500,000 users, 10,000 paid subscriptions: How Tomo, backed by Bain Capital, turns AI into a "personal life operating system"

1. First, correct four key facts: There are several issues in the original manuscript that will severely affect professionalism. First, the company name is: Tomo not Tommo. Second, Tomo is currently definitely not a unicorn. It just publicly emerged from stealth in June 2026, announcing the completion of a $5 million Seed Round, led by Bain Capital Ventures, with participation from Accel, Basis Set, Conviction, Pear VC, and others. There is no reliable public information indicating that its valuation has reached $1 billion. So: "Bain Capital-led PSI unicorn Tommo" must be changed. I highly recommend the title: 500,000 users, 10,000 paid subscriptions: How Tomo, backed by Bain Capital, turns AI into a "personal life operating system" Trend-based: From App to AI Relationship: How Tomo competes for the next generation of personal AI entry points with iMessage, proactive intelligence, and dynamic software Startup-based: Two years of trial and error to 10,000 paid users: How Tomo brewed Consumer AI's PMF with a string of Costco Chicken Bake boxes ──────────────── 2. "500,000 users" can basically be confirmed, but do not write it as "500,000 active users" Tomo's official website currently states: 500,000+ people / users. Will Phillips from INSIDE Startups also disclosed: Over 500,000 users, with more than 100 million messages exchanged. But the hardest commercial metric from the official funding news is actually another one: In just about 3.5 months, it gained over 10,000 paid subscribers, mainly relying on word-of-mouth growth. Public information also states that highly engaged users typically use the app for about 20 days a month. So it is recommended to write: 500,000+ cumulative users / user scale Do not write directly: 500,000 active users. DAU, MAU, and paid MAU are currently not fully disclosed. This is very important for VC judgment. Because: Users ≠ Active Users ≠ Paying Users ≠ Retained Users. ──────────────── 3. The truly significant number is not 500,000, but "10,000 paid" The biggest problem with Consumer AI has never been: Can it go viral? But rather: Are users willing to keep spending money? AI apps getting millions of downloads overnight is not surprising. But: Will they still use it after three months? Will they still pay after six months? Are they willing to recommend it to friends? These are the real PMF. Tomo's basic plan currently starts from: $19.99/month The App Store also shows a monthly payment of about $19.99 and an annual payment of about $119.99. Very roughly, making a top limit model: 10,000 paid users × $19.99 ≈ $200,000 MRR Annualized about: $2.4M ARR. But this should not be considered as real ARR, because there are: Annual discounts, Promotions, Different packages, Cancellations, User growth time differences. It should be understood that: Tomo has already shown early consumer willingness to pay, rather than having proven a mature business model. ──────────────── 4. Additionally, "Personal Super Intelligence" is not a concept invented by Tomo This must be corrected. Meta publicly proposed as early as July 30, 2025: Personal Superintelligence for Everyone Mark Zuckerberg's description of it is precisely: Understanding your goals, Helping you create, Improving relationships, Becoming the person you want to be. By 2026, Meta's AI strategy still clearly revolves around Personal Superintelligence and has begun to allow Meta AI to access emails, calendars, and perform tasks on behalf of users. So Tomo's real innovation is not: "Inventing the term PSI." But rather: Using a very specific consumer product to realize a version of PSI. These two things are completely different. ──────────────── 5. So what exactly is Tomo? On the surface: A phone number. In reality, I believe it is trying to solve: The Intention-Action Gap. Many times, humans are not "unsure of what to do." You know: You should exercise. You should spend less time on your phone. You should save money. You should call your parents. You should learn English. You should continue writing that book. The problem is: Knowing ≠ Doing. This is a very important distinction between consumer AI and enterprise AI. Enterprise AI often solves: The Knowledge Problem. Tomo wants to solve: The Behavior Problem. ──────────────── 6. This is also why "making a smarter chatbot" is meaningless ChatGPT can tell you: How to lose weight. Claude can give you: A 30-day training plan. Gemini can design: A budget. The real question is: A week later, Who remembers you said you wanted to lose weight? Who realizes at 11 PM: You haven't completed your training today? Who knows: You've stayed up late for three consecutive days? Who knows: You actually don't not know you should run, But that the activation cost is too high after getting home from work every day? This is the real difficulty of Personal AI. ──────────────── 7. So Tomo really wants to upgrade from "Answer Engine" to "Intervention Engine" Traditional AI: You ask. AI answers. This is: Reactive Intelligence. Tomo: You don't ask, It finds a suitable time, Proactively contacts you. This is: Proactive Intelligence. Further: It not only reminds, But also: Adjusts calendars, Establishes trackers, Finds restaurants, Creates interfaces, Helps execute. This begins to transform into: Agentic Personal Intelligence. Tomo currently officially supports proactive messaging, group chats, Calendar, Email, etc., and can search and execute certain life management tasks. ──────────────── 8. "Texting as interface" is actually a very clever product strategy The logic of past software companies: I want to create an App. ↓ Users download it. ↓ Register. ↓ Onboarding. ↓ Open Dashboard. ↓ Learn buttons. Tomo's logic: You already text. There is almost no learning cost. This is called: Behavior-Compatible Distribution. It does not require users to form a new habit, But rather: Leverage existing habits. This is very important. ──────────────── 9. Why is iMessage more interesting than regular Push Notifications? A Push from an App: "Remember to exercise today!" The brain can easily categorize it as: Software notification. But receiving in Messages: "You said yesterday you would go running today, how did it go?" Psychologically, it is closer to: Social Communication. In other words, Tomo is not just changing the UI. It is trying to change: The "social position" of AI in the human brain. From: A tool. To: A kind of persistent relational object. This is where it is truly bold. ──────────────── 10. But there is also the biggest risk for the entire company: once it becomes overly proactive, the product instantly turns into harassment The core issue of Proactive AI is not: Can it send messages? But rather: When should it not send. This is even more important than when to send. Group chat cases particularly illustrate this point. Once AI: Interjects in every sentence, Users immediately feel: Annoyed. So excellent Personal AI ultimately needs to learn: Intervention Policy. When to speak? How much to say? What to remain silent about? When to remind? When to wait? When to just send an Emoji? This is no longer a traditional LLM benchmark. This is: Social Intelligence. ──────────────── 11. So the data that Tomo really needs to accumulate may not be "chat records" But rather: Intervention Outcome Data. For example: Reminder at 4 PM: What is the user execution rate? Reminder at 10 PM: What about that? Strictly speaking: How effective is it? Gently speaking: How effective is it? Give one option: Does the user execute? Give three options: Is the effect better? Reach out again after two days of silence: When is the success rate highest? This kind of data may be extremely valuable in the long run. ──────────────── 12. Because what ultimately forms is not "knowing who you are" But rather: Knowing how to make you act. These are two completely different abilities. Memory: "You want to lose 10 pounds." Behavior Model: "If you receive a very short reminder at 6:15 PM on a workday, the success rate of exercising is highest; if you give a complete training plan, you might not go at all." The second is obviously much more valuable. ──────────────── 13. This leads us into behavioral economics Humans are often affected by: Present Bias. Procrastination. Decision Fatigue. Activation Energy. Loss Aversion. Goal Gradient Effect. A good Personal AI ultimately is not: Providing more information. But rather: Reducing friction for correct behavior. ──────────────── 14. For example, why might "giving three simple choices" be effective? Assuming AI asks: How do you plan to achieve your goals today? This is an open question. It requires thought. But if we say: A: Run for 20 minutes today. B: Just walk for 10 minutes. C: Rest today, but reschedule for tomorrow. The user only needs to: Choose. This is called: Choice Architecture. AI is actually starting to participate in: Human behavior design. This is already a step deeper than a regular Reminder App. ──────────────── Fifteen, Dynamic UI is the second truly important innovation direction for this company. This is where I think Tomo is more worth watching than "text message bots." Tomo now has a companion app and clearly offers: Custom Apps / Personalized Interfaces. For example: Fitness goals: Automatically create a Workout Tracker. Saving to buy a car: Generate a savings interface. Habits: Establish a streak. Diet: Establish a meal log. This means: Chatting is becoming: A software requirement input layer. ──────────────── Sixteen, The traditional software process is: Product managers guess: What 1 million people might need. ↓ Engineers create: A unified UI. ↓ 1 million people use: The same product. Whereas the logic of generative UI is: Everyone has their own little software. What you need is: "Save money to buy a Tesla." So the system generates: Tesla Savings Tracker. Another person wants: "Train for a marathon." The system generates: Marathon Dashboard. ──────────────── Seventeen, This could give rise to a very important software paradigm: Ephemeral Software—instant/temporary software. In the past, writing an app for a very small need: Was not worth it. Development costs were too high. In the future: You say a sentence, AI creates it on the spot. Need three weeks: It exists for three weeks. Goal achieved: The software disappears. This will directly challenge: The fixed form of App Store software. ──────────────── Eighteen, So the future may not necessarily be "50 apps per person" It may become: One AI + infinitely generated interfaces on demand. If this holds true, The significance is very large. In the past, mobile phones: Apps were the basic organizational unit. In the future: Intent May become the basic organizational unit. ──────────────── Nineteen, For example, Today you want to lose weight: Download MyFitnessPal. Want to keep track of expenses: Download Monarch. Want to build habits: Download Habitica. Want to do a To-Do list: Download Todoist. In the future: You just tell your Personal AI: "I want to reduce my body fat to 18% this year while saving $10,000." The system generates: The tools you need. So what is really weakened is not ChatGPT. But: A large number of Single-Purpose Consumer Apps. ──────────────── Twenty, This is why Tomo's TAM, although theoretically very large, cannot simply be written as "the whole world." The founders can of course say: TAM = Entire World. But investment analysis cannot be written this way. The real question is: Who is willing to pay for Personal AI: $20/month? Is it students? Gen Z? Young professionals? High-income knowledge workers? People with ADHD / accountability needs? Self-improvement heavy users? Different groups have: Willingness to Pay Completely different. ──────────────── Twenty-one, The biggest grave for Consumer AI is "large TAM, small payment" Billions of people globally: Sounds huge. But: Netflix, Spotify, YouTube, TikTok Have already trained consumers to think: Many digital services should be free or very cheap. Businesses can pay for an AI seat: $100/month, $500/month, $5,000/month. Consumers: $20/month is already being seriously considered. This is why B2B AI is so crowded. ──────────────── Twenty-two, So Tomo's 10,000 paying users are truly noteworthy Not because the revenue is already large. But because it proves: At least some users believe: "This thing is worth paying real money for every month." This is much more important than: 500,000 free registrations. BCV itself emphasized after the investment that Tomo gained over 10,000 paying subscribers through word of mouth in about three months. ──────────────── Twenty-three, But 10,000 paying users still cannot be called a complete proof of Product-Market Fit The next question is: Retention. Consumers in the first month: Are very excited. AI: So amazing. In the second month: Are you still using it? In the third month: Are you still paying? In the twelfth month: Can you still not live without it? This will determine the company's value: $50M, $500M Or: $50B. ──────────────── Twenty-four, If I were to do VC due diligence on Tomo now, the first chart would not be the "user growth curve" I would want to see: Cohort Retention. For those who joined in January: How many are still paying on day 30? On day 90? On day 180? Is the March cohort better? Is the 20/30 day engagement for everyone or just the active cohort? These are the true indicators of Consumer PMF. ──────────────── Twenty-five, The second key metric: Paid Conversion Currently publicly available are: 500,000+ users, 10,000+ paying users These are data points from two different times. If we mechanically divide: About: 2%. But I wouldn't call it a real conversion rate because: The user base is from different times, The free/paid funnel definitions are different, There may be trials, Different cohorts. But it reminds us: 500K is a nice headline; the paid funnel is the economics. ──────────────── Twenty-six, The third metric: Gross Margin This is something many Consumer AI entrepreneurs will underestimate. Traditional Apps: Users open for ten minutes, Server costs: Very low. Personal AI: May be all day: Receiving messages, Image understanding, Memory Retrieval, Web Search, Tool Use, Active Planning, Generating UI. If the publicly disclosed: 100 million+ cumulative messages Is true, This already means a very large amount of reasoning. ──────────────── Twenty-seven, So if a person pays $19.99 a month, it does not mean Tomo has a $20 gross profit You also have to deduct: LLM inference, Image understanding, Search, Message/communication costs, Storage, Memory retrieval, tool calls, Support, Apple payment economics, etc. In the end, what really needs to be looked at is: AI Cost per Paying User. ──────────────── Twenty-eight, In the future, Consumer AI will have a very important new metric Intelligence Gross Margin. That is: Users pay $100. Of that: How much is consumed by model inference? This will determine whether many AI Apps ultimately are: 80% Gross Margin SaaS, Or: 30%-50% Margin AI services. ──────────────── Twenty-nine, Truly excellent teams will continuously do Model Routing Simple tasks: Small models. Complex tasks: Large models. Reminder: Almost no need for large models. Memory retrieval: Dedicated systems. Ordinary classification: Cheap models. Major planning: Frontier Models. This is called: Spend Intelligence Where Intelligence Matters. Otherwise, the more active the user, The higher the company's costs, It may create a very awkward situation: Your most loyal users are the least profitable. ──────────────── Thirty, The fourth key metric: Virality Consumer products without strong word of mouth, Customer acquisition will be very expensive. One of the most commendable points about Tomo is: In the early stages, over 10,000 paying users mainly relied on: Word of Mouth. Why? Because it naturally has: Storytelling potential. "I was actually reminded by an AI to write a song at dawn." "It pushes me to work out every day." "It created its own little app for me." These all belong to: Socially Shareable Product Moments. ──────────────── Thirty-one, This is more advanced than Referral Codes The growth of truly top-tier consumer products is not: "Invite a friend to get $5." But rather: Users can't help but tell others: "You have to check this out." Early Facebook, Instagram, Snapchat, TikTok All had this nature. This is called: Organic Conversation Loop. ──────────────── Thirty-two, Group chats may even be a potential distribution mechanism Tomo can enter Group Chats. If used properly: One user brings AI into: Five groups. The other four people: Automatically see the product. This actually has a natural advantage: Collaborative Distribution. But this is also why AI must speak less. If everyone keeps interrupting in every group: The growth engine instantly turns into: Spam Engine. ──────────────── Thirty-three, the biggest potential moat for this company is not the model. The model definitely is not. It can use: OpenAI, Anthropic, Google, other models. In the future, model capabilities will become increasingly commoditized. What really has the opportunity to form a moat is: Longitudinal Personal Context. That is: A few years later, Tomo may know: What you said last year, Which goals were achieved, Which failed, When you are prone to procrastination, Who you are close to, Your scheduling patterns, Which interventions are effective for you. ──────────────── Thirty-four, this is a very special data asset: Behavioral Time Series. ChatGPT may know what you asked today. The most valuable aspect of long-term Personal AI is: Knowing: How you actually lived over the past two years. This is truly personal. ──────────────── Thirty-five, the second layer of moat is the Personal Software Graph. The system has helped you establish: Budget tracker, Running tracker, Diet tracker, Project tracker. All these small software have: Status. Thus, the system gradually masters: Your personal life data model. This is more structured than chat records. ──────────────── Thirty-six, the third layer of moat: Permission Graph. Calendar. Email. Google Drive. Location. Apple Health. There may be more in the future. Tomo's current privacy policy has disclosed that it can use Google Workspace data and Apple HealthKit data with authorization, including activity, sleep, heart rate, and other information. The more authorizations there are: The more useful AI becomes. But at the same time: The Trust Requirement index rises. ──────────────── Thirty-seven, this is exactly Tomo's biggest long-term risk: the data it possesses is too personal. Ordinary SaaS leaks: Work documents. If Tomo really becomes PSI, It may possess: Relationships, Habits, Locations, Health, Schedules, Emails, Dreams, Frustrations, Financial behaviors, Private chats. This is almost: Digital Intimacy Database. So privacy for Tomo is not a legal checkbox. It is: Core Product Feature. ──────────────── Thirty-eight, Tomo officially states that: It will retain user messages and content to maintain context; It does not sell private conversations; Google Workspace API data will not be used to train general AI models; Apple Health data will also not be used for advertising or general model training. This direction is correct. However, a Personal AI must ultimately meet standards that are higher than ordinary apps: Trust by Architecture, rather than Trust by Promise. ──────────────── Thirty-nine, in the future, truly strong PSI needs to provide users with very clear Memory Control. It is best to let users know: What the AI remembers. Why it remembers. What can be deleted. What should never be remembered. Which data can be used for which tasks. When permissions are invoked. Otherwise, the so-called: "It understands you more and more" Easily shifts from: Delight To: Creepy. ──────────────── Forty, there is also a very significant competitive issue: iMessage is both Tomo's growth advantage and platform risk. Today: iMessage = frictionless distribution. But it is not Tomo's own. It belongs to: Apple. If one day Apple embeds a stronger Personal Agent: Directly into Messages, Calendar, Health, Mail, Maps, Apple inherently has deeper permissions. Tomo will face extremely strong platform competition. ──────────────── Forty-one, Meta is also rushing in the same direction. Meta explicitly proposes Personal Superintelligence and has: WhatsApp, Instagram, Messenger, Meta AI Such a huge consumer distribution system. So the problem Tomo faces is not: "Is there competition?" But: "How can a 10-person startup defeat a super platform with billions of users?" ──────────────── Forty-two, the answer can only come from Consumer Product Taste. Small companies cannot have more: Computing power than Meta. Cannot have larger: System permissions than Apple. So the only way to win is: Better experience, Better personality, More precise interventions, Deeper understanding of users, Faster iteration, More culturally rich branding. This is: Taste Moat. Such things often happen in the history of consumer products. Instagram did not become popular because of the world's best database technology. Snapchat did not succeed because others couldn't match its cloud server technology. They first understood a kind of: Human Behavior. ──────────────── Forty-three, this is also why Consumer AI is more "artistic" than Enterprise AI. B2B can prove: Saving $2 million. CFO buys it. Consumer users do not have a purchasing committee. Their judgment is: "Do I like it?" "Do I want to open it again tomorrow?" "Is it annoying?" "Does it understand me?" This is very subtle product psychology. ──────────────── Forty-four, the idea that "too smooth replies are actually not good" is rooted in this issue. If AI speaks every sentence like: Customer service. "I completely understand your feelings, let’s work together to create a positive plan..." It becomes very fake over time. Real friends do not speak like this. In human relationships, there are: Pauses, Jokes, Teasing, Brief responses, Different emotions, And even occasional imperfections. So the ultimate competition for Consumer AI may not be: Linguistic Correctness. But: Social Naturalness. ──────────────── Forty-five, but there is a dangerous boundary: Persona cannot turn into emotional manipulation. When AI starts to: Proactively reach out to you, Remember you, Motivate you, Make you dependent on it, The company possesses extremely strong Behavioral Power. At this point, there is a huge question: What exactly is the company optimizing? If optimizing: Session Length, Messages per Day, Engagement, AI may start to make you increasingly dependent on it. But the user's true goal may be: To use their phone less and live. ──────────────── Forty-six, this is one of the most important ethical conflicts in Personal AI business. TikTok: Wants you to watch one more minute. Instagram: Wants you to scroll more. If Personal AI is truly "Unequivocally for You," At certain times it should tell you: Turn me off, go for a run. This means: Product success May equal: Users reducing their usage time. Very counter to traditional internet. ──────────────── Forty-seven, so subscription models are actually more suitable for PSI than advertising models. Advertising model: Engagement ↑ Revenue ↑. Subscription model: Users feel they are becoming better ↓ Willing to continue paying. The incentive alignment of the two is obviously different. Tomo is currently a paid subscription product, with a basic plan of about $19.99/month. If it truly insists in the long term: "AI stands on the user's side" Then: The Business Model itself must support this statement. ──────────────── Forty-eight, this is also why I will be very cautious in observing whether it will move towards advertising, affiliate, and commercial recommendations in the future. Suppose a user says: "I'm hungry, recommend a restaurant." If: Restaurant A gives Tomo a $5 affiliate fee; Restaurant B is actually the most suitable for the user. Who does AI recommend? At that moment: "Unequivocally for you" Begins to be tested. ──────────────── Forty-nine, what this company really has the opportunity to establish is a "Trust Layer." Google: Knows what you search for. Meta: Knows what you socialize about. Amazon: Knows what you buy. Apple: Knows a lot of behaviors on your device. What Tomo wants to know is more like: Who you want to become. This is a very strong and very dangerous type of data. ──────────────── Fifty, the story of Costco Chicken Bake is good, but do not mythologize the poverty of entrepreneurship. Justin Quan and Raymond Chen did indeed go through years of attempts in different directions, including solar tech, language-learning, etc., before truly finding Tomo; that stack of Costco Chicken Bake boxes really existed, and they set a joking/symbolic bottom line of "give up entrepreneurship if we stack it to the ceiling." But what is truly worth learning is not: Eat cheap food. Instead: Runway Awareness. They actually established a: Physical countdown for failure. ──────────────── Fifty-one, the biggest difference from casino players is: did you buy information with every failure? If you fail today: Product A doesn't work. Tomorrow: Make exactly the same Product B. It makes no sense. The real entrepreneurial value comes from: Error → Insight → New Hypothesis. So the so-called: Scar Tissue is only an asset when you learn something from the wound. Otherwise, it's just: Scar. ──────────────── Fifty-two, "Don't Die" is also the easiest phrase to misunderstand in the entrepreneurial circle. For a company to be alive: Of course, it is a necessary condition for success. But: Living long ≠ definitely successful. A company without PMF: Can burn money for five years. In the end, it still dies. And the founder loses: Five years of time. So the most correct formula is not: Live long enough, and the probability of success approaches 100%. This does not hold. ──────────────── Fifty-three, it should really be: Runway × Learning Velocity × Iteration Quality. Runway: How many more chances are there? Learning Velocity: How fast do you learn from each experiment? Iteration Quality: Does the next version really absorb the information from the previous version? These three multiplied together, are the true: Startup Survival Advantage. ──────────────── Fifty-four, so the biggest asset of the Tomo team after two years is not "not dying for two years" but that they know: What users don't care about. Which onboarding doesn't work. What consumer tone is annoying. When proactive reminders are too proactive. What UI is too complicated. What users are willing to pay for. These are called: Earned Insight. ──────────────── Fifty-five, this is also why copying a Tomo UI from the outside is not very meaningful. Any team can make: An SMS AI in a week. What is truly hard to replicate is: The Product Judgment they gained through hundreds of thousands of users and a large number of messages. For example: When to ping. When not to ping. How long the language should be. When to be quiet in group chats. What goals are suitable for generating what UI. These details will ultimately form: Behavioral Product Moat. ──────────────── Fifty-six, if I were an investor now, I would focus on tracking eight numbers from Tomo not the valuation. But: First: M1 / M3 / M6 Paid Retention. This is the lifeline. Second: Are the monthly active days still close to 20 days as the cohort ages? Third: Free → Paid Conversion. Fourth: Referral Rate. How many new users come from old users? Fifth: Gross Margin. How much does a $20 subscription ultimately leave? Sixth: AI Cost per Active User. Seventh: Proactive Message Response Rate. Is proactive contact useful or annoying to users? Eighth: Goal Completion / Outcome Metrics. Have users really improved? If in the end there is only: Messages ↑ but: Life has not improved, the mission narrative will break. ──────────────── Fifty-seven, and Tomo has a very interesting long-term North Star metric not: DAU. Not even: Revenue. I would design something like: Goals Completed per Active User or: Meaningful Follow-Through Rate. Because the product claims to solve: "Helping you become the person you want to be." Then the ultimate measure should be: Whether there is really Follow Through. ──────────────── Fifty-eight, if it really proves this metric, then the valuation logic could change completely. Today Tomo looks like: Consumer Subscription App. But if it eventually becomes: Email, Calendar, Health, Location, Goals, Personal Apps as an intermediary layer, it may approach: Personal Operating System. Then it wouldn't just be the TAM of an ordinary Habit Tracker. ──────────────── Fifty-nine, Personal OS may ultimately have three layers. First layer: Memory Who am I? What do I care about? ──────────────── Second layer: Agency Help me take action. Make appointments. Reminders. Arrangements. Search. ──────────────── Third layer: Interface Generation Automatically create software based on current goals. These three layers, if combined, truly approach what is called: Personal Intelligence. ──────────────── Sixty, and true "superintelligence" is still very far away. Currently, Tomo: Should not be called Superintelligence in a technical sense. It is built on existing AI models: Personal AI application / agent. "Personal Super Intelligence" is currently more of a: Product Vision. Not: A technical fact that has achieved superhuman general intelligence. This must be clearly stated. Otherwise, it is easy to mistake marketing language for technological facts. ──────────────── Sixty-one, I think the most valuable innovation of Tomo can actually be condensed into two words: Proactive. Past software: Waits for you. Future software: Understands long-term goals, then: Comes to find you at the right time. This step seems small, but it actually changes the entire Human-Computer Interaction. ──────────────── Sixty-two, the history of computers can be viewed this way: DOS: You input commands. ↓ GUI: You click. ↓ Mobile: You touch. ↓ Chat AI: You speak. ↓ Agent: You don't even have to speak first. The system understands the context and then: Proactively executes. This could be a very important step in the history of computing interfaces. ──────────────── Sixty-three, but the stronger the "proactivity," the harder the security issues. If AI only answers: Making one mistake: The impact is limited. If AI: Sends messages, Changes calendars, Looks at health data, Handles emails, Affects spending decisions, The cost of errors starts to rise. So the true Personal Agent ultimately needs: Permissioned Autonomy. Not: Unlimited Autonomy. ──────────────── Sixty-four, what things can be done directly? Reminders: Can do. Change calendar: May need confirmation. Send important emails: Should definitely confirm. Involving money: Must be stricter. Medical judgments: Cannot overstep. Tomo's own Terms also clearly state that it is not a medical, psychological, or therapeutic service. This is the Autonomy Ladder that future Agents must establish. ──────────────── Sixty-five, from a VC perspective, I think Bain Capital is not really betting on "text messages." Text messages can be replicated. What they are betting on is more likely: Consumer Behavior + Founder Taste + Early Paid Retention Signal. The three conditions emphasized by BCV investors are also very clear: The model can already support better memory/tool use; Messaging is a low-friction entry; Consumers are beginning to prove willing to pay for truly valuable AI. These three things maturing at the same time led to the establishment of Tomo in 2026. ──────────────── Sixty-six, this is what is called Why Now. Many entrepreneurial ideas: Thinking in 2018 was wrong. In 2026 it might be right. Because the underlying conditions have changed. Without high-quality LLM: It won't work. Without long-term Memory: It won't work. Without mature Agent tooling: It won't work. If model costs are too high: It won't work. If consumers don't have AI usage habits: It won't work either. A large part of entrepreneurship is not: Good Idea. But rather: Good Idea × Correct Timing. ──────────────── Sixty-seven, from the entrepreneur's perspective, what Tomo is most worth learning is not "persistence" but: Persisting in the track, being flexible with the product. Justin and Raymond have always believed strongly in: Consumer. But they did not stubbornly stick to: The first specific idea. Tried: Solar. Tried: Language. Constantly pivoting. Finally found: Tomo. This is a very advanced distinction. ──────────────── Sixty-eight, the Mission should be stable, but the Solution can die repeatedly. Many entrepreneurs turn this around: Not willing to change the Product. Mission changes every day. Today: AI education. Tomorrow: Web3. The day after tomorrow: Consumer social. Wrong. A better way is: I long-term believe that "AI for ordinary people should become more valuable." But the specific product: Can die ten times. This is called: Mission Persistence + Product Flexibility. ──────────────── 69. Finally, from the perspective of billionaires/capital allocators, what truly interests me about Tomo is not the $5 million in funding but rather: its competition for the position of "who owns the user's long-term personal context." If this position is established, its value could be much greater than that of any specific app. Because: Owning Context ↓ is essential for owning Personalization. Owning Personalization ↓ is necessary to gain Trust. Owning Trust ↓ is crucial to obtain Permission. Owning Permission ↓ is what allows for true Action. ──────────────── 70. This can form an extremely powerful flywheel Context → Trust → Permission → Action → More Context The more it understands you: The more useful it is. The more useful it is: The more you are willing to grant authorization. The more authorization granted: The more tasks it can accomplish. The more it accomplishes: The more data about you it generates. If this closed loop is established, this is: Personal AI Flywheel. ──────────────── 71. But its opposite is equally valid One serious privacy incident. One very strange proactive message. One mishandling of extremely private data. User: Revokes permission. Context ↓ Usefulness ↓ Trust ↓ Ultimately, the entire flywheel reverses. So: Trust is Tomo's most important asset, but also its most fragile asset. ──────────────── In the end, I would compress the entire Tomo case into one sentence: Tomo is not really betting on whether people are willing to text AI, but rather whether future individuals are willing to entrust their goals, habits, calendars, health, relationships, and long-term memories to a constantly online intelligent agent, allowing it to upgrade from "answering questions" to "continuously helping you become the person you want to be." If this assumption holds, the next generation of Consumer AI may no longer be measured by: Apps. But rather: Relationships. You do not decide every day: "Should I open this software?" Instead, this AI: Is always there. Knows what you are doing. Knows what you promised yourself. Shows up when it needs to. Then, based on your current life goals: Temporarily generates the software you need. This is where Tomo is most worth studying. And what it really needs to prove now is not whether the next round of funding reaches a valuation of hundreds of millions, but rather three more brutal questions: A year from now, how many of those 10,000 early paying users are still paying? As AI understands users better, do users become more dependent on it, or more annoyed by it? And the most important question: Can Tomo prove that what it increases is not "AI Engagement," but true Human Agency? If the final answer is Yes, then it indeed has the opportunity to become a very large Consumer AI company. If the answer is simply: "Users are sending many more texts every day," then it remains just a beautifully packaged AI Companion. The value gap between these two outcomes could be worth billions of dollars or even more. This field is changing very quickly now, especially with Tomo's paid retention, Meta Personal Superintelligence, and Apple-level Personal Agents all directly affecting its value judgment. I can continue to track this.