Y Combinator Spring 2026 Batch Sees Native Mobile Apps Drop to 4%, Founders Shift Focus to AI and Agents
In Y Combinator's Spring 2026 batch, only 4% of companies consider native mobile apps as their main product, down from 15% in 2013. Founders have shifted their main focus to artificial intelligence and agents.
Bloomberg columnist Parmy Olson notes that the app economy is being dismantled by the same wave of technology: on one hand, the number of new apps in the Apple App Store surged by about 80% earlier this year due to "ambient programming," while on the other hand, the most promising startups are actively moving away from mobile screens. The public list for the Spring batch includes about 194 companies, with enterprise services accounting for about 61% (119 companies), leaning towards infrastructure and tools rather than end-user applications; consumer products number around 13, making up about 6%, fluctuating between 4% and 10% in the last five batches without a rebound.
The demand within the batch has shifted to agents. Many companies are selling phone numbers, payments, identities, memory, sandboxes, and insurance, treating agents as new clients rather than just providing buttons and forms. Analysis breaks down the 194 companies into approximately 55 focused on agent infrastructure, about 56 on AI labor, around 44 on care and capital, and about 39 on AI interfacing with the physical world. About half of the companies mention AI in their one-sentence introductions, over a third reference agents or agent-like services; about 13% are solo founders. Geographically, the vast majority have moved to San Francisco, with about 91% registered as U.S. companies.
The same filtering logic continues into adjacent batches. In the Winter 2026 batch of nearly 190 companies, about 64% are enterprise services, around 5% are consumer-facing, and nearly 60% have been AI-native from day one, with 14 companies already surpassing $1 million in annual revenue as of the demo day. Among the 208 companies sampled from the 2026 directory, about 69.7% carry AI labels, with a median team size of two; the number of robotics and physical AI companies has equaled that of AI infrastructure, with both at 32 companies. The industrial category for Summer 2026 has risen to about 23%, while software as a service has dropped to about 12%.
The app store itself is also pushing startups out. Y Combinator's public policy head points out that reviews often take more than two weeks, and system-level interfaces are not open to third-party assistants, forcing some companies to pivot to USB-C hardware to execute multi-step tasks on desktops. Agents can now directly click interfaces, fill out forms, and bypass CAPTCHAs on Android; if apps cannot expose core capabilities as interfaces, they will be bypassed.
Buyers include enterprise clients purchasing identities, sandboxes, payments, and oversight tools for agents, as well as seed funding that continues to write checks for Y Combinator batches; sellers are founders no longer viewing the app store as the main distribution channel. Funding is shifting from consumer apps measured by downloads and in-app purchases to enterprise workflows measured by seats, tasks, and agent calls. Beneficiaries are model companies and agent infrastructure, while native apps that rely on store revenue and screen time are under pressure—store listings are increasing, but the main product forms chosen by incubators are withdrawing from mobile.
Source: Public information
ABAB AI Insight
Y Combinator previously sent Instagram, Dropbox mobile, and a wave of local lifestyle apps into batches around 2011 to 2013, when native apps equated to customer acquisition channels. Thirteen years later, the same screening table has nearly eliminated this category, not because apps cannot be made, but because making apps no longer proves you have captured the next tier of customers. Garry Tan cites examples of teams of a dozen achieving annual revenues in the tens of millions to over a hundred million, rendering the old path of "first make an app, then raise funds and expand" obsolete.
Capital follows customer forms. Users want icons, push notifications, and store ratings; agents want APIs, machine-readable documentation, programmable access, and auditable identities. Thus, checks are shifting from consumer subscriptions to agent supply chains: whoever issues commands to agents, collects payments, insures, and enters sandboxes will reap new tolls. Apple and Google still collect store taxes, but Y Combinator batches have voted with their feet, rewriting distribution from the two stores to model context protocols and command lines. The median two-person teams can sustain themselves because coding and customer service have been absorbed by models, eliminating the need to maintain a separate client for each operating system.
The historical parallel is the switch from web to mobile. In 2013, not making native apps was seen as a lack of understanding of distribution; in 2026, making the main product a native app instead appears as if one is still optimizing for the previous generation's entry points. Uber Eats' logistics network will remain, while the ordering interface may become an agent-callable API; assistants like Blue selling hardware to circumvent system blocks indicate that old operating systems are using review queues to protect their default assistants. The industry is not in an expansion phase for consumer apps but is reconstructing the software economy into "result delivery" and "tools for machines."
This is technological substitution. What is being replaced is the human-machine interface layer, not the demand itself. The mechanism is that models turn "button-clicking users" into scriptable clients, while store reviews and system interface blocks raise the cost of creating human-machine interfaces, simultaneously allowing platforms' own assistants to capture the benefits of such interfaces. Thus, the cheapest entrepreneurial action becomes bypassing screens and directly linking value to ports callable by agents. The 4% does not indicate that apps are dead, but that apps are no longer considered the default shell for entrepreneurship.