Will the A.I. Assistant Wars Produce a SuperApp?
Observer ·
The race to build an A.I. assistant for everything is accelerating, but consumer habits and the challenges of real-world execution could make a single superapp unlikely. The next opportunity, argues Premise Ventures’ Mercedes Bent, may lie in specialized agents that can reliably complete tasks and work together.
Meta, OpenAI and a growing field of startups are racing to move A.I. beyond chatbots and into everyday life—on our phones, in our text messages. Products such as Meta’s Muse, OpenAI’s Dots, Wajo’s Fo, Town, Hark and Instinct are designed to help users manage email, make purchases, plan travel and complete tasks through familiar messaging interfaces. Their ambition is to become the layer that people rely on to complete nearly everything on their to-do lists, creating a new category called “SMS-based A.I. assistants.”
The market is already attracting significant attention, growing from non-existent to millions of users in a matter of weeks. Meta’s Muse, which launched in September, reached 10.9 million downloads in its first 40 days, reaching approximately 2.5 million daily active users. Those early numbers suggest that customers are willing to try A.I. tools that promise to act on their behalf, not just generate answers.
But I don’t think this is the moment when one A.I. SuperApp takes over everything. More likely, the market will fragment into specialized categories, consolidate around a few leaders in each and eventually develop an orchestration layer that coordinates different agents. Here’s why.
No one actually wants a super app, at least in the U.S. American consumers have spent decades building digital lives around specialized products. We use separate apps for messaging, managing money, booking travel and doing our jobs. Even when one platform can handle multiple tasks, we often prefer the tool that does a particular job best.
Specialization is as core to the American psyche as our Protestant roots and the Industrial Revolution. The logic is familiar from economics. In The Wealth of Nations , Adam Smith , credited with founding modern economics, argued that the division of labor could dramatically increase productivity by breaking work into specialized tasks. Software has largely followed the same pattern wherein products win by solving particular problems well, then build durable customer relationships around those use cases.
A.I. is already showing similar signs of specialization. People use Claude for coding, ChatGPT for general questions and other tools for domain-specific research or workflow automation. Power users are willing to pay for multiple subscriptions when each product offers something the others don’t. A.I. won’t necessarily erase this preference. As underlying models become more interchangeable, the advantage of an application increasingly comes from specialized workflows, integrations, user experience and proprietary data. The model matters, but so does how effectively a product applies it to a particular task.
Most software categories eventually consolidate around a handful of dominant players. The third-place competitor often struggles to win and retain customers. We should expect a similar pattern for the A.I. SMS-based assistant space: fragmentation across categories, consolidation within them. Leadership in one category, though, doesn’t guarantee leadership in another. A company that dominates coding assistance may not be the best choice for travel planning or managing a busy inbox.
The same pattern could emerge among A.I. assistants that operate through text messages and other lightweight interfaces. Some will specialize in email and online tasks, while others focus on routines, purchasing, travel or coordinating across connected apps. A few may become dominant within their respective categories while coexisting with competitors that solve different problems.
The early field is crowded, with dozens of competitors vying for the top spot. Assistant Benchmark, which tests A.I. assistants on every tasks such as booking travel, managing schedules and making purchases, currently ranks Meta’s Muse first, followed by Hark and Instinct, with OpenAI’s Dots, Town, and Pally also among leading contenders. Its rankings offer a snapshot of a rapidly evolving market, rather than a definitive forecast. Products are tested on different subsets of tasks, and their relative performance can change as the tools improve. That uncertainty is itself revealing. The market is still working out what people want an A.I. assistant to do, which interfaces they prefer and which products can reliably carry out tasks beyond simply recommending what to do next.
One big question is what opportunities do SMS-based A.I. assistants create that browser and app-based assistants do not? Messaging-based assistants can make A.I. useful in the moments when opening an app, navigating a website or coordinating several services feels like too much work. Based on how I’m using these tools, I’m seeing three key categories stand out.
Handling tasks on the go, in-transit. I recently used an agent to plan a theme and order supplies for my daughter’s second birthday party. After connecting my Amazon account, I asked the assistant to send me a screenshot for approval before completing the purchase. I did all of this throughout the morning, while getting ready in the morning, at red lights when stopped in traffic and while picking the dog up from daycare. The appeal is that an A.I. assistant can carry a task from idea to a proposed purchase, with the user retaining control over the final decision.
Coordinating multistep plans. Similar to the birthday party, travel and dining are natural use cases for the SMS-based assistants. These require many small logistical details and hand-off coordination points, including comparing options, checking schedules, making reservations and coordinating with other people. An assistant that can manage those steps across services may offer more value than one that produces a list of recommendations.
Ordering services that require a phone call. If you’re like me, waiting on hold or calling a business during its operating hours can turn a simple errand into a time-consuming chore. I recently had an agent call United Airlines to cancel a flight and request a refund, saving me about 30 minutes on the phone. On another occasion, I had it call my nail salon to book an appointment. Neither task required sophisticated reasoning, but both involved navigating a process that would otherwise have taken my time.
There are still a number of bottlenecks holding back this burgeoning category. Real-world execution requires access to users’ accounts, appropriate permissions, secure payment capabilities and a way to handle exceptions when a website, business or booking process doesn’t behave as expected. These are not glamorous problems, but they determine whether an agent finishes a task or merely tells a user what to do. An assistant might identify the right flight but fail at checkout, find an available appointment but be unable to confirm it or prepare a purchase without knowing when it should ask for approval.
If these SMS-based assistants can solve these obstacles, it creates an opportunity for startups beyond competing directly with the biggest A.I. labs. Companies that solve the messy last mile—connecting agents to services, managing permissions and payments and reliably completing transactions—could become essential infrastructure for the broader ecosystem.
As an investor, I am always on the hunt for new categories that are being created, and the leaders who champion the category into a recognizable new modality. A.I. assistants are still early enough that the leading business models and interfaces are not settled. Over the next year, we will get to find out which assistants can turn initial curiosity into habitual use, which categories can support durable businesses and who will ultimately win this space.