Original Article Title: Big Ideas 2026: Part 1
Original Article Author: a16z New Media
Translation: Peggy, BlockBeats

Abstract: Over the past year, AI breakthroughs have shifted from model capabilities to system capabilities: understanding long sequences, maintaining consistency, performing complex tasks, and collaborating with other intelligent agents. The focus of industrial upgrades has also shifted from singular innovations to redefining infrastructure, workflows, and user interaction.

In the annual "Big Ideas 2026," a16z's four investment teams have provided key insights for 2026 across four dimensions: infrastructure, growth, healthcare, and the interactive world.

Essentially, they collectively outline a trend: AI is no longer just a tool but an environment, a system, an acting entity parallel to humans.

Below are the four teams' assessments of the structural changes in 2026:

As investors, our job is to delve into every corner of the tech industry, understand its operational context, and anticipate the next evolution. Therefore, every December, we invite each investment team to share their "big idea" that they believe tech entrepreneurs will tackle in the coming year.

Today, we bring you viewpoints from the Infrastructure, Growth, Bio + Health, and Speedrun teams. The opinions of other teams will be released tomorrow, so stay tuned.

Infrastructure Team

Jennifer Li: Startups Will Tame the “Chaos” of Multi-Modal Data

Unstructured, multi-modal data has always been the biggest bottleneck for enterprises and remains the largest untapped resource. Every company is inundated with PDFs, screenshots, videos, logs, emails, and various semi-structured "data mud." Models are becoming increasingly intelligent, but input is becoming more chaotic—leading to the illusion of the RAG system, causing intelligent agents to err subtly and expensively, and keeping critical workflows highly reliant on manual quality checks.

Today, the true limiting factor for AI companies is data entropy: in an unstructured world that holds 80% of a company's knowledge, freshness, structure, and truthfulness continue to decay.

As a result, unraveling the "tangled mess" of unstructured data is becoming a generational entrepreneurial opportunity. Companies need a persistent way to clean, structure, validate, and govern their multi-modal data to make downstream AI workloads truly effective. Application scenarios are everywhere: contract analysis, user onboarding, claims processing, compliance, customer service, procurement, engineering retrieval, sales enablement, analytics pipelines, and all intelligent agent workflows that rely on reliable context.

A platform-based startup that can extract structure from documents, images, and videos, reconcile conflicts, fix data pipelines, and maintain a fresh and searchable data platform will hold the "key to the kingdom" of enterprise knowledge and processes.

Joel de la Garza: AI Will Reshape the Recruiting Dilemma of Security Teams

Over the past decade, the most significant headache for CISOs has been recruiting. From 2013 to 2021, the global cybersecurity workforce gap surged from just under 1 million to 3 million. The reason is that security teams require highly specialized technical talent but have them engage in exhausting Tier 1 security work like log parsing, which few are willing to do.

The deeper root of the problem is that security teams have created their own grunt work. They bought tools for "indiscriminate detection of everything," forcing the team to "review everything," which in turn created artificial "labor scarcity," forming a vicious cycle.

By 2026, AI will break this cycle by automating most repetitive and redundant tasks, significantly narrowing the talent gap. Anyone who has been in a large security team knows that half the work can be automated entirely; the issue is when you are overwhelmed with work daily, you can't step back to think about what should be automated. Truly AI-native tools will do this for security teams, allowing them to finally focus their energy back on what they wanted to do in the first place: track attackers, build systems, and patch vulnerabilities.

Malika Aubakirova: Intelligent-Agent-Native Infrastructure Will Become Standard

In 2026, the most significant infrastructure shake-up won't come from the outside but from within. We are transitioning from "human speed, low concurrency, predictable" traffic to "intelligent agent speed, recursive, bursty, massive" workloads.

Today's enterprise backend is designed for a 1:1 "from human action to system response" model. It's not suited to handle a single-minded intelligent agent triggering 5000 subtasks, database queries, and internal API calls in a millisecond-level recursive storm. When an intelligent agent tries to refactor a codebase or fix security logs, it doesn't behave like a user; for traditional databases or rate limiters, it's more akin to a DDoS attack.

To build systems for 2026's intelligent-agent workloads, the control plane must be redesigned. "Agent-native" infrastructure will begin to emerge. The next-generation systems must consider the "herd effect" as the default state. Cold starts must be shortened, latency spikes must converge, and concurrency limits must be raised by orders of magnitude.

The real bottleneck will shift towards coordination itself: routing, lock control, state management, and policy enforcement in large-scale parallel execution. Only platforms that can survive the deluge of tool invocations will emerge as the ultimate winners.

Justine Moore: Creative Tools Embrace Multimodality

We already have the basic building blocks for AI storytelling: generative sound, music, images, and video. However, as long as the content is more than just a short clip, achieving close to director-level control is still time-consuming, painful, and even impossible.

Why not let a model take a 30-second video, have it create a new character using the reference images and sound we provide, and continue shooting the same scene? Why not have the model "reshoot" from a new perspective or match the action to the reference video?

2026 will be the year when AI truly achieves multimodal creation. Users will be able to feed all forms of reference content to the model to collaboratively generate new works or edit existing scenes.

We have already seen the emergence of first-generation products such as Kling O1 and Runway Aleph, but this is just the beginning—both the model layer and the application layer require new innovation.

Content creation is one of AI's "killer applications," and I anticipate multiple successful products emerging for various user groups—from meme creators to Hollywood directors.

Jason Cui: AI-Native Data Stack Will Continue to Iterate

Over the past year, the "modern data stack" has been visibly integrated. Data companies are transitioning from modular services such as collection, transformation, and computation to bundled and unified platforms (e.g., Fivetran/dbt merger, Databricks expansion).

Despite the ecosystem maturing, we are still in the early stages of a truly AI-native data architecture. We are excited about how AI will continue to reshape multiple aspects of the data stack and are beginning to see irreversible deep integration of data and AI infrastructure.

We are especially focused on the following directions:

How data will continue to flow to high-performance vector databases beyond traditional structured storage

How AI agents will address the "context problem": maintaining continuous access to the correct data semantics and business definitions, enabling applications similar to a "data dialogue" to maintain consistent understanding across multiple systems

As data workflows become more agent-centric and automated, how traditional BI tools and spreadsheets will evolve

Yoko Li: We Will Truly "Step Inside the Video"

By 2026, video will no longer be a passive viewing content but will start to become a place we can "step into." Video models will finally be able to understand time, remember what has been presented, and react to our actions while maintaining a level of realism and coherence closely resembling the real world, rather than just outputting a few seconds of unrelated images.

These systems will be able to sustain characters, objects, and physical laws over extended periods, allowing actions to have real consequences and causality to unfold. Video will thus transition from being a medium to being a space where things can be constructed: robots can train within it, game mechanics can evolve, designers can conduct prototype experiments, and agents can learn by "doing."

The presented world will no longer be like a short video but more like a "living environment," starting to narrow the gap between perception and action. This will be the first time humans can truly "inhabit" the video they have generated themselves.

Growth Team

Sarah Wang: The Role of Enterprise "Record Systems" Will Begin to Shake

2026 will see the true transformation of enterprise software coming from a fundamental shift: the central role of record systems will finally begin to decline.

AI is compressing the distance from "intent" to "execution": models can directly read, write, and reason through enterprise operational data, transforming ITSM, CRM, and other systems from passive databases to self-executing workflow engines.

With the rapid advancement of inference models and intelligent workflow agents, these systems will no longer just respond to demands but will be able to predict, coordinate, and execute end-to-end processes.

Interfaces will become dynamic layers of intelligent agents, with the traditional system of record layer gradually becoming a "cheap persistence storage," relinquishing strategic control to the players who manage the intelligent execution environment.

Alex Immerman: Vertical AI Upgrading from "Information Retrieval and Inference" to "Multiplayer Collaboration Mode"

AI is driving explosive growth in vertical industry software. Companies in healthcare, law, and real estate have quickly surpassed $100M ARR; finance and accounting are closely following suit.

The initial revolution was in information retrieval: finding, extracting, and summarizing information.

2025 brought about breakthroughs: Hebbia parsing financial statements, Basis reconciling trial balances across multiple systems, EliseAI diagnosing maintenance issues and scheduling vendors.

2026 will unlock the "multiplayer mode."

Vertical software inherently has industry-specific interfaces, data, and integration capabilities, and vertical industry work is fundamentally collaborative: buyers, sellers, tenants, consultants, suppliers, each with different permissions, processes, and compliance requirements.

Today, individual AIs operate in silos, leading to chaotic handoffs and lack of authority: an AI analyzing contracts cannot communicate with the CFO's modeling preferences; a maintenance AI is unaware of on-site personnel's commitments to tenants.

Multiplayer-mode AI will break this paradigm: automatically coordinate among parties; maintain context; synchronize changes; automatically route to domain experts; allow adversarial AIs to negotiate within boundaries and flag asymmetries for human review.

As transactions enhance in quality due to "multi-agent + multi-human" collaboration, switching costs will soar — this layer of collaborative network will become the long-missing "moat" of AI applications.

Stephenie Zhang: The Future Canvas Will No Longer Be Human, But Agents

By 2026, people will interact with networks through agents, and human-centric content optimization will lose its significance.

We used to optimize for predictable human behavior: Google rankings; Amazon top items; news articles' 5W+1H and engaging introductions.

Humans may overlook profound insights buried on the fifth page, but agents won't.

Software will also change accordingly. Applications used to be designed for human eyes and clicks, where optimization meant better UI and processes; but as agents take over retrieval and interpretation, the importance of visual design diminishes: engineers no longer stare at Grafana, AI SREs will automatically parse telemetry and provide insights on Slack; sales teams won't manually scroll through CRM, agents will automatically summarize patterns and insights.

We are no longer designing for humans, but for agents. The new optimization is not at the visual level, but at machine readability. This will completely transform content creation methods and toolsets.

Santiago Rodriguez: KPIs Based on "Screen Time" Will Disappear

Over the past 15 years, "screen time" has been the gold standard for measuring product value: Netflix's viewing time; mouse clicks in healthcare systems; minutes users spend on ChatGPT

But in the upcoming era of "outcome-based pricing," screen time will be completely phased out.

We are already seeing hints of this: ChatGPT's DeepResearch queries require almost no screen time but provide immense value; Abridge automatically records doctor-patient conversations and handles follow-up work, allowing doctors to hardly look at the screen; Cursor completes full application development, and engineers are already planning the next phase; Hebbia automatically generates pitch decks from a large number of public documents, allowing investment banking analysts to finally get some sleep

Challenges follow suit: businesses need to find more sophisticated ROI measurement methods—physician satisfaction, developer productivity, analyst well-being, user happiness... all of these are on the rise with AI.

Companies that can tell the clearest ROI story will continue to succeed.

Bio+Health Team

Julie Yoo: "Health MAUs" Becoming the Core User Group

By 2026, a new healthcare user group will take center stage: "Health MAUs" (monthly active users who are healthy and not ill).

Traditional healthcare mainly serves three types of people:

- Sick MAUs: high-cost, cyclical demand

- Sick DAUs: such as those in long-term intensive care

- Healthy YAUs: individuals who rarely seek medical care

Healthy YAUs can transition to Sick MAUs/DAUs at any time, and preventative care could have delayed this transition. However, due to the current "treatment-focused" healthcare system, proactive testing and monitoring are not adequately covered.

The emergence of Health MAUs has changed this structure: they are not ill but are willing to regularly monitor their health status, making them the largest potential group.

We anticipate that AI-native startups and "repackaging" of traditional institutions will join forces to provide periodic health services.

As AI drives down healthcare delivery costs, preventative-focused insurance products emerge, and users are willing to pay for subscription services, "Health MAUs" will become the most promising customer segment for the next generation of health tech—actively engaged, data-driven, and focused on prevention.

Speedrun Team (Gaming, Interactive Media, and World Modeling Direction)

Jon Lai: World Modeling Will Reshape Narrative

By 2026, AI world modeling will fundamentally change storytelling through interactive virtual worlds and the digital economy. Technologies like Marble (World Labs) and Genie 3 (DeepMind) can generate complete 3D worlds from text, allowing users to explore like in a game.

As creators adopt these tools, a new form of narrative will emerge—potentially giving rise to a "generated version of Minecraft," where players collaboratively build a vast, evolving universe.

These worlds will blur the line between players and creators, creating a shared, dynamic reality. Different genres such as fantasy, horror, and adventure can coexist; within them, the digital economy will thrive, enabling creators to earn income through asset creation, player guidance, and interactive tool development.

These generated worlds will also serve as training grounds for AI agents, robots, and even potential AGI. The world modeling not only brings a new category of games but also a new frontier of creative media and economics.

Josh Lu: "My Year"

2026 will be "My Year": products will no longer be mass-produced for the "average consumer" but tailored to "you."

In education, AlphaSchool's AI mentor will match each student's pace and interests.

In health, AI will tailor supplements, exercise plans, and diet programs for you.

In media, AI will remix content in real-time to your taste.

The giants of the past century 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 for "you."

Emily Bennett: The First AI-Native University Will Emerge

In 2026, we will witness the first truly AI-native university—an institution built around intelligent systems from the ground up. Traditional universities have already used 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 a university where courses, guidance, research collaboration, and campus operations are all continuously adjusted in real time based on a feedback loop; schedules self-optimize; reading lists dynamically update as new research emerges; and each student's learning path evolves 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 integrated AI literacy into general education.

In an AI-native university:

- Professors become "learning system architects": designing data, tuning models, and teaching students how to scrutinize machine reasoning.

- Evaluation methods will shift to "AI mindfulness" assessments: not asking whether students used AI, but how they used AI.

With industries in dire need of talent capable of collaborating with intelligent systems, this university will become the "talent engine" of the new economy.

[Original Article Link]

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia