新智元报道
妈妈还有10分钟到家，女孩却找不到另一只高跟鞋了。
「我另一只高跟鞋呢？」
墙上的屏幕开始回放客厅画面。AI提示：鞋可能被踢到了沙发底下。她随即俯身去找。
女孩没有交代鞋的颜色、款式，也没有回忆最后一次在哪儿见过。WallEX把她的话和客厅里的画面联系起来，给了她一条线索。
找鞋是件小事，但要AI办成，得同时听懂人话、看懂画面，还得把两件事对上号。走出客厅，它面对的还有更嘈杂的环境、更复杂的任务。
10月22日，讯飞1024全球开发者节在北京举行。
主论坛上，科大讯飞董事长刘庆峰发布了基于全国产算力升级的星火X2.6，并首次推出端云一体协同架构。
AI可以根据任务需要，选择在本地处理、交给云端，或由两端协作完成。
开发者生态也迎来升级：讯飞发布星辰TaaS平台，配合APA智能体流程自动化和讯飞CLI，支持Agent操作业务系统、完成复杂任务。
开发者由此可以从调用模型，进一步走向把AI能力组合成可交付的行业方案。
同日下午的开发者生态发布会，聚焦开发者怎样接入这些能力、把应用做出来。
机器人要从几个人的声音里分清谁在提问，办公室里的Agent则要替员工核价、填单，电视要记住厨房里刚聊过的片单。
讯飞给新一代人机交互定了三个方向：拟人、普适、易用。
- 拟人，是让对话更像人，靠多通道端到端模型、端云协同和多设备联动；
- 普适，是从安静的房间走进噪音环境，从国内走向多语种，从单一场景走向多场景；
- 易用，是让开发者接入更省事，靠软硬一体模组、整机参考设计、交互研发数字员工和平台的命令行版本。
拟人、普适、易用，也贯穿了家庭、企业和公共服务这几类场景。
把人说的话、镜头里的画面放在一起理解，就是多模态交互。
OpenAI在做实时语音对话，Google把视频理解和屏幕共享带进Gemini Live。用户可以拿起手机，边给AI看、边跟它聊。
讯飞这次拿出来的，是让对话在多个设备之间互通。
用户在厨房说了一句「放假了，想放松一下」，冰箱先应答，电视上随即出现推荐片单。走到客厅后，他接着说：「看第二部，之前看过一点，帮我回顾剧情。」
用户没再报片名，电视也能明白他说的是刚才那张片单里的第二部电影。WallEX把厨房里的对话和推荐结果同步给了客厅设备，用户换个房间说话，不必从头再讲一遍。
WallEX多设备联动
能这样接话，底层是讯飞的多通道端到端模型：端侧快速应答，云侧深度理解，再把结果在设备之间同步。
WallEX背后，是讯飞面向智慧空间做的一套软硬件方案，叫SpaceMind。
在商业、居住、办公这些空间里，SpaceMind通过开放的IoT协议和Harness平台，兼容全球主流通信标准。
它已经和海尔、美的等家电品牌，以及西门子、GVS等电工电气品牌的设备打通。
不同品牌、不同品类的设备都能接进来，智能就不再停在某一台设备上，而是整个空间一起协同。
讯飞的目标就是，用端云协同和软硬一体，把智能硬件的开发门槛再降一截，让AI进到更多家庭和行业里。
从相对安静的家里走进展厅、酒店，机器人面对的声音就复杂多了。
几个人围在机器人面前说话，旁边还放着音乐。提问声混在聊天声里，机器人既要听清问题，也要分清是谁在问。
讯飞用全景摄像头配合环形麦克风阵列，把人像和声音对应起来，在360°范围内支持5人多模态降噪。
在-10dB噪声测试中，这套方案的语音识别准确率为90.1%。
到了展厅或酒店，访客还可能换一种语言开口。平台支持40多种语言，其中30多种可以免切换交互。
问完一句话，等上好几秒才有回应，人很容易失去继续聊的耐心。
讯飞让设备在本地处理即时交互和离线任务，把复杂理解交给云端。这种「端云融合」方案，端侧平均响应时间为0.85秒。
给孩子讲故事，AI得跟着情节调整语气；在门店接待顾客，又要换一种问问题、接话的方式。开发者可以选用儿童陪伴、门店营销等场景的交互模板，再按需要调整。
同样是讲故事，语音合成模型还要把情节里的情绪读出来。
在真实场景中，语音依然是人与AI最自然、最直接的交互方式。除了听得清、反应快，AI还得说得准、说得自然。
而在这个讯飞深耕多年的领域，它的技术实力再次得到了验证。
讯飞的语音合成模型ULingo-TTS-1.0-Preview，在10月16日SuperCLUE-TTS测评总榜中拿到71.39分。
这一总分为全部参测模型最高。
交互能力有了，接下来是把它装进产品。
把AI装进眼镜、录音环或导览机器人，开发团队还要调试麦克风收音、摄像头取景，以及模型怎样控制设备。
换一种硬件，这些部分就可能要重新适配。
开发者可以先拿适配好的硬件试一轮。讯飞提供交互板、软硬一体模组和一体机，团队验证效果后，再按自己的产品调整。
团队也可以参考讯飞自有产品的整机设计，确定软硬件怎么组合。
接入和适配时，开发者可以直接描述需求，由Agent调用AIUI Code插件辅助完成。这些插件封装了讯飞的交互开发经验。
在家里说「看第二部」，电视接着响应就好。到了办公室，一句「采购显示器」，还要把事情交到有权限处理的同事手里。
会调用工具、按目标执行任务的AI，被称为Agent，也就是智能体。
OpenAI的开发框架已经支持Agent之间交接任务、记录过程。进了公司，这些交接还得遵守公司的分工。
这也是这次发布会最重的部分：三款面向企业的首发产品——Agent OS、APA和星辰TaaS，分别解决分工、操作和方案三件事。
预算由谁确认、合同由谁审核，不能让Agent临时猜。
此次首发的Agent OS，就是给人、多个Agent和业务系统定分工的。
员工提出采购显示器，系统整理用途与规格，登记进办公系统；到了审批环节，通知负责人处理，再把结果反馈给员工。
Agent接手时，还要知道这笔采购办到了哪一步。它按权限读取项目、人员和历史记录，接着处理自己负责的任务。
员工的纠正意见和任务结果，也会经过评测审核，用于调整后续技能和流程。
Agent OS自动登记采购申请
Agent OS审批协同与结果回传
分工定了，Agent还得会操作公司的各个系统。
查库存、填表、提审批，员工平时怎么操作，就可以把过程录下来，再用对话补充要求，教给Agent。
这次首发的APA，全称智能体流程自动化，把软件操作变成Agent可调用的技能。
员工在星辰Cowork工作台里录制、描述操作过程，Agent也在这里调用技能，跨系统办事。
客户要10台无线接入设备和2台交换机，Agent先到经营管理和仓库系统核对价格、库存，再查合同中的付款与交付条款。
核对完成后，它生成16800元的报价，并送进办公系统审批。
录好的技能可以分享给同事，免得每个人再教一遍。
再换一家客户，业务系统和行业规则又不同了。开发者能调模型、接工具，也还得把这些能力组织成客户用得上的方案。
此次首发的星辰TaaS，让开发者复用模型、工具和行业知识。企业在AI实验室提出需求，开发者据此组合方案，在线验证后，再推进交付。
TaaS的全称是Token as a Solution。讯飞希望，模型消耗的每个Token都能用于推进具体业务，直到任务完成。
这套方案由模型供给、能力编排和场景应用三层配合：
- 模型供给：星辰MaaS提供底层模型能力。
- 能力编排：通过模型调度、智能体引擎、知识工程、工具编排和运行治理，组织模型与工具，让Token用得准、用得省。
- 场景应用：已沉淀同传翻译、数字员工、办公提效、知识问答、智能营销等应用，面向企业服务、金融、医疗、教育、能源等行业。
开发者可以复用现成体系，减少从零搭建的工作；企业则按业务需求获得可以直接使用的能力。
讯飞自己的总结是一句话：让Token不空转，让价值看得见。
星辰TaaS组合模型、知识与工具
星辰TaaS生成方案
做招采评审时，审核人尤其需要知道AI凭什么下结论。讯飞招采助手让多个专业Agent并行处理材料，附上引用依据，关键节点交给人确认。
这套系统已在50多家单位落地。在国家能源集团相关项目中，智能评审准确率达到97%。
与合肥公共资源交易中心联合打造的公共资源交易大模型，则让评标评审时长缩短90%以上。
海外广告投出去以后，团队要盯着成本和转化，排查流量、素材出了什么问题。
「超级效果优化师」让多个Agent分工检查、诊断和提出调整策略，投手确认后再执行，已用于电商大客户的海外投放。
在某电商KA客户的海外投放案例中，人均操盘账户的投放规模提升40%，ROAS（广告支出回报率）提升50%。
Agent在公司里干活，也要用讯飞开放平台上的能力。但它找工具的方式，跟人不一样。
开发者让Agent识别一段录音，首先得让它找到语音识别服务，知道怎么调用，并拿到相应权限。
Anthropic提出MCP协议，推动AI用统一方式连接外部数据和工具。
讯飞这次将推出XFYUN CLI，把平台功能做成可以用一句话调用的工具。
开发者完成安装、授权后，Agent就能用这些指令请求服务。
语音识别、语音合成等700多项AI功能，将通过这套方式供Agent调用，适配范围覆盖200多种智能体。
工具找到了，最后一步是把Agent装进现场的设备。
过去，给智能硬件接入交互能力，开发者要自己写代码，再完成编译、调试和端侧部署，链路长，联调也复杂。
前面提到的AIUI Code，通过Agent调用插件辅助开发，减少手工编码和联调。
交互研发数字员工则进一步支持用自然语言配置机器人。开发者发出语音指令，就能按场景调整配置并即时更新。
企业还需要把设备配置和实际服务衔接起来。
企业的定制需求往往很零散：这里要个数字人，那里要个问答，常常需要分开采购。
GuideX平台这次升级，把分散的需求组合成一套可交付、可持续运营的方案。
方案里有两个角色。
- AStudio是讯飞新一代的全场景AI生产力工作台，装在开发者的电脑上，负责接需求：梳理要求、搭建Agent。
- GuideX是讯飞面向公共服务场景的数字员工平台，接手从配置组装到交付运营的后续流程。 开发者通过对话式开发，搭建主动迎宾、多语种问询、业务办理等功能，再接入带触控屏、能对话的终端。
发布会上要现场演示的，是给机场定制一个服务管家。
开发者先在AStudio里梳理需求、搭建Agent，再交给GuideX配置，接入数字人和终端设备，测试、调试之后交给机场使用。
在肯尼亚运营商Safaricom的总部智慧展厅，GuideX数字导览员已用于介绍通信业务、M-PESA移动金融服务和财报内容。
项目团队为此整理专属知识库，适配行业名词、财报缩写和当地表达。
10月24日，讯飞AI开发者大赛年度决赛将在合肥举行。今年的大赛吸引了全球15个国家、351座城市的52120支团队，规模创下新高。
参赛团队的构成，也反映了产业落地和参与门槛的变化。
一是AI创新在往产业里走：应用赛里65%的参赛团队拿到了国家资质认证，讯飞也在通过技术支持、路演和资本对接，帮这些团队把作品变成产品。
二是门槛在降：算法赛里43%的选手不是算法专业出身，过去要靠多年积累才能碰的算法优化，现在借助AI工具，跨专业的人也能上手。
有些需求很小，小到不值得专门开发一套软件，却会每天耗掉一些人的时间。
如果搭建应用能更省力，熟悉这些麻烦的人就有机会把自己的办法做出来，再交给更多人用。我们下一次觉得「这东西真方便」，也许就来自某个人对身边一件小事的不满意。
讯飞把开放平台往Agent这边再推一步，就是想让这件事更容易发生。
从客厅里寻找一只高跟鞋，到机场里为成千上万的旅客提供服务，看似毫不相干的场景，背后其实是同一场变革：AI正在从屏幕里的回答者，变成现实世界的参与者。
而决定这场变革能走多远的，是能否让技术走出实验室，让开发者更容易创造，让企业真正用起来，让普通人的想法也有机会变成产品。
讯飞此次开发者生态升级，也在推动开发方式变化：人们从学习使用AI工具，走向与Agent共同创造应用。
当开发AI应用不再只是少数程序员的专长，更多懂得实际问题的人就能参与创造。技术突破要转化为生产力，还得看这些应用能在真实场景里解决多少问题。
编辑：摩西 所罗门
不喜欢
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
New Ji-won.
Mom's home in 10 minutes and the girl can't find another heels.
"Where's my other heels?"
The screen on the wall began to replay the living room. AI tip: Shoes could be kicked under the couch. She fell down and looked for it.
The girl did not speak of the colour of the shoes, the style, nor did she recall where she had last seen them. Wallex linked her words to the scene in the living room and gave her a clue.
Looking for shoes is a small matter, but for AI to do it, you have to understand people and visualize them. Out of the living room, it also faces more noisy environments and more complex tasks.
On 22 October, the iFlytek1024 Global Developer Festival was held in Beijing.
At the main forum, Liu Qingfung publish, Chairman of the Board of Directors of iFlytekK, Kodak, presented Starfire X2.6, based on the upgrading of the country ' s production capacity, and launched, for the first time, a cloud-like synergetic architecture.
AI may choose to handle it locally, hand it over to the clouds, or be done in collaboration with both ends, depending on the mandate.
Developer ecology is also being upgraded: the iFlytekpublish Star TaaS platform, which supports Agent ' s operating system and complex tasks in conjunction with APAAI agent process automation and iFlytekCLI.
This allows developers to move further from a model to a portfolio of AI capabilities into a deliverable industry programme.
In the afternoon of the same day, the Developer Eco-publish meeting focused on how developers could access and apply these capabilities.
The robot needs to know who's asking questions from the voices of a few people, and Agent in the office has to write down the price for the employee, fill out the bill, and the TV has to remember the list we just talked about in the kitchen.
iFlytek interacts with the new generation of human beings in three directions: designer, universal, easy to use.
- The terminator is to make the dialogue more human, based on multiple end-to-end models, cloud-to-end synergies and multi-equipment connections;
- Generalization, from a quiet room into a noise environment, from home to multilingualism, from a single scene to a multiple scene;
- Easy to use is more economical access for developers, based on soft and hard integrated modules, in-house reference designs, interactive development of digital staff and platform version command line.
It is easy, easy to use, and it cuts across family, business and public services.
The combination of what people say, the images in shot, is a multi-modular interaction.
Open AI is making live voice conversations, and Google brings video understanding and screen sharing into Gemini Live. Users can pick up the phone, show it to AI, talk to it.
iFlytek came out this time to allow the dialogue to interact with multiple devices.
In the kitchen, the user said, "Take a break, try to relax," and the fridge responded first, and the recommendation sheet appeared on TV. And when he went to the living room, he said, “Look at the second part, a little something before it, and help me recall the story.”
The users didn't report the film, and the TV could understand that he was talking about the second movie on that list. WallEX synchronized the conversation and recommendations in the kitchen with the living room, and users spoke in a different room, without having to start over.
WallEX multi-equipment connection
To answer this, the bottom is a multi-channel-to-end model of iFlytek: the end-side rapid response, cloud-side depth understanding, and synchronization of the results between the devices.
Behind WallEX is a package of software and hardware that iFlytek makes for intelligent space, called SpaceMind.
In the commercial, residential and office spaces, SpaceMind is compatible with global mainstream communication standards through open IOT and Harness platforms.
It has been connected to electrical brands such as Hair, U.S.A. and electrical brands such as Siemens and GVS.
When different brands, different types of equipment are available, intelligence is no longer on a particular device, but the whole space is working together.
The goal of iFlytek is to reduce the threshold for the development of smart hardware to a greater number of households and industries, using cloud co-ordinated and hard and hard.
The voices of robots are much more complex when walking from relatively quiet homes to progress halls and hotels.
Several people speak in front of robots and there's music. The questions are in the conversation, and the robot needs to hear the questions and tell who's asking.
iFlytek uses a panoramic camera to match a circular microphone array, matching the human image with the sound and supporting five people at 360° level with multimodular noise reduction.
The speech recognition accuracy rate for the programme was 90.1 per cent in the 10-dB noise test.
Visitors may speak in another language when they reach the gallery or hotel. The platform supports more than 40 languages, of which more than 30 are non-interchangeable.
After a few seconds to respond, it is easy to lose the patience to continue talking.
iFlytek allows the equipment to handle instant interaction and offline missions locally and hand over complex understandings to clouds. This "end cloud integration" program has an average end-side response time of 0.85 seconds.
To tell a story to a child, AI has to adapt its tone to the circumstances; to receive customers at a door shop, and to ask questions and answer questions. Developers can select interactive templates for scenarios such as child companionship, doorshop marketing, and then adjust them as needed.
It's also a story, and the speech synthesis model has to read the emotions in the plot.
In the real scene, voice remains the most natural and direct interaction between people and AI. Apart from hearing and responding quickly, AI has to be clear and natural.
And in this area where iFlytek has been working for many years, its technical strength has once again been tested.
iFlytek ' s speech synthesis model ULingo-TTS-1.0-Preview received 71.39 points in the SuperCLUE-TTS General Review on 16 October.
This overall division is the highest for all reference models.
There's interactive power, and then it's loaded into the product.
Put AI in glasses, audio loops, or guide robots, and the development team is also going to debug microphones, camera viewing, and how models control the equipment.
In the case of hardware, these components may have to be refitted.
The developer can first try with the right hardware. iFlytek provides interactive panels, soft and hard integrated modules and single machines, which are then adjusted to their own products after the team validates their effects.
The team could also use the whole machine design of iFlytek's own product to determine how software and hardware are combined.
When accessing and fit, the developer can directly describe the demand, supported by Agent calling the AIUI Code plugin. These plugins contain the interactive development experience of iFlytek.
At home, you say, "Look at the second one," and then the TV responds. At the office, a "shopping monitor" was used to put things in the hands of colleagues with permission.
AI, known as Agent, known as AI agent, which calls for tools to carry out targeted tasks.
Open AI ' s development framework has supported the transfer of tasks between Agent and the recording process. Into the company, these handovers are subject to the division of labour.
This is also the most important part of this publish: three of the first products for the enterprise — Agent OS, APA and StarTaas — address the division of labour, operation and programme, respectively.
Who confirmed the budget and who audited the contract, and Agent could not be given an ad hoc guess.
The launch of Agent OS was designed to create a division of labour between people and multiple Agent and business systems.
The staff member submits the purchase monitor, systematizes the use and specifications and registers it into the office system; at the time of approval, the responsible person is notified of the process and the results are fed back to the staff member.
When Agent took over, it was also important to know what step the procurement had taken. It reads projects, personnel and historical records according to permission and then handles its own mandate.
Staff corrections and mission results are also evaluated and reviewed for subsequent skills and processes.
Agent OS automatically registers procurement requests
Agent OS Approval Synergy and Results Relay
There is a division of labour, and Agent has to operate the company's systems.
Checking inventory, filling out forms, filing approvals, and how staff normally operate, will allow them to record the process and use the dialogue to supplement the requirements and teach Agent.
This launch of APA, fully known as AI agent, automates the software operation into Agent-able skills.
Staff recorded and described the operation at the StarCowork workstation, where Agent called on skills to cross-systems.
The customer required 10 wireless access equipment and 2 exchanges, and Agent checked the price, inventory and payment and delivery terms of the contract with the operating management and warehouse system.
Upon completion of the reconciliation, it generated a quote of $16,800 and sent it to the office system for approval.
Good recording skills can be shared with colleagues, lest everyone teach them again.
Another customer, the business system and industry rules are different. Developers are able to adapt models, connect tools and organize these capabilities into programmes that clients can use.
The launch of the Star TaaS gave developers access to models, tools and industry knowledge. Enterprise demand is presented at the AI laboratory, whereby developers group the programme and validate it online before advancing delivery.
Taas' full name is Token as a Solution. iFlytek hopes that every Token consumed by the model will be used to advance specific operations until the mission is completed.
The package is complemented by three tiers of model supply, capability organization and scenario application:
- Model supply: Star Maas provides bottom modelling capability.
- Capacity development: Token is used properly and cost-effectively through modeling, AI agent engine, knowledge engineering, tool organization and running governance, organizational models and tools.
- Site applications: co-translators, digital workers, office effects, knowledge questions, smart marketing, etc., for business services, finance, health, education, energy, etc.
Developers can reuse off-the-shelf systems to reduce the number of zero-cost jobs; enterprises can acquire direct-use capabilities according to business needs.
iFlytek's own summary is a sentence: let Token not be empty and value visible.
StarTaaS combination models, knowledge and tools
StarTaaS Generation Program
In particular, the reviewers need to know what AI ' s conclusions are when conducting the appeal. iFlytek recruitment assistants have multiple professional Agents handle parallel materials, with references, and key nodes are confirmed.
The system has landed in over 50 units. In projects related to the National Energy Group, the accuracy rate of the smart appraisal was 97 per cent.
LLM, a joint venture with the Fertilizer Public Resource Trading Centre, reduced the length of evaluation of bids by more than 90 per cent.
After overseas advertising, the team had to look at the costs and transformations, check the traffic, and what happened to asset.
The Super Effects Optimist has enabled various Agents to work together to examine, diagnose and propose adjustment strategies, which are confirmed by the pitchers and are then implemented and used for overseas placements of large electrician customers.
In the case of the overseas placement of an electrician, KA client, the per capita launch of a master account increased by 40 per cent and ROAS by 50 per cent.
Agent works at the company, using the capabilities of the iFlytek open platform. But it's not the same way to find tools.
The developer asked Agent to identify a recording, first to find a voice recognition service, know how to call and get the corresponding permission.
Anthropic presented the MCP agreement to promote a unified approach by AI to external data and tools.
iFlytek will launch the XFYUN CLI this time, making the platform function a tool that can be called in one sentence.
Upon completion of the installation and authorization by the developers, Agent can use these instructions to request services.
More than 700 AI features, such as voice recognition and speech synthesis, will be made available to Agent for call, covering over 200 AI agents.
The tools were found, and the last step was to put Agent in the equipment.
In the past, the ability to interact with smart hardware access has required developers to write their own codes, complete compilation, debugging and end-side deployments, chain lengths, and complex interlinking.
APUI Code, referred to earlier, is supported by Agent calling plugins to reduce manual coding and connection.
The interactive development of digital workers is further supporting the deployment of robots in natural languages. The developers give voice commands to adjust the configuration to the scene and update it instantaneously.
Enterprises also need to link equipment configuration and actual services.
The customized needs of enterprises are often fragmented: a digital human is required here, there is a question-and-answer period, and often separate purchases are required.
This upgrade of the GuideX platform combines dispersed needs into a package of deliverables and sustainable operations.
There are two roles in the programme.
- AStudio, a new generation of iFlytek ASI productivity workshops, installed on the developers ' computers, is responsible for demand: combing requirements and building Agent.
- GuideX is a digital employee platform for iFlytek public service scenes to take over the follow-up process from configuration assembly to delivery. Through dialogue-based development, the developer develops active hospitality, multilingual queries, business processes, etc., and re-enters the terminal with touch screens and dialogue.
At the publish conference, a live presentation was made for a service housekeeper at the airport.
The developers first comb the demand in AStudio, build Agent, then hand it over to GuideX for configuration, access to digital human and terminal equipment, testing, debugging and then handing it over to the airport.
The GuideX digital guide has been used to present communications operations, M-PESA mobile financial services and financial reporting at the headquarters wisdom fair in Safaricom, Kenya.
For this purpose, the project team compiles an exclusive knowledge base that adapts trade terminology, financial abbreviations and local expressions.
On 24 October, the annual finals of the iFlytek AI developer competition will be held at the Fats. This year ' s competition attracted 52120 teams from 15 countries and 351 cities around the world, at a new level.
The composition of participating teams also reflects changes in the threshold of industrial landing and participation.
One is that AI innovation is moving towards industry: 65 per cent of the teams in the application competition have been certified as national qualifications, and iFlytek is helping these teams turn their work into a product through technical support, road shows and capital matching.
The second is that the threshold is falling: 43 per cent of the players in the algorithms are not professional in algorithms, and the algorithms that used to be accumulated over many years have been optimized, and now, with the help of AI tools, cross-professionals.
Some of the needs are so small that they do not warrant the development of a specialized set of software that they consume some people's time every day.
If build-up applications are more energy-efficient, those who are familiar with these problems have the opportunity to make their own solutions available to more people. The next time we think it's convenient, maybe it comes from someone who's not happy with a little thing.
iFlytek took the open platform to Agent to make it easier.
Looking for a high heels in the living room to serve thousands of tourists at the airport seems to be a irrelevant scene behind the same change: AI is turning from a screen respondent to a participant in the real world.
How far this change can go is to get technology out of the lab, make it easier for developers to create, make businesses really useful, and give ordinary people the chance to turn their ideas into products.
This eco-upgrading of iFlytek by developers is also driving a change in the way development works: from learning to using AI tools to creating applications with Agent.
When the development of AI applications is no longer the expertise of a small number of programmers, more people with practical knowledge can be involved in creating them. Technological breakthroughs translate into productivity, depending on how many problems these applications can solve in the real scene.
Editor: Moses Solomon
I don't like it.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.