Imagine 2026: Part One
God Translation Bureau is a translation and compilation team under 36Kr, focusing on fields such as technology, business, the workplace, and life. It mainly introduces new technologies, new ideas, and new trends from abroad.
Editor's note: In 2026, screen time is dead, and the record - keeping system has stepped down. When AI starts to design infrastructure for AI, the ultimate moat of business is shifting from traffic to collaboration and personalization. This is the first part of a series of articles, and the article is from a compilation.
As investors, our responsibility is to conduct in - depth research on every corner of the technology industry and gain insights into future development trends. Therefore, every December, we invite our investment teams to share a major topic that they believe technology developers will strive to tackle in the coming year.
Today, we will share the insights from the Infrastructure, Growth, Bio + Health, and Speedrun teams. The views of other teams will be released later, so stay tuned.
Infrastructure
Jennifer Li: Start - ups Tame the Chaos of Multimodal Data
Unstructured, multimodal data has always been the biggest bottleneck for enterprises and also their largest untapped treasure. Every company is drowning in a "quagmire" of PDFs, screenshots, videos, logs, emails, and semi - structured data. Although models are becoming smarter, the input data is becoming messier, which causes RAG (Retrieval - Augmented Generation) systems to generate hallucinations, agents to malfunction in subtle and costly ways, and key workflows still rely heavily on manual quality assurance (QA). Currently, the factor limiting AI companies is "data entropy": in the unstructured world that contains 80% of corporate knowledge, the real - time nature, structuring level, and authenticity of data are constantly deteriorating.
This is why organizing unstructured data has become an epoch - making opportunity. Enterprises need a continuous method to clean, structure, validate, and manage their multimodal data so that downstream AI tasks can be truly implemented. Application scenarios are everywhere: contract analysis, onboarding processes, claims processing, compliance checks, customer support, procurement management, engineering searches, sales enablement, analysis pipelines, and every agent workflow that relies on reliable context. If start - ups can build platforms that extract structured information from documents, images, and videos, reconcile conflicts, repair pipelines, or keep data real - time and retrievable, they will hold the key to the treasure chest of corporate knowledge and processes.
Joel de la Garza: AI Transforms Cybersecurity Recruitment
For most of the past decade, the biggest challenge faced by Chief Information Security Officers (CISOs) has been recruitment. From 2013 to 2021, the number of unfilled cybersecurity positions increased from less than 1 million to 3 million. This is because security teams hired highly skilled technicians but made them deal with exhausting front - line basic work (such as viewing logs) all day long, and no one wants to do such work. The core of the problem is that security teams bought products that can detect everything, artificially creating a heavy workload - which means the team has to review every alert, leading to a false shortage of manpower. This is a vicious cycle.
By 2026, AI will break this cycle and fill the recruitment gap by automating most of the repetitive and redundant work for security teams. Anyone who has worked in a large security team knows that half of the work can be completely solved by automation, but when you are deeply stuck in complicated affairs, you simply don't have the energy to think about which parts to automate. AI - native tools that sort out these problems for security teams will finally free them from trivial matters and allow them to do what they really want to do: track down bad guys, build new systems, and fix vulnerabilities.
Malika Aubakirova: Agent - Native Infrastructure Becomes the Entry Barrier
By 2026, the biggest infrastructure impact will no longer come from external companies but from within. We are transitioning from predictable, low - concurrency "human - speed" traffic to recursive, sudden, and large - scale "agent - speed" loads.
Today's enterprise back - ends are designed for a 1:1 ratio of "one human operation to one system response." It doesn't take into account that an agent's "goal" can trigger 5,000 subtasks, database queries, and internal API calls in a recursive fan - out within milliseconds. When an agent tries to refactor a codebase or fix security logs, its behavior is completely different from that of an ordinary user. To traditional databases or rate limiters, it looks more like a DDoS attack.
Building systems for agents in 2026 means re - engineering the control plane. We will witness the rise of "agent - native" infrastructure. Next - generation systems must treat the "thundering herd" pattern as the default state. Cold - start times must be shortened, latency fluctuations must be eliminated, and concurrency limits must be increased by an order of magnitude. The bottleneck will shift to coordination: routing, locking, state management, and policy execution in large - scale parallel execution. The winning platform will be the only one that can survive the ensuing flood of tool executions.
Justine Moore: Creative Tools Go Multimodal
We now have the cornerstones for storytelling with AI: generative voice, music, images, and videos. But apart from generating single - shot segments, getting the desired output is often time - consuming, frustrating, and almost impossible - especially when you want to achieve the level of control of a traditional director.
Why can't we give a model a 30 - second video and ask it to add a new character to continue the plot based on reference pictures and sounds? Or reshoot a segment so that we can view the scene from a different angle or make the action match the reference video?
2026 will be the year when AI moves towards multimodality. You can provide the model with any form of reference content and collaborate with it to create new works or edit existing scenes. We have already seen some early products, such as Kling O1 and Runway Aleph. But there is still a lot of work to be done, and we need to innovate at both the model and application levels.
Content creation is one of the killer applications of AI. I foresee that multiple successful products will emerge for different user groups and application scenarios, from meme creators to Hollywood directors.
Jason Cui: The AI - Native Data Stack Continues to Evolve
In the past year, we have seen a large - scale integration of the "modern data stack." Data companies are shifting from specialized divisions in ingestion, transformation, and computation to unified bundled platforms. For example, the merger of Fivetran and dbt, and the continuous rise of unified platforms such as Databricks.
Although the ecosystem seems to have matured significantly, we are still in the early stages of a true AI - native data architecture. What excites us is that AI will continue to transform multiple parts of the data stack, and we are also beginning to see that data and AI infrastructure are becoming inseparable.
Several directions we are focusing on include:
How data will continue to flow into high - performance vector databases in parallel with traditional structured data;
How AI agents will solve the "context problem": by continuously accessing the correct data context and semantic layer, building robust applications (such as "data conversations") and ensuring that the correct business definitions are always available across multiple record - keeping systems;
How traditional business intelligence (BI) tools and spreadsheets will change as data workflows become more intelligent and automated.
Yoko Li: The Year of Stepping Inside Videos
By 2026, videos will no longer be just content that we passively watch but more like a space that we can truly step into. Video models will finally be able to understand time, remember what has been shown, and react when we operate, maintaining an internal consistency similar to the physical world. These systems will no longer just generate a few seconds of intermittent images but will be able to maintain characters, objects, and physical laws for a long time, making actions meaningful and results visible. This transformation will turn videos into a medium for development: robots can be trained in it, games can evolve in it, designers can create prototypes, and agents can learn through practice. What is produced will no longer be just a segment but more like a living environment, beginning to bridge the gap between perception and action. For the first time, we will feel that we can "live" in the generated videos.
Growth
Sarah Wang: The System of Record Loses Its Dominance
In 2026, the real disruption in enterprise software is that the "System of Record" will finally lose its core position. AI is shortening the distance between intention and execution: models can now directly read, write, and reason across business data, transforming ITSM (IT Service Management) and CRM (Customer Relationship Management) systems from passive databases into autonomous workflow engines. With the continuous progress in the fields of inference models and agent workflows, these systems can not only respond but also predict, coordinate, and execute end - to - end processes. The interface will evolve into a dynamic agent layer, while the traditional record - keeping system will degenerate into a general persistent storage layer in the background - its strategic leverage will give way to those who control the intelligent execution environment actually used by employees.
Alex Immerman: Vertical AI Evolves from Information Retrieval and Inference to a "Multi - Person Collaboration" Model
AI has driven the growth of vertical industry software at an unprecedented speed. Companies in the medical, legal, and real estate fields have exceeded $100 million in Annual Recurring Revenue (ARR) in just a few years; the finance and accounting fields are also following closely. This evolution initially involved information retrieval: finding, extracting, and summarizing the correct information. In 2025, reasoning ability was introduced: Hebbia started analyzing financial statements and building models, Basis cross - checked trial balances across systems, and EliseAI diagnosed maintenance problems and dispatched appropriate suppliers.
In 2026, the "multi - person collaboration" model will be launched. Vertical software benefits from domain - specific interfaces, data, and integrations. However, the work in vertical industries is essentially multi - party involved. If agents are to represent the labor force, they must collaborate. From buyers, sellers to tenants, consultants, and suppliers, each party has unique permissions, workflows, and compliance requirements, which only vertical software can understand.
Currently, each party uses AI in isolation, which leads to a lack of authoritative hand - offs. The AI that analyzes procurement agreements won't communicate with the Chief Financial Officer (CFO) to adjust the model; the AI in charge of maintenance doesn't know what promises the on - site staff made to the tenant. The multi - person collaboration model will change this situation by coordinating among stakeholders: routing to functional experts, maintaining context consistency, and synchronizing changes. The other party's AI will negotiate within the parameters and mark information asymmetries for manual review. The annotations of senior partners will train the company's entire system. Tasks executed by AI will have a higher success rate.
When value is increased through the collaboration of multiple people and multiple agents, the switching cost will also increase. At that time, we will see the network effect that has been lacking in AI applications: the collaboration layer will become the moat for enterprises.
Stephenie Zhang: Create for Agents, Not Humans
By 2026, people will start to interact with the Internet through their own agents. What was important for human consumption in the past may not play the same role for agent consumption.
For years, we have been optimizing for predictable human behavior: ranking high on Google, appearing in the top few results of Amazon searches, or starting with "TL;DR". When I took journalism classes in high school, my teacher taught us to write news with the 5W + H formula and start feature stories with an eye - catching "hook". Humans may miss in - depth, relevant, and insightful statements buried on the fifth page, but agents won't.
This change is also reflected in software. Apps were designed for human eyes and clicks, and optimization meant beautiful UIs and intuitive processes. As agents take over retrieval and interpretation, the importance of visual design for understanding will decrease. There will no longer be a need for engineers to stare at Grafana dashboards; AI Site Reliability Engineers (SREs) can interpret telemetry data and publish insights in Slack. There will no longer be a need for sales teams to flip through CRM; agents can automatically present patterns and summaries.
We are no longer designing for humans but for agents. The new optimization direction is no longer visual hierarchy but "machine readability" - which will change the way we create and the tools we use for creation.
Santiago Rodriguez: AI Applications Bid Farewell to the "Screen Time" KPI
In the past 15 years, screen time has been the best indicator for measuring the value output of consumer - level and enterprise - level applications. We have always lived in a paradigm where the streaming time of Netflix, the number of clicks on Electronic Health Record (EHR) systems (to prove effective use), and even the usage time of ChatGPT are used as Key Performance Indicators (KPIs). As we move towards a future of pricing based on results (a model that perfectly aligns the incentives between suppliers and users), we will first abandon screen - time reports.
We have already seen relevant practices. When I use ChatGPT for a DeepResearch query, I gain great value even though it hardly takes up any screen time. When Abridge magically captures doctor - patient conversations and automates downstream activities, doctors hardly need to look at the screen. When Cursor develops an entire application end - to - end, engineers are already planning the next feature development cycle. And when Hebbia drafts a financing presentation based on hundreds of public documents, investment bankers are enjoying a long - awaited sleep.
This poses a unique challenge: how much an application should charge each user requires a more complex method for measuring return on investment (ROI). The satisfaction of doctors, the productivity of developers, the physical and mental health of financial analysts, and the well - being of consumers will all improve with the use of AI applications. Companies that can tell the ROI story in the most concise way will continue to lead their competitors.
Bio + Health
Julie Yoo: Healthy Monthly Active Users (MAUs)
By 2026, a new group of healthcare customers will take center stage: "Healthy Monthly Active Users" (Healthy MAUs).
The traditional healthcare system mainly serves three user groups: (a) "Sick Monthly Active Users (Sick MAUs)": people with sudden, high - cost needs; (b) "Sick Daily Active Users (Sick DAUs)": for example, those receiving intensive care or long - term care; (c) "Healthy Yearly Active Users (Healthy YAUs)": relatively healthy people who rarely see a doctor. Healthy YAUs are at risk of turning into Sick MAUs or DAUs, and preventive care can slow down this transition. However, our medical insurance reimbursement system based on "reactive treatment" rewards treatment rather than prevention, so the priority of proactive examination and monitoring services is not high, and insurance rarely covers these items.
Here come the "Healthy MAUs": they are not sick but want to regularly monitor and understand their health status. This group represents the most promising part of the consumer group. We foresee that a number of companies - including both AI - native start - ups and transformed established institutions - will start to provide circular services to meet the needs of this user group.
As AI has the potential to reduce the cost structure of care services, new types of medical insurance products focusing on prevention emerge, and consumers are becoming more accustomed to paying out - of - pocket for subscription - based services, "Healthy MAUs" will become the next highly promising customer group in the healthcare technology field: they are continuously engaged, data - driven, and prevention - oriented.
Speedrun
Jon Lai: World Models Shine in Narrative
By 2026, AI - driven world models will completely transform the way of storytelling through interactive virtual worlds and the digital economy. Technologies like Marble (World Labs) and Genie 3 (DeepMind) can already generate complete 3D environments based on text prompts, allowing users to explore them like playing a game. As creators adopt these tools, new narrative formats will emerge and may eventually evolve into a "generative Minecraft," where players jointly create a vast and ever - evolving universe. These worlds can combine game mechanics with natural language programming, for example, by giving instructions like "Create a paintbrush that turns everything I touch pink."
Such models will blur the line between players and creators, transforming users into co - authors of a dynamic shared reality. This evolution may give rise to an interconnected "generative multiverse" where different genres such as fantasy, horror, and adventure can coexist. In these worlds, as creators earn income by creating assets, guiding newbies, or developing new interactive tools, the digital economy will flourish. In addition to entertainment, these generative worlds will also serve as rich simulation environments for training AI agents, robots, and possibly even Artificial General Intelligence (AGI). Therefore, the rise of world models not only marks a new game genre but also foreshadows a brand - new creative medium and economic frontier.
Josh Lu: The "Year of the Self"
2026 will be the "Year of the Self": from this moment on, products will no longer be mass - produced but will be customized for you.
This trend is already everywhere.
In the field of education, start - ups like Alphaschool are building AI tutors that can adapt to each student's progress and curiosity, enabling each child to receive an education that matches their rhythm and preferences. Previously, this level of attention was impossible without spending