In 2010, Steve Jobs laid out a vision for Siri that has become a benchmark for what we now expect from AI assistantsAccording to Norman Winarsky, one of Siri's co-founders, Jobs understood that having a personal assistant capable of genuine human-like interaction would create unique valueThis assistant would not only chat but would also possess enough understanding to execute tasks effectively for users.
Fast forward fourteen years, and the anticipation remains unchangedPeople are still waiting for a true personal assistant, one that can “really understand them, complete a multitude of tasks, and operate across various domains.” Reports suggest that OpenAI is planning to launch an AI assistant product early next year that focuses on automating tasks such as sending work emails and creating travel plans.
The breakthroughs in large language models are believed to be rapidly propelling the reality of this expectationExecutives from OpenAI frequently emphasize that AI agents will be the next significant breakthroughIn a research report released earlier this November, Bank of America indicated that Agentic AI—with greater autonomous planning and action capabilities—is sparking a new cycle of innovation unlike anything we’ve seen with tools like ChatGPT and Copilot.
One of the platform-level opportunities that has emerged within this innovation cycle is the development of an AI assistant that is closely connected to users, offering extensive interaction and facilitating greater collaboration among various agents.
On one hand, the support from large models signifies the potential of natural language interaction to replace more graphic interface interactions, suggesting that any scenario could be reconstructed with this new form of engagementOn the other hand, shifts in interaction methods are poised to disrupt existing paradigms within the software and hardware landscapes, potentially catalyzing the decline of established beneficiaries while paving the way for newcomers.
The current competitive landscape is still in its early stages
Advertisements
It encompasses ambitious startups focusing on large models, major internet conglomerates that dominate the primary platforms, mid-sized companies that have survived between these giants, and hardware firms aspiring to capitalize on software profits, all of which are vying for a share of this burgeoning opportunityExplorations are evident in general AI assistants, vertical-specific AI assistants, and tool-oriented AI assistants.
Ultimately, the players who accumulate more advantages in areas including large-scale model technology, agent ecosystems, user acquisition efficiency, and business model innovation will be the ones poised to reap the new benefits of market entry, securing a significant position in this emerging food chain.
As noted in a report by Bank of America, the evolution of AI can be categorized into three waves: Pre-GenAI, GenAI, and Agentic AIThe Pre-GenAI phase, lasting from the 1940s until the release of ChatGPT in November 2022, featured various voice assistants like Siri, Alexa, Xiao Ai, and Tmall GenieHowever, the practicality of AI primarily served to leverage data-driven insights and decision-making in different contexts.
During the GenAI phase leading up to October 2024, applications based on GenAI focus predominantly on two aspects: fostering more natural human-machine interactions—with early iterations like ChatGPT, intelligent agents, and C.AI winning user acceptance through seamless communication—and enhancing personal productivity and efficiency through AI-driven tools for search, video generation, code generation, and more.
Silvio Savarese, the Executive Vice President and Chief Scientist of Salesforce AI Research, asserts that we are now moving into the Agentic AI phase, which marks a significant capability leap where AI will autonomously automate tasks and take action on our behalf, a far cry from the AI we have known previouslyGartner predicts that by 2028, at least 15% of everyday work decisions will be made independently by Agentic AI.
While in 2024 this figure remains zero, faint signs of this paradigm shift are already surfacing.
Firstly, enhancements in memory capabilities within large models are laying the groundwork for autonomous decision-making
Advertisements
Google’s AI assistant Gemini is already equipped to remember the life details, work content, and personal preferences of Google One AI Premium subscribers.
Secondly, technological advancements are endowing large models with multi-modal capabilities and the ability to operate applications (APPS), continuously expanding the boundaries of what AI assistants can accomplishWith multi-modal capabilities, these assistants can interact more naturally by processing auditory, visual, and textual inputsIdeally, users could achieve direct information retrieval via voice or video commands, based on visual data.
Recently unveiled technologies, such as AutoGLM from Zhihui and computer use by Anthropic, have demonstrated this capability in mobile and computing environments respectivelyAdditionally, phone manufacturers have showcased their AI capabilities through user scenarios, such as ordering takeout or booking restaurantsNevertheless, it’s important to acknowledge that these trials are still in their infancy, far removed from large-scale application.
Furthermore, a number of companies are beginning to leverage AI assistants to build their own third-party AI app ecosystemsMicrosoft has introduced the Azure AI Foundry, a platform aimed at assisting organizations in designing, customizing, and maintaining management of AI applications and assistantsBaidu has launched a new no-code tool called Miaoda, pledging to support the creation of millions of "super useful" applications.
A former Apple employee noted that the failure to swiftly develop a third-party ecosystem contributed to Siri’s stagnationIn essence, the third-party application ecosystem in the context of AI assistants represents a talent pool that can be called upon to satisfy user demands by comprehending their needs and intentionsThis might include other AI assistants and more functional AI capabilities.
The competition among AI assistants mirrors a fictional role-selection process seen in Johnnie To’s films, fraught with layers of vested interests
Advertisements
The role of 'leader' conveys power and often intertwines with profit-sharing rightsThe incumbents, being the existing internet platforms, are reluctant to relinquish their dominance, while the newcomers, embodying startups in the large model arena, are eager to enter the center of power.
Currently, the narrative of this multi-party struggle is still unfolding, with character introductions only just taking placeLarge model startups are the most proactive contenders in this upcoming power struggle, a sentiment echoed by numerous players in the industry, both in China and abroadAlmost every large model startup has unleashed its AI assistant product, continuously enhancing its content generation and interconnectivity features.
Firms boasting early advantages and heightened visibility, buoyed by capital and media attention, have shown quicker expansionData collected in October indicated that ChatGPT and Kimi’s AI assistant applications topped the global Apple AI app download charts, occupying first and third ranksHowever, a significant resource pivot from large model development toward applications might lead to diminishing returns on some AI assistant products.
Former leaders, such as Alibaba, Baidu, Tencent, and ByteDance, holding significant internet platform power, are not taking the competition for the next leadership role lightlyTheir respective AI assistants—Tongyi from Alibaba, Xinyoubao from Alipay, Wenxiao from Baidu, Yuanbao from Tencent, and Doubao from ByteDance—have established themselves within the framework, demonstrating deeper accumulations of user data, scenarios, and resources compared to less established startups.
These industry titans can leverage their extensive resources to advance their products more seamlesslyAccording to statistics from Quantum Computing Research, as of October, Doubao has garnered over 100 million cumulative downloads, leaving second-place Kimi trailing significantly at 57 million
Alibaba and Tencent, which wield extensive platform scenarios and cloud services, are poised to leverage stronger systemic advantages in the AI assistant arena.
The hardware sector missed out on the mobile internet explosion in previous timesFrom shopping to payments, and from video to music, hardware manufacturers attempted various layouts, with Xiaomi even launching the prominent e-commerce application YoupinYet, to this day, hardware firms remain tunneled by their positional paradigms, with their internet services frequently becoming merely another revenue stream fueled by advertising.
AI assistants present an opportunity for hardware manufacturers to integrate and leverage user habits, data accumulation, screen interaction, and app operation more cohesivelyIn contrast to software, hardware can facilitate a more immediate pathway for users to summon AI assistantsInitiatives like Microsoft adding a Copilot key to keyboards and the dedicated camera control button on iPhone 16 exemplify the pathways to invoking AI assistantsMoreover, hardware manufacturers are often better positioned to implement mixed cloud-edge AI solutions that ensure data security and user privacy.
Tool-based products are facing two potential routes in relation to AI assistantsOne trajectory, akin to DingTalk, Zhihu, Quark, and Meitu, involves leveraging their existing competency to introduce vertical AI assistants designed for specific scenarios, aimed at optimizing user experiences through natural language interaction and automationThe alternative approach, exemplified by Shenzhizhibo, integrates its content actively into other AI assistants from the outset.
Ultimately, who will emerge as the foremost authority hinges upon the control of the pivotal assetsIn a sense, the question is not who will wield this authority, but rather who possesses the capability to do soThis capability epitomizes the cumulative strength of various factionsReturning to the contest over AI assistants, this comprehensive strength encompasses the competence of base models, business scenarios, user acquisition, and integration of hardware and software.
First and foremost, enhancements in foundational model capabilities remain crucial
Whether it is accurately grasping user intent or flexibly deploying applications and other agents, increased dependence on advancements in foundational models is inevitable.
Especially as alternatives to graphical interactions depend on models’ capabilities to comprehend and manipulate screens and windowsRecent reports indicate that Google’s Project Jarvis is aimed at developing a new model that will operate atop the Gemini model, executing functions like screenshot interpretations, button clicks, and text inputs within browser environments.
Secondly, the breadth of capabilities that assistants can link to determines the level of access they commandIt is improbable that any single AI assistant will dominate the landscape; instead, multiple lower-tier assistants will acquire greater distribution rightsA sufficiently rich agent ecosystem indicates ample capability supply, allowing for a diverse array of functions to be integrated within a unified framework, thereby positioning the ecosystem assistants as foundationally more potent.
We see that, akin to the native AI assistants, platform-capable products attempt to evolve while providing convenient tools that allow for greater: integration of diverse agent capabilitiesMajor players like Microsoft, Baidu, and ByteDance are actively incubating ecosystems through tools, pushing to infuse the outputs of more developers into their AI assistantsSimultaneously, products like Slack and DingTalk are opening up vertical scenarios to developers, enriching their capabilities.
Thirdly, integration between hardware and assistants will be essentialCompanies must find cooperation strategies in the era of AIAccording to Zhao Ming, CEO of Honor, creating the next generation of AI operating systems involves understanding how to serve consumers, navigate inter-system operations, and collaborate with cloud-based AI agents, offering a novel opportunity for competitive advancements.
Similarly, firms such as Xiaomi, Vivo, and OPPO are keenly positioned to capitalize on restructuring AI operating systems to gain greater authority over system-level interactions historically dominated by Apple and Google while playing a more proactive role in their engagements with internet platforms
Furthermore, offerings such as AI headphones from ByteDance and AI glasses from Baidu reflect aspirations to achieve cohesive software-hardware integration.
Lastly, in situations where model capability, product experience, and ecosystem maturity are comparably aligned, the conflict between assistants often boils down to traffic acquisitionGaining dominion over the AI assistant narrative necessitates maintaining control over substantial traffic at lower costsHere, hardware manufacturers and internet platforms retain intrinsic advantagesThe overwhelming download figures for Doubao, surpassing Kimi, alongside META AI reaching over 500 million users, are manifestations of this cost-effective traffic advantage.
Conversely, the pressure for emerging players to monetize traffic is intensifyingData from App Growing reveals Kimi's investments surged in October, tallying 110 million yuan within 20 days, nearing the total investments for the preceding three-month periodDuring the same timeframe, Doubao invested 15 million yuan, and Tencent's Yuanbao allocated about 30 million yuan through its advertising servicesThroughout the entire third quarter, Kimi invested 150 million yuan, whereas Doubao allocated 200 million yuan.
A Deloitte report highlights that, in comparison to past technological revolutions, advancements in application values are inextricably linked to enhancements in foundational technologies: the upgrading of data resource infrastructures can effectively drive the expansion of application scenarios and foster innovative functionality experiences, culminating in the formation of new ecosystems that create super apps for user engagementIn the realm of AI, the quest for consumer attention has emerged as a focal point for application development, paving the way for the emergence of new user interfaces.
As such, the contest among AI assistants ultimately represents a battle for user interfaces and distribution rightsYet, this time around, the victors stand a better chance of serving both consumer and business needs, propelled by advancements in interaction, decision-making, and execution capabilities.
Advertisements
Advertisements