1886 年,奔驰造出的第一辆汽车「奔驰一号」,本质上是一辆「无马的马车」。
发动机挂在马车底盘上,转向沿用舵柄,因为没人知道「汽车」应该长什么样。十余年后,人类才回过神来:汽车不该对马车照猫画虎,而应该完全另起炉灶。于是,「梅赛德斯 35 HP」应运而生。
毕竟,固守马车时代的外观理念,是造不出真正面向将来的汽车的。

今年 WAIC 期间,类似的一幕在智能体手机上重演。
今年很多手机所做的「AI 化」,无非是外挂 AI 助手、调用云端服务。这两者本质上都是「无马的马车」,就像是在旧底盘上挂一台新引擎;而极少数玩家挑选了「AI 原生」,一如梅赛德斯 35HP 那样,尝试从模型、平台到硬件重新外观,另辟蹊径从头来,以模型作为第一性原理,造一台为了 AI 而生的新手机。
当时在 WAIC 上亮相的 STEPX Neo,挑选的正是第三条路。

三个月后的今天,阶跃终端正式发布首款大模型原生智能体手机 STEPX Neo,以及住在它里面的个人智能体 Amoo,将「模型、软件、硬件」三位一体的工艺主张落到了实际感受上。
提前拿到真机、深度使用一段时间后,我发现它不是一台面面俱到的「水桶机」,也不是急于证明自己完美成熟的流水线产物。它兑现了「AI 原生」的承诺,跳出了传统赛道,开辟了一条以「使用者创造力」为核心的 AI 原生智能体手机探索路线。
智能体手机的野心,长在复古将来主义里
告别大众旗舰千篇一律的玻璃曲面,STEPX Neo 的背板刻意弱化了传统多摄的夸张凸起,顶部使用跑道型横向双摄模组,平整克制,整机重心居中、不坠手。
灰黑基底上点缀亮橙色按键与像素 Logo,整台机器展现出一种近似 Teenage Engineering 周边机型的独特科技玩味,喜欢的人会一眼认出这是同类。
背部还有一块兼顾状态反馈与个性化展示的背屏。

使用者可以 DIY,并把自己喜欢的文案放到手机背面,让它成为日常表达的一部分。

尤其当 STEPX Neo 静止或倒扣在桌面时,除了来电或消息,如果 Amoo 正在后台执行多步骤长任务,背部会泛起细腻的动效,可以让「AI 正在替你工作」变得更可感知,而不是只能打开 App 查看进度条。

区别于传统音量键和电源键,机身侧边还设立了一颗 AI 实体键,可以随时呼出 Amoo,总体的盲操过程也极为顺畅:短按呼出平台级意图流面板与多模态常驻感知,长按直接进入端侧语音实时双工交互。

当然,鲜明的外观也带来了取舍。
STEPX Neo 握在手里仍带棱角感,而总体的外观则以「复古将来主义」呼应 Amoo「住在手机里」的设定:背板上下分区,摄像头代表视觉感知,膜片纹理借推理与记忆的意象,电路纹理将二者相连。
背面代码则从 stepfun 到 user_potential,再到 human_creativity + code 的幂运算,层层指向工艺迭代的底座永远是人的想象力的愿景。

每次写到智能体手机时,总有人认为,评价智能体手机,应该先看它的显示屏、电池续航、表现等基本感受能否达到日常使用的要求,再看 AI 做得怎么样。这个判断是对的,手机每天都要拿在手里使用,基础感受如果显著拖后腿,再强的 AI 也很难弥补。
但基础感受过关之后,决定人们为什么要买一台「智能体手机」的,终究还是 AI。专门挑选这类机型的人,多半期待它能完成普通手机做不到、或者做起来很费劲的事。
了解当下整个消费电子行业的大背景,有助于笔者理解 STEPX Neo 的诸多全新定义。而苹果正是一个关键参照。
在今年的苹果秋季发布会上,新任 CEO John Ternus 将 iPhone 定义为「智能个人中枢」:依托本地算力和隐私上下文,联动 Apple Watch、AirPods 等硬件,成为个人数字设备的核心调度枢纽。

设备中枢是一条稳健的路线,但使用者取得什么能力,主要仍由厂商外观的平台能力和设备协作规则决定。
STEPX Neo 想进一步探索的是另一个难题,使用者能否参与塑造手机理解自己、处理任务的方式,让个人智能体随着使用,逐步形成只属于这个人的能力。
这也是阶跃终端对 Step AOS 的定位:「模型原生」。
阶跃终端是当下市面上唯一一家来自基础模型公司背景的手机公司,「模型原生」是它对 AI 手机的天然认知:要做真正的个人智能体,就一定需要硬件容器,让智能体可以伴随使用者进入物理世界中千千万万的情形。
所以它率先提出了「软硬模三位一体」:
模型负责理解与决策(Step 1+N 模型全家桶),软件负责调度与执行(Step AOS),硬件负责感知与承载(AI 原生终端),而三者的中心,个人智能体必须具备自己的「身体」——智能体手机。
要贯彻「模型原生」,智能体必须跨过三堵墙:记忆墙、决策墙和行动墙。记忆墙来自分散在各处的个人上下文,决策墙涉及一句需要背后的步骤规划,行动墙则横在计划与实际操作之间。

界面也随之发生改变。STEPX Neo 弱化了让人逐个寻找 App 的传统图标逻辑,把负一屏的「开展中任务」和「上下文意图」放到更显眼的位置。
接下来,笔者就用真实操作看看,它具体能做些什么?和普通手机又有什么区别?
记住偏好,是一切个性化智能的起点
一个人的日程、照片、文档和使用习惯散落在手机各处。智能体如果只能听见眼前的一句指令,就很难知道使用者说的「那张照片」是哪张,安排出行时也得反复询问时间与偏好。
Step AOS 为此外观了统一语义数据层和「双域六层三步记忆」:
前者负责整理与任务有关的信息,后者分别积累使用者偏好和 Amoo 的任务经验,让已经给出过的上下文,有机会在下一次任务中继续发挥作用。

Amoo 也有自己的记忆平台,启动记忆共享后,它会保留交流偏好、备忘事项等信息,随着使用逐步了解使用者,让后续沟通和处理任务更顺手。

例如,在一次闲聊中,它就能清楚记得我爱喝蜜雪冰城的棒打鲜橙。

另外,在实际感受中,Amoo 还可以搜索、查看和整理相册照片,管理联系人与日历日程。
如果说上述表现跨越了长期记忆的墙,那么更容易当场验证的即时上下文连贯,则体现在打破「感知墙」的「圈屏提问」上:按下 AI 实体键呼出语音输入后,可以直接在显示屏内容上圈画,再问 Amoo 圈中的东西是什么。
刷到一张小狗的照片,按下 AI 实体键,在显示屏上圈住它,直接问:这种狗好养吗?Amoo 认出犬种,接着从性格、运动量、掉毛程度给出饲养建议。

虽然圈选这个动作本身也不是新发明,三星 Galaxy S24 的「即圈即搜」两年前就做了。但「即圈即搜」的终点是搜索结果页,圈完一次就结束;STEPX Neo 的圈屏提问,终点是多模态问答和后续任务,圈中的内容直接进入 Amoo 的上下文。
期间,使用者不必费力描述对象的外观、位置和所在页面,圈出的内容本身就成为提问的线索。
不止于「答」,更在于「谋」
得到上下文之后,Amoo 还要判断怎样处理任务。设闹钟、查照片可以迅速响应;订票、处理文件或安排周期任务,则可能涉及查找信息、比较选项和规划多个步骤。
Step AOS 使用端云协同:即时操作的简单任务交给端侧模型 Step Edge,复杂推理与多步骤规划交给云端模型。在基础手机操作中,Amoo 能打开或关闭各种 App,设置闹钟,查看天气与位置,拨打电话、发送短信,管理联系人和日历日程。

需要外部信息时,它可以搜索新闻、资料与实时信息,借助东方财富查看股票和基金行情,借助什么值得买查询商品售价与建议,定时提醒和周期任务也在当下兼容的范围内。
最能说明难题的是一次订票。我给出的需要是:帮我订明早从广州飞北京的机票,选国航的;之类,如果明早北京大雨,就直接改订高铁商务座;对了,总体预算不能超过 1500 块。

一句话里埋了三个考点:明确偏好、条件分支、预算上限。Amoo 把它拆成几步,先查航班与售价,再查北京明早的天气,依据天气结果在机票和高铁商务座之间做取舍,全程把预算卡在 1500 元以内,最后停在付款前等我确认。
从我的感受看,简单任务它判断果断,多步骤任务中它倾向于在关键节点停下来确认,这种「克制的聪明」反而是合理的。
我还试了一个周期任务,创建一个定时任务,每天早上 9 点推送新闻简报,精选过去 24 小时最值得关注的 5 条新闻,每条附简短摘要和来源链接。创建之后,第二天早晨简报准点浮现。

这本质上已经不是传统意义上的「设置提醒」,而是使用者借助自然语言给自己「编程」,创造出了一个专属能力。这种低门槛的自定义感受,正是智能体手机走向「使用者创造力」的第一步。
从行动交付到共同创造,接住 Builder 的想象力
智能体即便想好了步骤,也可能被应用之间的边界拦住。打车、订票、处理文档各有自己的入口、数据和权限,知道「应该怎么做」并不等于可以替使用者完成。
相比于普通手机的 AI 助手大多困在对话框里,能「说」不能「做」,Step AOS 的原子能力引擎尝试把这些能力组织成可调用的服务,让 Amoo 以平台级身份安排操作。
最常见的办公情形就涉及 Word、Excel、PPT、PDF 的创建与编辑,飞书文档、表格、日历、任务和审批的读写,以及金山文档处理。
我让它新建一个 Word 文档,写一份关于 2026 年折叠屏手机市面趋势的调研大纲,写完直接导出成 PDF。一句话交代了格式、主题和交付形态,Amoo 依次完成大纲撰写、文档生成和 PDF 导出。

生活服务也可以看作一个完整闭环,我让它帮我在美团点一杯奶茶,它没有让我输入地址,而是先问我:送到住处,还是公司。它知道我有两个常用地址,在涉及付款、删除数据等敏感步骤时仍需使用者确认。
打车时,我也试过让 Amoo 从爱范儿公司地址叫一辆快车,目的地是广州白云机场 T2 航站楼。

在生态合作维度,阶跃 Amoo 已与支付宝、腾讯、携程、滴滴、剪映、高德、美团等近百家企业及机型建立合作关系,将服务能力延伸至本地生活、旅游出行、办公协作、内容创作等多个情形,推动不同平台与服务之间的能力互通。

除调用现有服务外,行动力的边界不仅在于跨应用执行,更在于无中生有的创造。
当现有应用无法满足情形需要时,Amoo 还兼容生成图片、制作 HTML 页面或小工具,手机的行动能力也就从「替使用者点击按钮」,自然延伸到了「按需生成工具」。
Skill 给出了进一步扩展的入口,让一些特定流程有机会成为手机上可重复调用的能力:当下可见的 Skill 涵盖飞书 CLI、企业微信 CLI、Skill 安全助手,以及英语学习、网页抓取、PDF 转 Markdown、Word 转化等不同用途;

例如 APPSO 此前报道过的微信读书 Skill,如今只需在手机上完成简单规格,就能接入使用,更方便地获取阅读相关信息。

开发者也可以在 AmooStore 上传已有 Skill 包,经过格式与安全检查后提交审核。这是「使用者参与塑造手机」那句话最具体的落点。

对照普通手机看,这一步的意义更清楚:传统手机的能力边界由 App Store 和厂商划定,使用者只能在货架上「选」,不能自己「造」。Skill 是「使用者参与塑造手机」那句话最具体的落点。

Amoo Store 里还可以借助智能体能力连接远程智能体。当下兼容远程连接的工具涵盖阶跃 AI、Claude Code、Codex、豆包、DeepSeek 和 WorkBuddy 等,让手机也成为调用这些远程智能体的入口。
对于希望更从容开发能力的使用者,阶跃终端还带来了 Step Studio,并将它定位为首款 AI 终端原生开发工具。
使用者可以借助对话描述需要,构建自己的 App、Skill 或 Agent,再推送到手机上测试和使用。
我输入「构建一个全自动科技深度报道主编智能体」,大约两分钟后,Step Studio 就生成了相应 Agent,将选题判断、素材研究、交叉验证和撰写成稿写入工作流程,并加入注明来源、区分事实与分析等编辑要求,供我继续调整。
如果希望智能体调用手机能力,还可以按需添加连接器。界面中可选的能力涵盖文件管理、定位、图库、日历、信息和电话等,前面提到的原子能力,由此成为开发专属能力时可以组合使用的资源。
完成构建后,点击「真机测试」,就能将智能体推送到 Amoo 空间的「我的作品」中,在手机上测试和修改,再借助「发布并上架」供更多使用者安装使用。
从让 Amoo 写一次大纲,到为自己创建一个遵循固定编辑流程的 Agent,使用者可以定制的内容,也从单次任务的结果延伸到了处理任务的方法。Step Studio 为这种定制给出了从构建、修改到手机端测试和发布的流程,也让按需生成工具有了更完整的开发入口。
从使用能力到创造能力,模型、平台和开发工具也成为手机价值的一部分。
基于这一判断,STEPX Neo 先行版推出「软硬模」全栈定价,将 Step 系列模型、Step AOS、Step Studio 与手机硬件整合定价,全栈价为 3999 元,给出碳黑、硅银两种配色。
购买使用者可享连续 12 个月、每月百亿 Token 额度的 STEPX Token Plan,兼容原生智能体应用的开发与使用。
「阶跃 Amoo 成长计划」还将为前 1 万名任务达标使用者给出「星辰船票」,将来五年内可凭 STEPX Neo,以 0 元置换任意一款 STEPX 新款终端,具体权益以商品页面为准。
上市后,STEPX Neo 将持续更新平台、拓展任务情形,并借助 Step Studio 吸引更多开发者和创客参与生态共建,让手机能力随着使用者的使用与创造不断多元。
把「塑造手机」的权利,第一次交到笔者手上
当大多数厂商挑选在 Android 外壳上套一个 AI 助手时,STEPX Neo 从硬件外观、底层平台 Step AOS、端云协同、到开放生态全部重做了一遍。
但「重做」本身并不稀奇,真正值得分辨的是:这台机器是追热点的产物,还是一条早就想清楚的路线的必然结果。
今年 9 月,各家厂商开始陆续押注 Personal Agent;而早在 7 月,阶跃终端带来个人智能体 Amoo 时,就已经讲出了逻辑一致的故事——只是将它放在了一台更切实有感的设备当中。
更关键的是,阶跃终端想清楚的不只是概念,还有概念的物理前提:个人智能体必须具备自己的「身体」,才能伴随使用者进入物理世界中千千万万的情形。
先有路线,再有架构,最后才有 STEPX Neo 这台真机。这个先后顺序也意味着,当行业刚开始讨论 Personal Agent 时,阶跃终端已经在回答下一个难题了——当智能体有了身体,手机竞争的靶心该放在哪里?
STEPX Neo 给出的回答是从「卷镜头」转向「卷创造力」。
2000 年,夏普 J-SH04 第一次把摄像头塞进手机,11 万像素,拍出来的照片糊成色块,被嘲笑为「给玩具装镜头」。谁也想不到,这个被嘲笑的开局,拉开了智能手机史上最漫长的一场军备竞赛:
此后二十多年,从诺基亚 808 PureView 的 4100 万像素传感器,到不断发展的计算摄影、潜望长焦和一英寸大底,厂商们在方寸之间的镜头上缠斗了整整二十年。
为什么偏偏是影像?
因为在移动互联网时代,「拍照记录」是普通人表达创造力门槛最低的出口。按下一张好照片、剪一段生活视频,本质都是人对「创造」与自我表达的本能追求。
影像时代的手机让笔者成为世界的「观察者」和「记录者」;而当大模型的意图理解与行动能力恰好在这个节点成熟,创造欲开始寻找新的出口:
人们不再满足于拍一张好看的照片,而是希望手中的终端能替自己应对复杂难题、编排专属逻辑,甚至像搭积木一样定制只属于自己的数字能力。应运而生的 STEPX Neo,也就第一次以开放模型、平台与 Skill 生态,把「塑造手机」的权利交到了使用者手里。
至于那些尚未磨平的生态断点、仍需优化的发热与电池续航,也都是开拓者留在路上的路标。但就像 1886 年那辆笨拙驶过街头、沿用舵柄却不再依靠马匹的先行者一样——
它的价值从不在于第一天就跑赢马车,而在于它终于摆脱了马车的惯性。
本文来源:https://www.ifanr.com/1684008
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。
Mercedes-Benz One, the first car built by Mercedes-Benz in 1886, was essentially a carriage without horses.
The engine hangs on the chassis of the carriage and the steering follows the tiller, because no one knows what the “car” should look like. More than ten years later, human beings returned to their senses: cars should not paint cats and tigers on carriages, but should start a completely new stove. The result was the Mercedes 35 HP.
After all, sticking to the exterior concept of the carriage era is not really a car for the future.

A similar scenario was repeated on smartphones during this year's WAIC.
The "AIization" done by many mobile phones this year is nothing more than plugging in AI assistants and calling cloud services. Both are essentially "carriages without horses", like hanging a new engine on an old chassis; and a very small number of players have chosen "AI native", like Mercedes 35HP, to try to reappear from model, platform to hardware, and create a new phone for AI from scratch with the model as the first principle.

The STEPX Neo, which was unveiled on WAIC at the time, was the third way to go.
Today, three months later, Step Terminal officially released the first large model native smartphone, STEPX Neo, and the personal agent Amoo living in it, bringing the "model, software, hardware" trinity process proposition to the actual feeling.
After getting the real machine in advance and using it in depth for a period of time, I found that it was not a "bucket machine" in all aspects, nor was it a product of the assembly line eager to prove its perfect maturity. It delivers on the promise of “AI-native”, jumping off the traditional track and opening up a route to explore AI-native smartphones with “user creativity” at their core.

Saying goodbye to VW's flagship uniform glass curved surface, the back plate of the STEPX Neo deliberately weakens the traditional multi-shooting exaggerated bulge. The top uses a runway-type horizontal double-camera module, which is flat and restrained. The center of gravity of the whole machine is centered and does not fall.
Dotted with bright orange buttons and pixel logos on a gray-black substrate, the entire machine exhibits a unique technological play similar to Teenage Engineering's surrounding models, and people who like it will recognize it at a glance.

There is also a back screen on the back that combines status feedback with a personalized display.
Users can DIY and put their favorite copy on the back of their phone to make it part of their daily expression.
Especially when STEPX Neo is stationary or inverted on the desktop, in addition to incoming calls or messages, if Amoo is performing multi-step and long tasks in the background, the back will be delicate, which can make "AI is working for you" more perceptible, rather than just opening the app to view the progress bar.

Different from the traditional volume key and power button, there is also an AI physical key on the side of the fuselage, which can exhale Amoo at any time. The overall blind exercise process is also extremely smooth: short press the outgoing platform-level intention flow panel and multi-modal permanent perception, and long press the voice to directly enter the end-side real-time duplex interaction.
Of course, the distinctive appearance also brings trade-offs.

STEPX Neo is still angular in the hand, while the overall appearance echoes Amoo's "living in the phone" setting with "retro-futurism": the upper and lower partitions of the backplane, the camera represents visual perception, the diaphragm texture connects the two with the image of reasoning and memory, and the circuit texture connects the two.
The back code ranges from stepfun to user_potential to the power operation of human_creativity + code, and the layers point to the base of process iteration, which is always the vision of human imagination.
Every time I write about smartphones, some people always think that when evaluating smartphones, I should first look at whether its display, battery life, performance and other basic feelings can meet the requirements of daily use, and then look at how AI is doing. This judgment is correct. The mobile phone should be used in hand every day. If the basic feeling is significantly delayed, no matter how strong the AI is, it will be difficult to make up for it.

But after passing the basic feelings, it is still AI that decides why people want to buy a "smart phone". Most people who choose this type of model expect it to do things that ordinary mobile phones can't do or are hard to do.
Understanding the current background of the entire consumer electronics industry helps me understand the many new definitions of STEPX Neo. And Apple is a key reference.

At this year's Apple fall conference, new CEO John Ternus defined the iPhone as a "smart personal hub": relying on local computing power and privacy context, linking Apple Watch, AirPods and other hardware to become the core scheduling hub for personal digital devices.
The equipment hub is a robust route, but what capabilities users achieve is still mainly determined by the vendor's platform capabilities and equipment collaboration rules.
What STEPX Neo wants to further explore is another puzzle, whether the user can participate in shaping the way the phone understands itself and handles tasks, so that the individual agent gradually develops the ability to belong only to this person as it is used.

This is also the positioning of the step terminal to Step AOS: "model native".
Step terminal is the only mobile phone company on the market today that comes from the background of a basic model company. "Model native" is its natural understanding of AI mobile phones: to be a real personal agent, it must need hardware containers, so that the agent can accompany users into thousands of situations in the physical world.

So it pioneered the "soft and hard mold trinity":
The model is responsible for understanding and decision-making (Step 1 + N model family bucket), the software is responsible for scheduling and execution (Step AOS), and the hardware is responsible for sensing and carrying (AI native terminal). At the center of all three, the personal agent must have its own "body" - smart phone.

To implement Model Native, the agent must cross three walls: the memory wall, the decision wall, and the action wall. The memory wall comes from scattered personal contexts, the decision wall involves the planning of the steps behind a sentence, and the action wall is between the plan and the actual operation.
The interface also changes. Stepx Neo weakens the traditional icon logic that allows people to look for apps one by one, and puts negative one-screen "active tasks" and "contextual intent" in a more prominent position.
Next, I will use the real operation to see what it can do? What's the difference with a regular phone?

A person's schedule, photos, documents, and usage habits are scattered throughout the phone. If the agent can only hear one command in front of it, it is difficult to know which "photo" the user said, and it is necessary to repeatedly ask the time and preferences when arranging travel.
Step AOS has a unified semantic data layer and "double-domain six-layer three-step memory" for this appearance:

The former is responsible for collating information related to the task, while the latter accumulates user preferences and Amoo's task experience respectively, giving the context that has been given the opportunity to continue to play a role in the next task.
Amoo also has its own memory platform. After memory sharing is started, it will retain information such as communication preferences, notes, etc. As you use it, you will gradually understand the user, making it easier to communicate and handle tasks.
For example, in a small chat, it can clearly remember that I like to drink the sticks of the ice city of honey and beat fresh oranges.

And in real life, Amoo searches, views, and organizes album photos to manage contacts and calendar schedules.
If the above performance spans the wall of long-term memory, then the instant context coherence that is easier to verify on the spot is reflected in the "circle screen question" that breaks the "perception wall": after pressing the AI physical key to exhale the voice input, you can directly circle the display content and ask what is in the Amoo circle.

Brush on a picture of a puppy, press the AI entity key, trap it on the display, and ask directly: Is this dog good to keep? Amoo recognizes the breed of dog and then gives breeding advice based on personality, amount of exercise, and degree of hair loss.
Although the circle selection action itself is not a new invention, the Samsung Galaxy S24's "circle search" was done two years ago. However, the endpoint of "Instant Search" is the search results page, which ends after one round; the endpoint of STEPX Neo's circle screen asks questions, and the endpoint is multi-modal Q&A and follow-up tasks. The contents of the circle directly enter the context of Amoo.
During this period, the user does not have to describe the appearance, location and page of the object, and the circled content itself becomes a clue to ask the question.

Once the context is obtained, Amoo also decides how to handle the task. Setting an alarm clock, checking photos, and booking tickets, processing documents, or scheduling recurring tasks can involve finding information, comparing options, and planning multiple steps.
Step AOS uses end-to-end cloud collaboration: the simple task of immediate operation is handed over to the end-side model Step Edge, and the complex reasoning and multi-step planning are handed over to the cloud model. Amoo opens and closes apps, sets alarms, checks weather and location, makes phone calls, sends texts, and manages contacts and calendar schedules in basic mobile operations.

When external information is needed, it can search for news, materials, and real-time information. With the help of Oriental Wealth, it can check stock and fund quotes. With the help of what is worth buying, it can inquire about commodity prices and suggestions. Timing reminders and periodic tasks are also compatible with the current scope.
The hardest part is booking a ticket at once. My needs are: help me book a flight from Guangzhou to Beijing in the morning and choose Air China; etc. If it rains heavily in Beijing tomorrow morning, I will directly rebook the high-speed rail business seat; by the way, the overall budget can not exceed 1500 yuan.
There are three test points buried in one sentence: clear preferences, conditional branches, and budget ceilings. Amoo breaks it down into a few steps, first checks the flight and price, then checks the weather in Beijing tomorrow morning. According to the weather results, it makes a trade-off between the ticket and the high-speed rail business seat. The budget is stuck within 1500 yuan for the whole process, and finally stops and waits for me to confirm before paying.

Judging from my feelings, it is decisive in simple tasks. In multi-step tasks, it tends to stop at key nodes to confirm that this "restrained cleverness" is reasonable.
I also tried a cycle task, creating a timed task, pushing a newsletter every day at 9am, and featured 5 of the most noteworthy news from the past 24 hours, each with a short summary and source link. After creation, the next morning the briefing appears on time.

In essence, this is no longer a "set reminder" in the traditional sense, but a user creates an exclusive ability by "programming" himself with natural language. This low threshold of customization is the first step towards “user creativity” for smartphones.
Even if the agent thinks of steps, it may be blocked by the boundary between applications. Taxi booking, ticket booking, and processing documents each have their own entrance, data, and permissions. Knowing "what to do" does not mean that it can be completed for the user.

Compared to ordinary mobile phones, AI assistants are mostly trapped in dialog boxes and can "say" and "do". Step AOS's Atomic Capabilities Engine attempts to organize these capabilities into callable services, allowing Amoo to arrange operations as a platform-level identity.
The most common office situations involve the creation and editing of Word, Excel, PPT, and PDF, the reading and writing of flying book documents, forms, calendars, tasks, and approvals, and the processing of San Francisco documents.
I asked it to create a new Word document, write a research outline on the market trends of foldable phones in 2026, and export it directly to PDF. In one sentence, describing the format, subject, and delivery form, Amoo completes the outline, document generation, and PDF export in turn.

Life service can also be seen as a complete closed loop, I asked it to help me order a cup of milk tea in Meituan, it did not ask me to enter the address, but first asked me: delivered to the place of residence, or the company. It knows I have two primary addresses and still requires user confirmation when it comes to sensitive steps like payment, deleting data, etc.
When I took a taxi, I also tried to get Amoo to call a express from the address of Aifan 'er, and the destination was the T2 terminal of Guangzhou Baiyun Airport.

In the dimension of ecological cooperation, Step Amoo has established cooperative relationships with nearly 100 companies and models such as Alipay, Tencent, Ctrip, Didi, Shearing, Gaode, Meituan, etc., extending service capabilities to local life, travel, office collaboration, content creation and other situations, and promoting the interoperability of capabilities between different platforms and services.
In addition to invoking existing services, the boundary of mobility lies not only in cross-application execution, but also in the creation of nothing.
When the existing application cannot meet the needs of the situation, Amoo is also compatible with generating images, making HTML pages or widgets, and the mobile ability of the mobile phone extends naturally from "clicking buttons for users" to "on-demand generation tools".

Skill gives a further expanded entrance, giving some specific processes the opportunity to become repeatable on mobile phones: currently visible Skill covers Feishu CLI, Enterprise WeChat CLI, Skill Security Assistant, as well as English learning, web scraping, PDF to Markdown, Word conversion and other different purposes;
For example, WeChat Reading Skill, previously reported by APPSO, can now be accessed and used simply by completing simple specifications on a mobile phone, making it easier to access and read relevant information.

Developers can also upload existing Skill packages to AmooStore for format and security checks before submitting for review. This is the most concrete landing point of the phrase “user participation in shaping the mobile phone”.
Compared with ordinary mobile phones, the meaning of this step is clearer: the boundaries of the ability of traditional mobile phones are defined by the App Store and manufacturers, and users can only "select" on the shelves, not "build" themselves. Skill is the most specific landing point for the phrase “user engagement shapes the phone.”
Remote agents can also be connected in the Amoo Store with the help of agent capabilities. Today's compatible remote connectivity tools include Step AI, Claude Code, Codex, Beanbag, DeepSeek, and WorkBuddy, making mobile phones the gateway to these remote agents.

For users who want to develop more leisurely, Step Terminal also brings Step Studio and positions it as the first AI terminal native development tool.
Users can use conversations to describe their needs, build their own apps, Skills, or Agents, and then push them to their phones for testing and use.

I entered "Build a fully automated in-depth science and technology report editor-in-chief agent". About two minutes later, Step Studio generated the corresponding agent, judging the topic, researching the material, cross-validating and writing the manuscript into the workflow, and adding editorial requirements such as annotating the source, distinguishing the facts and analyzing, etc., for me to continue to adjust.
If you want the agent to call the phone's capabilities, you can also add connectors as needed. The optional capabilities in the interface cover file management, positioning, gallery, calendar, information and telephone, etc. The atomic capabilities mentioned above become resources that can be used in combination when developing exclusive capabilities.
After completing the build, click "Real Machine Test" to push the agent to "My Work" in the Amoo space, test and modify it on your phone, and then use "Publish and List" for more users to install and use.

From having Amoo write an outline once, to creating an agent for themselves that follows a fixed editing process, users can customize content that extends from the results of a single task to the way the task is handled. Step Studio gives this customization a flow from build, modification to mobile testing and release, and a more complete development portal to on-demand generation tools.
From usability to creativity, models, platforms, and development tools are also part of the value of mobile phones.

Based on this judgment, STEPX Neo Advanced Edition introduced "soft and hard mold" full-stack pricing, integrating the Step series model, Step AOS, Step Studio and mobile phone hardware pricing, with a full stack price of 3,999 yuan, giving two color schemes of carbon black and silver.
Purchase users can enjoy the STEPX Token Plan with a monthly quota of 10 billion tokens for 12 consecutive months, which is compatible with the development and use of native agent applications.

The "Step Amoo Growth Plan" will also give the first 10,000 mission-compliant users the "Starship Ticket". In the next five years, any new STEPX terminal can be replaced with STEPX Neo for 0 yuan, and the specific rights and interests will be subject to the product page.
After the launch, STEPX Neo will continue to update the platform, expand the task situation, and attract more developers and makers to participate in ecological co-construction with Step Studio, so that mobile phone capabilities will continue to diversify with the use and creation of users.
When most vendors choose to set up an AI assistant on the Android shell, STEPX Neo redo everything from the hardware appearance, the underlying platform Step AOS, end-cloud collaboration, to the open ecosystem.

But "redoing" itself is not unusual. What is really worth distinguishing is whether this machine is the product of chasing hot spots or the inevitable result of a long-thought-out route.
In September this year, various manufacturers began to bet on Personal Agent; and as early as July, when Step Terminal brought the personal agent Amoo, it already told a logical story - just put it in a more tangible and sensible device.

More importantly, the step terminal thinks clearly not only about the concept, but also the physical premise of the concept: the individual agent must have its own "body" to accompany the user into the physical world.
First there is a route, then there is a structure, and finally there is the STEPX Neo real machine. This sequence also means that when the industry began to discuss Personal Agent, the step terminal was already answering the next puzzle - when the agent has a body, where should the target of the mobile phone competition be placed?
The answer given by STEPX Neo is to move from “roll lens” to “roll creativity”.

In 2000, Sharp J-SH04 tucked the camera into the mobile phone for the first time, 110,000 pixels, and the photos taken were glued into a color block, which was ridiculed as "putting a lens on the toy". Unexpectedly, this ridiculed start kicked off the longest arms race in the history of smartphones:
From the Nokia 808 PureView's 41-megapixel sensor to evolving computational photography, periscope telephoto, and one-inch outsoles, manufacturers have wrestled with lenses between square inches for more than two decades.

Why images?
Because in the era of mobile Internet, "photo recording" is the lowest threshold for ordinary people to express their creativity. Pressing a good photo and cutting a life video are essentially people's instinctive pursuit of "creation" and self-expression.
The mobile phone of the image age makes me the "observer" and "recorder" of the world; and when the intention of the big model and the ability to act are mature at this point, the creation wants to start looking for a new exit:

People are no longer satisfied with taking a good-looking photo, but hope that the terminal in their hands can cope with complex problems, arrange their own logic, and even customize their own digital capabilities like building blocks. For the first time, STEPX Neo, which was born in response, handed the right to "shape mobile phones" to users with an open model, platform and Skill ecology.
As for those ecological breakpoints that have not yet been flattened, heat generation and battery life that still need to be optimized, they are also signs for pioneers to stay on the road. But just like the pioneer who in 1886 drove clumsily through the streets and used the tiller but no longer relied on horses -

Its value is never in outperforming the carriage on the first day, but in finally getting rid of the carriage's inertia.
This article is from: https://www.ifanr.com/1684008 For copyright issues, please write to: 13951431913#139.com or submit a review on the right.