Original Title: Big Ideas 2026: Part 1
Original Author: a16z New Media
Translated by: Peggy, BlockBeats
Summary: Over the past year, breakthroughs in AI have shifted from model capabilities to system capabilities: understanding long sequences, maintaining consistency, executing complex tasks, and collaborating with other agents. The focus of industrial upgrades has thus shifted from point innovations to redefining infrastructure, workflows, and user interactions.
In the annual "Big Ideas," a16z's four investment teams provide key insights for 2026 from four dimensions: infrastructure, growth, healthcare, and interactive worlds.
Essentially, they collectively depict a trend: AI is no longer just a tool, but an environment, a system, and an agent acting in parallel with humans.
Below are the judgments of the four teams regarding structural changes in 2026:
As investors, our job is to delve into every corner of the tech industry, understand its operational context, and judge the next evolutionary direction. Therefore, every December, we invite each investment team to share what they believe will be a "big idea" that tech entrepreneurs will tackle in the coming year.
Today, we present the views of the Infrastructure, Growth, Bio + Health, and Speedrun teams. Insights from other teams will continue to be released tomorrow, so stay tuned.
Infrastructure Team
Jennifer Li: Startups Will Tame the "Chaos" of Multimodal Data
Unstructured, multimodal data has always been the biggest bottleneck for enterprises and the largest untapped treasure. Every company is inundated with PDFs, screenshots, videos, logs, emails, and various semi-structured "data mud." Models are becoming smarter, but the input is increasingly chaotic—leading to hallucinations in RAG systems, causing agents to make subtle and costly errors, and keeping critical workflows highly dependent on manual quality checks.
Today, the real limiting factor for AI companies is data entropy: in a world of unstructured data that contains 80% of a company's knowledge, freshness, structure, and authenticity are continuously deteriorating.
For this reason, untangling the "mess" of unstructured data is becoming a generational entrepreneurial opportunity. Companies need a continuous method to clean, structure, validate, and govern their multimodal data, allowing downstream AI workloads to truly take effect. Application scenarios are everywhere: contract analysis, user onboarding, claims processing, compliance, customer service, procurement, engineering retrieval, sales enablement, analytics pipelines, and all workflows of agents that rely on reliable context.
Platforms that can extract structure from documents, images, and videos, reconcile conflicts, repair data pipelines, and maintain data freshness and retrievability will hold the "keys to the kingdom" of enterprise knowledge and processes.
Joel de la Garza: AI Will Reshape the Recruitment Dilemma for Cybersecurity Teams
For the past decade, the biggest headache for CISOs has been recruitment. From 2013 to 2021, the global cybersecurity job gap skyrocketed from less than 1 million to 3 million. The reason is that security teams need highly specialized technical talent but have them engaged in exhausting tier-one security tasks, such as log analysis, which almost no one wants to do.
The deeper root of the problem is that cybersecurity teams have created their own drudgery. They purchase "one-size-fits-all detection" tools, forcing the team to "review everything"—which in turn creates an artificial "labor shortage," forming a vicious cycle.
In 2026, AI will break this cycle by automating the vast majority of repetitive and redundant tasks, significantly narrowing the talent gap. Anyone who has spent time in a large security team knows that half of the work could be solved through automation; the problem is that when you are overwhelmed with work every day, you cannot step back to think about what should be automated. Truly AI-native tools will accomplish this for security teams, allowing them to finally focus their energy back on what they originally wanted to do: tracking attackers, building systems, and fixing vulnerabilities.
Malika Aubakirova: Agent-Native Infrastructure Will Become the "Standard"
The biggest infrastructure upheaval in 2026 will not come from the outside but from within. We are shifting from "human speed, low concurrency, predictable" traffic to "agent speed, recursive, explosive, massive" workloads.
Current enterprise backends are designed for 1:1 "human action to system response." They are not suited to handle a single "goal" from an agent triggering 5,000 sub-tasks, database queries, and internal API calls in a millisecond recursive storm. When an agent attempts to refactor a codebase or fix security logs, it does not behave like a user; to traditional databases or rate limiters, it resembles a DDoS attack.
To build systems for the agent workloads of 2026, the control plane must be redesigned. "Agent-native" infrastructure will begin to rise. The next generation of systems must treat "herd effects" as the default state. Cold starts must be shortened, latency fluctuations must converge, and concurrency limits must be increased by orders of magnitude.
The real bottleneck will shift to coordination itself: routing, lock control, state management, and policy execution in large-scale parallel execution. Platforms that can survive the flood of tool calls will become the ultimate winners.
Justine Moore: Creative Tools Will Fully Transition to Multimodal
We already have the basic components for storytelling with AI: generative sound, music, images, and videos. However, as long as the content is more than just a short clip, achieving director-level control remains time-consuming, painful, and even impossible.
Why can't we let a model receive a 30-second video, create a new character using our provided reference images and sounds, and continue shooting the same scene? Why can't we let the model "re-shoot" from a new angle or match actions to a reference video?
2026 will be the year AI truly achieves multimodal creation. Users will be able to throw any form of reference content at the model, collaboratively generating new works or editing existing scenes.
We have already seen the emergence of initial products, such as Kling O1 and Runway Aleph, but this is just the beginning—both the model layer and application layer need new innovations.
Content creation is one of AI's "killer applications," and I expect multiple successful products to emerge across various user groups—from meme creators to Hollywood directors.
Jason Cui: AI-Native Data Stacks Will Continue to Evolve
Over the past year, the "modern data stack" has been noticeably consolidating. Data companies are moving from modular services for collection, transformation, and computation to bundled and unified platforms (such as the Fivetran/dbt merger and Databricks' expansion).
Although the ecosystem has matured, we are still in the early stages of a truly AI-native data architecture. We are excited about how AI will continue to transform various aspects of the data stack and are beginning to see data and AI infrastructure irreversibly move toward deep integration.
We are particularly focused on the following directions:
How data continues to flow to high-performance vector databases beyond traditional structured storage.
How AI agents will solve the "context problem": continuously accessing the correct data semantics and business definitions, allowing applications that "converse with data" to maintain consistent understanding across multiple systems.
How traditional BI tools and spreadsheets will evolve as data workflows become more agentified and automated.
Yoko Li: We Will Truly "Step Inside the Video"
In 2026, video will no longer be a form of passive viewing content but will begin to transform into a place we can "step into." Video models will finally be able to understand time, remember previously presented content, and react as we take actions, while maintaining a stability and coherence close to the real world, rather than just outputting a few seconds of unrelated images.
These systems will be able to maintain characters, objects, and physical laws over longer periods, allowing actions to have real impacts and enabling causality to unfold. Video will thus shift from a medium to a space where things can be built: robots can train within it, game mechanics can evolve, designers can prototype experiments, and agents can learn by "doing."
The world presented will no longer resemble a short video but will look like a "living environment," beginning to bridge the gap between perception and action. This will be the first time humans can truly "inhabit" the videos they generate.
Growth Team
Sarah Wang: The Status of "Record Systems" in Enterprises Will Begin to Waver
In 2026, the true transformation of enterprise software will come from a core shift: the central status of record systems will finally begin to decline.
AI is compressing the distance between "intention" and "execution": models can directly read, write, and infer enterprise operational data, transforming ITSM, CRM, and other systems from passive databases into autonomous workflow engines.
With the rapid advancement of reasoning models and agent workflows, these systems are no longer just responding to demands but are capable of predicting, coordinating, and executing end-to-end processes.
Interfaces will become dynamic agent layers, while traditional system record layers will gradually retreat to being a "cheap persistent storage," with strategic dominance yielding to players who control the intelligent execution environment.
Alex Immerman: Vertical AI Will Upgrade from "Information Retrieval and Reasoning" to "Multiplayer Collaboration Mode"
AI is driving explosive growth in vertical industry software. Companies in healthcare, legal, and housing sectors have quickly surpassed $100 million ARR; finance and accounting are following closely behind.
The initial revolution was information retrieval: finding, extracting, and summarizing information.
2025 brought reasoning: Hebbia analyzes financial statements, Basis reconciles trial balances across multiple systems, and EliseAI diagnoses repair issues and schedules vendors.
2026 will unlock "multiplayer mode."
Vertical software inherently possesses industry-specific interfaces, data, and integration capabilities, while work in vertical industries is essentially multi-party collaboration: buyers, sellers, tenants, consultants, and suppliers, each with different permissions, processes, and compliance requirements.
Today, various AI systems are fighting their own battles, leading to chaotic and unauthoritative handoff points: AI analyzing contracts cannot communicate with the CFO's modeling preferences; maintenance AI is unaware of the commitments made to tenants by on-site personnel.
Multiplayer mode AI will break this situation: automatically coordinating among parties; maintaining context; synchronizing changes; automatically routing to functional experts; allowing opposing AIs to negotiate within boundaries and flagging asymmetries for human review.
As the quality of transactions improves due to "multi-agent + multi-human" collaboration, switching costs will soar—this layer of collaborative networks will become the long-missing "moat" for AI applications.
Stephenie Zhang: Future Creative Subjects Will No Longer Be Humans, But Agents
By 2026, people will interact with the web through agents, and the human-centric content optimization approach will lose its original significance.
We have optimized for predictable human behavior: Google rankings; top products on Amazon; the 5W+1H and catchy openings of news articles.
Humans may overlook deep insights buried on the fifth page, but agents will not.
Software will change accordingly. Applications were previously designed for human eyes and clicks, with optimization meaning better UI and processes; as agents take over retrieval and interpretation, the importance of visual design will decline: engineers will no longer stare at Grafana; AI SREs will automatically parse telemetry and provide insights in Slack; sales teams will not need to manually sift through CRMs, as agents will automatically summarize patterns and insights.
We will no longer design for humans but for agents. The new optimization will no longer be about visual hierarchy but about machine readability. This will fundamentally change the way content is created and the tools used.
Santiago Rodriguez: The KPI of "Screen Time" Will Disappear
For the past 15 years, "screen time" has been the gold standard for measuring product value: Netflix viewing time; mouse clicks in healthcare systems; the minutes users spend on ChatGPT.
However, in the upcoming era of "outcome-based pricing," screen time will be completely eliminated.
We are already seeing signs: ChatGPT's DeepResearch queries require almost no screen time yet provide immense value; Abridge automatically records doctor-patient conversations and handles follow-up tasks, with doctors hardly needing to look at screens; Cursor completes the development of entire applications, and engineers are already planning the next phase; Hebbia automatically generates pitch decks from a vast number of public documents, allowing investment analysts to finally get some sleep.
Challenges will follow: companies need to find more complex ROI measurement methods—doctor satisfaction, developer productivity, analyst well-being, user happiness… all of which will rise with AI.
Companies that can tell the clearest ROI stories will continue to win.
Bio+Health Team
Julie Yoo: "Healthy MAUs" Become the Core User Group
By 2026, a new healthcare user group will take center stage: "Healthy MAUs" (monthly active users who are healthy but not sick).
Traditional healthcare primarily serves three types of people:
- Sick MAUs: high-cost, periodic demanders
- Sick DAUs: such as long-term critical care patients
- Healthy YAUs: those who rarely seek medical attention
Healthy YAUs can easily become Sick MAUs/DAUs, and preventive care could delay this change. However, due to the current "treatment-oriented" healthcare system, proactive testing and monitoring are almost never covered.
The emergence of Healthy MAUs changes this structure: they are not sick but are willing to regularly monitor their health status, representing the largest potential user group.
We anticipate that AI-native startups and traditional institutions will both join in to provide periodic health services.
As AI reduces the cost of healthcare delivery, preventive-oriented insurance products emerge, and users are willing to pay for subscription services, "Healthy MAUs" will become the most promising customer group for the next generation of health technology—continuously active, data-driven, and preventive-oriented.
Speedrun Team
Jon Lai: World Models Will Reshape Narrative Methods
By 2026, AI world models will completely change storytelling through interactive virtual worlds and digital economies. Technologies like Marble (World Labs) and Genie 3 (DeepMind) can generate complete 3D worlds from text, allowing users to explore as if they were playing a game.
As creators adopt these tools, entirely new forms of storytelling will emerge—potentially even giving birth to a "generative version of Minecraft," allowing players to co-create vast, evolving universes.
These worlds will blur the boundaries between players and creators, forming a shared dynamic reality. Different genres such as fantasy, horror, and adventure can coexist; the digital economy within will thrive, allowing creators to earn income by producing assets, guiding players, and developing interactive tools.
These generated worlds will also become training grounds for AI agents, robots, and even potential AGI. The world models will bring not just a new game genre but a completely new creative medium and economic frontier.
Josh Lu: "My Year"
2026 will become "my year": products will no longer be mass-produced for the "average consumer" but will be tailored for "you."
In education, Alphaschool's AI tutors match each student’s pace and interests.
In health, AI customizes supplements, exercise plans, and dietary programs for you.
In media, AI remix content in real-time according to your tastes.
For the past century, giants have won by finding the "average user"; the giants of the next century will win by finding the "individual within the average user."
In 2026, the world will no longer optimize for everyone but will optimize for "you."
Emily Bennett: The First AI-Native University Will Emerge
By 2026, we will see the first truly AI-native university—an institution built from the ground up around intelligent systems. Traditional universities have applied AI for grading, tutoring, and scheduling, but now a deeper transformation is emerging: an "adaptive academic organism" that can learn and self-optimize in real-time.
You can imagine such a university: courses, guidance, research collaborations, and campus operations are adjusted in real-time based on feedback loops; the curriculum self-optimizes; reading lists dynamically update with new research; each student's learning path changes in real-time.
Precedents have already emerged: Arizona State University’s collaboration with OpenAI has produced hundreds of AI projects; the State University of New York has incorporated AI literacy into general education.
In an AI-native university:
- Professors become "learning system architects": curating data, tuning models, and teaching students how to examine machine reasoning.
- Assessment methods will shift to "AI awareness" evaluations: not asking students if they used AI, but how they used AI.
As various industries urgently need talent that can collaborate with intelligent systems, this university will become the "talent engine" of the new economy.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
