, anticipates your needs, and acts before you even ask.
  • It’s seamlessly integrated into your life, work, and decision-making processes, becoming an essential part of your digital existence.

Core AI Subscription: The Gateway to the AI-Native World

Sam Altman’s concept of the “Core AI Subscription” is not just visionary — it represents the key pathway to realizing the AI-native future. It refers to a highly personalized, continuously evolving AI assistant service that is deeply embedded in users’ daily lives, much like an operating system that runs across every aspect of work and life.

This service is far more than just a voice assistant or chatbot. It functions as an intelligent agent with the following capabilities:

  • Personalized customization based on your habits and behaviours;
  • Seamless integration with other applications and services;
  • Constant learning and self-improvement over time, becoming smarter and more efficient the more you use it.

In other words, whoever owns the user’s Core AI Subscription essentially controls the “operating system entry point” to the AI-native world.

What AI Capabilities Are Required for Core AI Subscription Services?

To bring “Core AI Subscription” to life, there’s a key question we might want to explore: What kind of AI is capable of supporting such a service?

OpenAI has proposed a tiered framework for Artificial General Intelligence (AGI) — AI systems with the ability to learn efficiently, generalize across tasks, and act autonomously in complex, dynamic environments. True AGI would possess a blend of perception, cognition, decision-making, learning, execution, and social collaboration, all while aligning with human emotions, ethics, and moral standards.

Here’s a breakdown of the AGI capability tiers:

  • Level 1: Chatbot — Basic conversational ability, like current GPT models.
  • Level 2: Reasoner — Can solve human-level problems — mathematics, logic, coding, and debugging.
  • Level 3: Agent — Acts on behalf of the user — booking travel, managing calendars, and automating task chains.
  • Level 4: Innovator — Capable of invention and creativity — designing new products, writing screenplays, composing music.
  • Level 5: Organizer — Manages teams, coordinates resources, sets strategies, and even runs companies.

What AGI Level Is Needed to Enable Core AI Subscription Services?

To bring Core AI Subscription services to life, the AI must reach at least Level 3 — Agent on the AGI scale. At this level, AI isn’t just passively responding to user commands — it must actively understand user needs, take initiative, trigger tools, execute task chains, and switch contexts fluidly across various scenarios.

Since 2023, Baidu founder Robin Li has echoed a similar vision, stating that “large models will usher in a flourishing ecosystem of AI-native applications.” He emphasized that AI-native applications are not simple replicas of mobile apps or desktop software — they are meant to “solve problems that were previously unsolvable or poorly solved.”

This vision aligns closely with the concept of Core AI Subscription: true AI-native products are those in which AI agents are deeply embedded in users’ lives and workflows as a systemic, always-on digital partner.

Open Evidence: Core AI Subscription in Action in Healthcare

Are there early pioneers building AI-native applications? Yes—and a standout is Open Evidence, a medical AI company founded in 2021. By February 2025, it had raised $75 million from Sequoia Capital and achieved unicorn status with a valuation surpassing $1 billion.

At the 2025 Sequoia AI Summit, co-founder Zach shared a real-world case showing how their Core AI Subscription model supports physicians:

Emergency In-Flight Medical Case

Dr. Susan Wilberg faced a medical emergency mid-flight: a 63-year-old male cancer patient on immunosuppressive therapy developed a severe rash. Suspecting shingles, she had to make a critical call—should the plane turn back? What immediate actions were needed onboard?

She turned to ClinicalKey AI, Open Evidence’s subscription-based platform built for medical professionals. It delivered instant, personalized guidance by combining:

  • CDC Yellow Book protocols,
  • The latest research on cancer immunotherapy, and
  • Patient-specific recommendations (based on age, history, treatment, etc.).

The platform:

  • Assessed the patient’s risk level given his immunosuppressed condition,
  • Offered specific and timely treatment guidance,
  • Helped avoid an unnecessary emergency landing while ensuring proper care upon arrival.

What Makes ClinicalKey AI a True Core AI Subscription?

Open Evidence’s AI assistant is more than a diagnostic aid—it functions like an intelligent agent that continuously learns, personalizes its output, and proactively supports users:

  • Hyper-personalization: Tailors suggestions based on user preferences and patient context.
  • Seamless integration: Connects effortlessly with existing medical systems and workflows.
  • Continuous evolution: Becomes smarter and more efficient through real-world interactions.

Business Model & User Growth

Over 25% of U.S. practicing physicians now rely on Open Evidence daily. The system handles more than 10 real-time clinical questions per second. While the service is free for doctors, revenue comes from medical device and pharmaceutical advertising, mirroring consumer internet models, but adapted for healthcare.

To deepen value and retention, Open Evidence is embedding top physicians’ expertise, starting with gastroenterology, into its AI, creating a collective intelligence layer. This not only strengthens its data advantage but enables constant answer refinement.

The Future: AI as an Indispensable Partner

Looking ahead, Open Evidence plans to integrate broader medical reasoning, research capabilities, and workflow tools to build a fully-fledged Core AI Subscription platform, ultimately becoming a mission-critical partner to doctors worldwide.

If you’re a business owner aiming to integrate an AI subscription model into your operations, here are three essential principles to keep in mind:

1. Shift from “function thinking” to “companionship thinking.”
Don’t just ask, “What can AI do for my business?” Instead, consider, “What do my users need AI to become?” A CFO doesn’t simply need a reporting tool—they need a proactive financial partner that can anticipate risks and guide decisions.

2. Capture high-frequency “scene entry points.”
Identify must-have, recurring scenarios—such as in healthcare, legal services, or vertical workflows—and embed AI deeply into those daily user moments. Your goal is to make AI a seamless, indispensable part of how users work.

3. Build a “subscription-based emotional account.”
Offer consistent, meaningful value—like weekly personalized insights—to create a sense of FOMO (fear of missing out). When users feel your AI is essential to staying ahead, loyalty follows naturally.

By applying these strategies, you can turn AI from a one-off tool into a trusted, subscription-powered companion that users depend on.

中文版

AI 原生商业洞察:核心 AI 订阅服务 KellyOnTech

你有没有拿到通往 AI 原生世界的门票 — “核心AI订阅服务”(Core AI Subscription)?

这个概念是 OpenAI 的 CEO 山姆·奥特曼(Sam Altman)在2025年红杉AI峰会上提出来的。

对投资人来说,这是一个信号:未来的赢家,不是卖模型的公司,而是能构建长期用户关系、具备平台级能力的AI服务商。

对创业者而言,这也是一个方向:谁能打造真正“贴身”的AI助手,并建立可持续的商业模式,谁就最有可能脱颖而出。

核心AI订阅服务”(Core AI Subscription)不仅是一个新的商业模式,更可能预示着人类正在加速进入一个全新的时代——AI 原生世界(AI-native World)。

一、什么是“AI 原生世界”?

“AI 原生世界”是指以人工智能技术为底层核心驱动力,深度融入社会各领域运行逻辑,从而形成一种全新生产生活方式与社会架构的未来世界形态。

在这个世界中,AI不再是简单的辅助工具,而是像电力、互联网一样成为基础设施,重塑人类的认知方式、经济模式、技术生态和社会关系。

想象一下,在“AI 原生世界”中:

  • 你不再只是使用AI来完成某个任务;
  • AI会主动理解你的意图,预测你的需求,并在你开口之前就完成任务;
  • 它无缝嵌入你的生活、工作和决策流程,成为你数字生活中不可或缺的一部分。

二、“核心 AI 订阅服务”:通往 AI 原生世界的入口

山姆·奥特曼所提出的“核心AI订阅服务”,正是实现这一愿景的关键路径。它是一种高度个性化、持续进化、深度嵌入用户日常生活的AI助手服务,就像操作系统一样,贯穿你工作的每一个环节。

这个服务不仅仅是语音助手或客服机器人,而是一个具备以下能力的“智能代理”:

  • 根据你的行为习惯进行个性化定制;
  • 能与其他应用和服务无缝对接;
  • 随着使用时间的增长不断学习、进化,变得更加智能和高效。

换句话说,谁掌握了用户的核心AI订阅服务,谁就掌控了 AI 原生世界的“操作系统入口”。

三、要实现“核心 AI 订阅服务”,需要什么样的 AI 能力?

这就引出了另一个关键问题:什么样的 AI 才能支撑这样的服务?

OpenAI 曾提出一套通用人工智能(AGI, Artificial General Intelligence)的能力分级体系。AGI 指的是具有高效学习和泛化能力的人工智能体,能够根据复杂动态环境自主生成任务、执行任务,并具备感知、认知、决策、学习、执行和社会协作等综合能力,同时符合人类情感、伦理与道德观念。

这套 AGI 等级标准如下:

  1. 第一级:聊天机器人– 具备基本对话能力,如当前的GPT系列模型。
  2. 第二级:推理者– 可以解决人类水平的问题,比如数学、逻辑推理、编程调试等。
  3. 第三级:代理(Agent)– 能够代表用户采取行动,例如预订行程、安排日程、自动完成任务链等。
  4. 第四级:创新者– 能够参与发明创造,如设计新产品、撰写剧本、创作音乐等。
  5. 第五级:组织者– 能够管理团队、协调资源、制定战略,甚至运营企业。

四、从 AGI 等级看“核心 AI 订阅服务”的门槛

要真正实现“核心AI订阅服务”,至少需要达到 AGI 第三级——代理级别。这意味着AI不仅要能理解用户的需求,还要能主动采取行动、调用工具、执行任务链,并能在不同场景中灵活切换。

从2023年开始,百度创始人李彦宏也多次强调:“大模型将开启一个繁荣的AI原生应用生态。”他指出,AI原生应用不是对移动互联网App和PC软件的简单重复,而是要能“解决过去解决不了或解决不好的问题”。

这与“核心AI订阅服务”的理念高度一致:真正的 AI 原生应用,是那些能让 AI 作为“代理”深度参与用户生活和工作的系统级产品。

有没有AI原生应用领域的先行者呢?有的。

五、Open Evidence:AI 订阅式服务在医疗领域的应用

成立于2021年的 Open Evidence是一家专注于医学人工智能的公司。截至2025年2月,Open Evidence 已获得红杉资本7500万美元的投资,估值突破10亿美元,成功跻身独角兽行列。

在2025年的红杉资本AI峰会上,Open Evidence 联合创始人 Zach 分享了一个真实的应用场景,展示了他们如何通过“核心AI订阅服务”为医生提供支持。

紧急空中救援案例

Susan Wilberg 医生在一次飞行中遇到一个紧急病例:一名63岁的男性患者出现了严重的皮疹。考虑到该患者正在接受前列腺癌治疗,处于免疫抑制状态,她怀疑是水痘。在这种紧急情况下她需要尽快判断是否需要返航?飞机航行过程中应采取什么紧急措施?

Susan 使用了 Open Evidence 的 ClinicalKey AI 平台,这是一个基于订阅的服务,专为医疗专业人士设计。该平台不仅即时提供了来自CDC《黄皮书》和最新癌症免疫治疗研究的信息,还根据患者的特定情况(如年龄、病史、当前治疗方案等)给出了个性化的建议:

  • 评估了免疫抑制患者的风险严重程度;
  • 提供了及时且针对具体情况的治疗建议;
  • 避免了不必要的迫降,并为落地后的进一步治疗做了规划。

核心AI订阅服务的特点

在2025年的红杉资本AI峰会上,Open Evidence 联合创始人Zach分享了一个真实的应用场景,展示了他们如何通过“核心AI订阅服务”为医生提供支持:更像是一个智能代理,能够持续学习、个性化定制,并主动为用户提供所需的知识和支持。

  • 高度个性化:根据用户的偏好和历史数据,提供定制化的信息和建议。
  • 无缝集成:与现有的医疗系统和工作流无缝对接,确保信息流通无阻。
  • 持续进化:随着用户互动的增加,系统不断优化,变得更加智能高效。

商业模式与用户增长

目前,超过25%的美国执业医生每天都在使用Open Evidence作为辅助诊疗工具,平台每秒处理十多个真实的医疗问题,展现了强大的规模化能力。

尽管产品对所有医生免费开放,但公司通过医疗器械和药品广告实现收入。这种模式类似于消费互联网公司,但在医疗领域内形成了独特的用户粘性机制。

为了增强用户粘性和提升服务质量,Open Evidence 正在将全球顶级医生的临床智慧编码进系统(首先从胃肠病学开始),形成“集体智能”。这一策略不仅增强了平台的数据优势,还促进了未来的答案持续优化。

未来,Open Evidence计划整合更多的医学推理模式、科研能力和工作流嵌入功能,致力于打造一个全面的 AI 订阅服务平台,真正成为医生不可或缺的工作伙伴。

如果你是企业主,想把 AI 订阅模式融入现有业务,记住三个关键点:

  1. 从 “功能思维” 转向 “陪伴思维”:别想 “我能让 AI 做什么”,先想 “用户需要 AI 成为什么”(比如财务总监需要的不是报表工具,而是能预警风险的财务管家);
  2. 抢占 “场景入口”:选择一个高频、刚需的场景(如医疗、法律、垂直行业办公),把 AI 深度嵌入用户的工作流;
  3. 设计 “订阅式情感账户”:通过持续的价值交付(比如每周生成专属行业洞察),让用户产生 “不用就亏” 的依赖感。

万亿 AI 市场大洗牌:创业者与投资者当下必须知晓的事 KellyOnTech

以长期主义和构建产业协同生态为核心价值观的红杉资本近期举行了第三届人工智能峰会。此次峰会汇聚了 150 位全球顶尖 AI 领域创始人,释放了三大核心信号。

一、AI 商业模式从 “工具导向” 向 “结果导向” 转型

这一转型的底层逻辑是实现商业模式闭环,将技术转化为可量化的业务成果。过去十年,企业软件的核心价值是“提升效率”——通过SaaS工具实现流程自动化,客户为此支付“软件费用”。但AI正在穿透这一逻辑,形成全新的商业闭环:

Software as a Tool(工具)→ Software as a Co-worker(协作)→ Software as an Outcome(成果)

对用户而言,付费逻辑从 “为模型能力买单” 转变为 “为实际问题解决能力买单”。例如,法律 AI 工具已从单纯提供 API 服务,升级为直接输出法律文书,通过结果交付实现价值闭环。

二、AI “操作系统” 级入口的竞争


这场竞争的核心在于通过增强用户粘性构建平台壁垒,而粘性的根基在于两大能力:记忆能力(用户与AI的历史交互数据沉淀)与执行能力(AI调用工具的效率)。

记忆能力:从“工具”到“数字伙伴”的进化

长期记忆功能使AI能够深度学习用户的习惯、偏好及场景需求,从而实现千人千面的个性化服务。

记忆能力的关键价值让AI从“被动响应指令”升级为“主动理解需求”,形成类似人类助理的“数字伙伴”关系。例如,在医疗领域AI系统基于患者病史、基因数据和用药记录,动态调整诊疗方案,避免“一刀切”的推荐。

执行能力:从“能力展示”到“任务闭环”的跃迁

高效调用工具(如API接口、硬件设备)的能力,决定了AI能否在复杂场景中完成端到端的任务闭环。执行效率是企业采纳AI的关键门槛。例如,一个法律AI工具若能自动调用合同模板库、检索判例数据库并协调律师团队,其效率远超依赖人工操作的传统模式。

当AI同时具备深度记忆与高效执行,将形成不可逆的用户依赖, 从而构筑真正的护城河。

三、智能体经济的崛起:从工具到自主经济参与者的跃迁

人工智能正从被动工具进化为具备自主决策与协作能力的经济主体,推动”智能体经济”形成。但实现这一愿景需满足三大核心要素:

1. 持久身份:智能体的“数字人格”构建

每个智能体需拥有唯一且可验证的标识,以建立其在经济体系中的身份基础。比如,中国国家网络身份认证体系的“网号+网证”双轨制,为智能体身份认证提供政策和技术框架。

2. 行动能力:打通数字世界与物理世界

智能体需具备调用物理工具(如机器人、IoT 设备)和数字工具(API、数据库)的能力,以完成闭环任务。这里面的挑战在于需要解决工具权限管理、实时响应性及容错机制(如失败任务的自动重试或人工介入)。

3. 可信协作:跨智能体的信任基础设施

智能体间的协作需依赖可信机制,确保交互的透明性、可追溯性和责任划分。关键问题是如何定义智能体间的责任归属?比如 LOKA协议,这是卡内基梅隆大学提出的分层架构(身份层+沟通层+伦理层),为智能体提供去中心化标识与伦理决策框架。另外,如何平衡隐私保护与数据共享?

竞争优势的本质是 “数据 – 场景 – 生态” 的三位一体

根据红杉资本 AI 峰会提出的未来 AI 产业三大核心信号,AI 产业的发展趋势为投资机构和创业公司提供了关键发力点与评估维度:能否在构建收入闭环的同时,建立可持续的竞争壁垒。红杉资本指出,AI 企业的可持续壁垒最终取决于:

  1. 数据飞轮:用户交互数据反哺模型优化的闭环效率;
  2. 场景深耕:在垂直领域(如医疗、法律)中解决复杂问题的不可替代性;
  3. 生态协同:通过入口级产品或标准协议构建跨主体协作网络。

创业公司需围绕上述维度,判断自身能否在 AI 从 “技术验证” 迈向 “价值创造” 的关键十年中,占据生态位并实现长期价值。

根据红杉资本的预测,到2026年,垂直领域智能体与混合治理框架将催生万亿级市场,而未能实现“工具→协作→成果”转型的企业将被淘汰。

AI Market Shake-Up: What Founders and Investors Must Know Now

With a core philosophy rooted in long-termism and building an integrated industrial ecosystem, Sequoia Capital recently hosted its third annual AI Summit. The event brought together 150 of the world’s leading AI founders and conveyed three powerful signals.

Three Core Signals for the AI Industry in the Next Decade

1. AI Business Models: The Shift From “Tool-Oriented” to “Outcome-Oriented”

This transformation reflects a deeper move toward closing the loop in AI-driven business models, where technology translates into measurable business outcomes.

Over the past decade, enterprise software has delivered value primarily by improving operational efficiency. Companies paid for SaaS tools that automated workflows — essentially, “software as a tool.” But AI is now breaking through that model and shaping a new paradigm:

Software as a Tool Software as a Co-worker Software as an Outcome

In this new framework, users are no longer paying for the AI model’s capabilities alone — they’re paying for its ability to solve real-world problems and deliver results.

Take a legal AI startup as an example. It has evolved from offering simple API-based services to delivering complete legal documents, shifting its value proposition from tool to outcome — and in doing so, creating a self-contained value loop.

2. Competition for AI “Operating System” — Level Entry Points

At the core of this race is the strategic goal of building platform-level moats through enhanced user stickiness. This stickiness is fundamentally driven by two key capabilities: memory (the accumulation of historical user-AI interactions) and execution (the AI’s ability to efficiently orchestrate tools and complete tasks).

Memory: Evolving from a “Tool” to a “Digital Companion”

Long-term memory enables AI systems to deeply learn a user’s habits, preferences, and contextual needs, unlocking truly personalized, one-to-one experiences at scale.

This memory capability shifts AI from being a passive command responder to an active need interpreter, evolving into a digital companion similar to a human assistant. For example, in healthcare, an AI system that incorporates a patient’s medical history, genetic profile, and medication records can dynamically adjust treatment recommendations — moving beyond one-size-fits-all suggestions.

Execution: Evolving from “Capability Showcase” to “Task Completion Loop”

An AI’s ability to efficiently call external tools (e.g., APIs, databases, hardware devices) determines whether it can execute end-to-end workflows in complex real-world scenarios. Execution efficiency is a critical adoption barrier for enterprise AI.

For instance, a legal AI tool that can autonomously pull from contract templates, search relevant case law, and coordinate with a legal team far outperforms traditional methods reliant on manual operations.

When deep memory and high execution efficiency come together, they form an irreversible user dependency, building a powerful moat for the AI platform. This is the foundation for the next generation of AI-native operating systems.

3. The Rise of the Agent Economy: From Tools to Autonomous Economic Actors

Artificial intelligence is rapidly evolving from a passive tool into an autonomous economic participant, giving rise to the “agent economy.” To realize this vision, three foundational pillars must be established:

1). Persistent Identity: Building the “Digital Persona” of Agents

Each agent must possess a unique and verifiable identity to establish its legitimacy within an economic system. For example, China’s dual-track digital identity framework — “Net ID + Net Credential” — offers a policy and technical foundation for authenticating AI agents.

2). Action Capability: Bridging the Digital and Physical Worlds

Agents must be able to orchestrate both physical tools (such as robots and IoT devices) and digital resources (like APIs and databases) to complete closed-loop tasks. Key challenges include permission management, real-time responsiveness, and robust fault tolerance (e.g., automatic retries or fallback to human intervention in case of failure).

3). Trustworthy Collaboration: A Trust Infrastructure for Inter-Agent Cooperation

Collaboration between agents requires a trustworthy infrastructure that ensures transparency, traceability, and accountability. A central question: How do we assign responsibility among autonomous agents? The LOKA Protocol, developed by Carnegie Mellon University, proposes a layered framework (Identity Layer + Communication Layer + Ethics Layer) that supports decentralized identity and ethical decision-making for agents.

Additionally, how can we balance data privacy and information sharing in a world of interacting agents? This tension will define the future landscape of agent-based collaboration.

The Essence of Competitive Advantage: A Trinity of Data – Use Case – Ecosystem

In summary, the evolution of the AI industry presents both strategic inflection points and evaluation criteria for investors and startup founders: the ability to build sustainable competitive moats while closing the revenue loop.

According to Sequoia Capital, long-term defensibility in AI companies hinges on three pillars:

  1. Data Flywheel:
    The efficiency of feedback loops where user interaction data continuously improves model performance.
  2. Scenario Depth:
    Irreplaceable value in solving domain-specific, complex problems—especially in verticals like healthcare and law.
  3. Ecosystem Synergy:
    Building a collaborative network across stakeholders through platform-level products or standardized protocols.

Sequoia predicts that by 2026, vertical AI agents and hybrid governance frameworks will drive the emergence of trillion-dollar markets. Conversely, companies that fail to evolve from tool → collaborator → outcome will be left behind.