After a trip to Silicon Valley, Timur Belikov gathered his observations on robots, venture money, and companies that get bought for their accumulated data. They turned into a conversation about why people need to be needed, whether it’s possible to forget how to communicate, and why, at the very center of the tech world, it still matters that people know you personally.
What phrase from the Valley would you bring back instead of a souvenir?
“Venture capital has three secrets, but nobody knows them.” I heard it from a managing partner at one of the venture funds in the U.S. I liked that someone who invests in the future for a living reserves the right not to know it. The Valley has no shortage of confidence, but this joke helps keep a sense of perspective about how far your own knowledge really goes.
Still, what do people there believe in right now?
Physical AI. During my trip, there was enormous interest in startups that enable artificial intelligence to act in the physical world. According to the people I spoke with, the hype has been especially intense over the past three months. Robots, devices with built-in portable “neural brains,” the interaction of AI with the environment — that’s where a lot of attention is going.
I think the next big step is tied to AI-native devices. They will be designed around AI from the very start. AI will determine how the device understands a person and what it can actually do. I’d like to see what kinds of things come out of it if design doesn’t start with the familiar screen and set of buttons.
Do developers already see the limit of what models can do?
One Anthropic developer told me: “We’re still very far from the limit of what AI models can do.” That exact phrasing stuck with me. When you hear it from someone who works on the models, you start to look differently at your own three-to-five-year plans. They may well have a limitation baked in that will soon be gone.
And has anyone on the inside already seen consciousness in AI?
I made a point of asking this question to people at OpenAI and Anthropic who work in development. I asked about self-awareness, too, and about a will of its own, similar to a human one. The people I talked to said they haven’t observed anything like that YET. They explained the stories about “runaway” agents by the way the task had been framed and the constraints in place: the program found a way to act that the developers hadn’t anticipated.
This is a retelling of my conversations. I wouldn’t pass it off as the companies’ official position or as a final answer to the question of consciousness.
But in Anthropic’s experiments, models sometimes violated even direct prohibitions. A technical spec alone won’t solve the problem. So I would separate two questions. Does the system have experiences of its own? And can we reliably control its behavior? The absence of proven consciousness doesn’t make the second question any less serious. A machine doesn’t need to want something the way a human does for the results of its actions to turn out harmful to us.
While we’re discussing machine consciousness, what’s happening to our own?
Little by little, we’re handing over to it work we used to take part in ourselves. A person gets an email and asks AI to read it and draft a reply. On the other end, a person does the same thing. Assignments to employees can also be written with AI, and the reports on those assignments can be prepared with AI, too.
One thought really stuck with me: people are becoming meat proxies between two AI bots. One wrote it, the other read it, and we hit “Send.”
On the bright side, everyone’s busy.
And the correspondence looks flawless. The question is whether you still understand what exactly you’re sending under your own name. Saving twenty minutes on an email is great. But ideally, you should still have your own judgment about what it says.
Then what profession will be left for people?
I’m thinking about a less pleasant scenario. Let’s say that in three to five years, a model trained for a specific task will be able to do almost everything a person of average ability does: write poetry, make films, solve research problems. I won’t presume to put an exact date on it, but as a thought experiment it’s useful. What, then, will set one specialist apart from another?
My hypothesis is a combination of areas of expertise. You have a deep understanding of one field, a second, a third, and you know how to connect them. For example, you know finance, how a particular business works, and what AI can do. At the intersection, you see a problem that a specialist from a single field simply won’t notice. Then you need to be able to frame it, evaluate the solution, and see it through to results.
Sounds like a new job description for an entrepreneur?
Yes, the Valley is discussing a scenario in which the main players of the future will be AI engineers and entrepreneurs. Engineers understand how to organize AI’s work. Entrepreneurs gather AI employees around themselves and create a product or service. A small team gains capabilities that used to require an entire organization.
And everyone else?
This is where Ancient Rome would come up in conversations: free Roman citizens, bread and circuses. People receive social payments, funded, for example, by taxes on large corporations’ use of AI. How to set up that kind of redistribution is a separate question, but it’s solvable.
What occupies me most in this scenario is where people will feel useful. Income is an important part of work. But work also answers the question of why you got up this morning and who needs the results of your efforts. And this trend could trigger an incredible rise in psychological problems among people.
Let’s get back to those who plan to make money. What are they hunting for?
Unique data. A company spends years providing services, running a CRM, keeping a history of its client relationships. Inside that body of data are real requests, decisions, mistakes, results. You can use it to teach a model how a specific industry works.
I was told that people are even looking for companies in bankruptcy. Figures of one to two million dollars came up in conversations: businesses are being bought for access to the information they’ve accumulated. I’d still like to look into specific deals. But the idea itself is remarkable: a company may already be earning poorly, yet its experience is still badly needed by someone.
So the company’s most valuable employee is its archive?
Sometimes it can turn out to be a very valuable asset. Although a folder full of documents doesn’t yet mean you have ready-made training material. You need to understand the quality of the data, what exactly it can tell you, and whether it can be used for this task. Value emerges when you manage to extract something useful from the accumulated information.
And which business model seemed the most straightforward to you?
Buying operating service companies and consolidating them while implementing AI. We discussed the appraisal business. A small company, a handful of people, a client base, a reputation in the region, accreditation with certain banks. Appraisals are needed for collateral, litigation, financial reporting. The clients already exist, they already pay, there’s a reputation.
You can buy companies like these, combine processes, automate part of the work. In the financial model we discussed, an individual business is acquired for roughly 3x EBITDA, while the combined group expects eight to ten. These are benchmarks for a specific scenario, not a price list for all appraisal companies. But the logic is clear: show exponential revenue growth by executing a consolidation strategy, improve the profitability of the acquired companies, and build a business the market will value more highly.
Sounds too good to be true. Where’s the work?
Inside the acquired companies. You have to retain clients and specialists, integrate processes, ensure quality. AI can help prepare materials, but someone has to be responsible for the final appraisal. Revenue growth from buying the next firm doesn’t yet prove that you’ve learned to run the previous one better.
The strategy itself already exists: General Catalyst, for example, publicly describes AI-enabled roll-ups — acquiring service businesses and expanding their capabilities with AI. What I find interesting is that here a new technology can be applied to an economy that’s already up and running.
Sometimes a startup founder lacks access to people who can understand their market and are willing to invest in it. Here, you buy an existing business and optimize it smartly with AI.
You mentioned more than a billion dollars in investment per day.
Let’s clarify the scale: NVCA and PitchBook estimated the entire U.S. venture market for 2025 at $320 billion — about $877 million per calendar day. What amazes me is the scale of the concentration of capital and of people creating new products and technologies. At the same time, a large amount of money around you, as, say, a startup founder, doesn’t yet mean it’s available to you. You need to figure out who you’re a fit for and why you’re worth dealing with.
So what should you do in your first month?
Don’t rush. That was one of the most useful pieces of advice. Go to events, meet people, figure out what makes you unique and where interests align. Look at accelerators: a good accelerator provides an environment and can serve as a mark of quality. But first, it’s worth figuring out exactly which environment you need to get into. A big conference or a small party? From what I heard, private events with a limited number of participants from the relevant field work especially well. Private gatherings of angels, funds, startups. There, you actually have a chance to talk properly. A person can ask you a second question, a third, and understand how you think. Many people said: you need to be known in the Valley. Otherwise, raising capital will be extremely difficult. This is important (and it’s another trend): you create AI employees and automate everything under the sun, but for an important conversation, you still need to show up in person.
On the stock exchange, you don’t need a personal introduction to the algorithm — not yet. Will robots make the market easier to read?
I’ve come up with a hypothesis: the volume of algorithmic trading is growing right now. Fewer and fewer people trade themselves, and more and more do it through trading robots. If many algorithms rely on similar technical analysis rules, they can amplify the effect of those rules. Everyone saw a similar signal, reacted in a similar way — and the move happened. A “self-fulfilling prophecy” effect kicks in. But I wouldn’t turn this into a promise of easy money. Robots have different strategies, and an identical reaction can amplify a move in either direction. The Bank for International Settlements also points to the risk of algorithms behaving in sync. There’s something to research here before trading on the strength of an attractive idea.
Where does the prospect of robots worry you the most?
In relationships. People often want to take the path of least resistance. Another person can be difficult: they have their own mood, their own desires, their own idea of what matters right now. You have to negotiate, sometimes give in. Now imagine a conversation partner who knows you very well and constantly adapts to you. Understands when to support you, when to argue, when to stay silent. Always available. Never brings their own rough day into the conversation. If a convincing physical embodiment of such an AI appears, the temptation will be even stronger.
You’re describing the ideal partner.
I’m describing a very convenient partner. And what worries me is that, after that, an ordinary person might start to seem unacceptably complicated. “Why are they tired? Why don’t they want to talk right now? Why do I have to explain something again?”
We could get used to communication that demands almost nothing of us. And then discover that we’ve forgotten how to put up with another person’s independence.
So a living, breathing human will become a luxury?
Perhaps there will be a countertrend toward imperfect relationships and the human origin of things. Made by Human. People will care that behind a work there’s someone’s lived life, and that beside them is a person who chose to be with them. But I also have the opposite thought: the new generation may not care who made a film or wrote the music. If you like the result, why bother finding out where it came from? Generational change could gradually devalue a label that seems important to us today.
So you’ve brought back two incompatible versions of the future?
So far, I don’t see a reason to throw out either one. Perhaps they’ll work at the same time. In one situation I need a good result; in another, specifically the human experience. The fact that a person made something with their own hands doesn’t mean anyone needs it. The value of that work will still have to be explained.
How do people even cope with so much uncertainty?
This question came up a lot in conversations. The pandemic, layoffs, the feeling that your profession could change abruptly. You studied, built up experience, and now you have to explain to yourself all over again what you can lean on. I wouldn’t blame all psychological problems on AI, but anxiety about the future was a noticeable theme. According to SAMHSA, 57 million American adults received mental health treatment in 2025.
Do you see a business opportunity here?
Yes. If you find a format that genuinely helps people, a service like that could be in high demand. Especially as help becomes more accessible. But talking convincingly and helping a person cope with a problem are different abilities. A product should be judged by what happens to the person after the conversation. I was also told about high occupancy at rehab centers and the healthy margins of this business. That’s what local contacts told me; the economics of specific centers need to be verified. As an idea for a chain business, the space deserves attention. Only here, it’s impossible to separate the financial result from the quality of help for people with addictions.
Funny, really: first we build machines that make life easier, then we rack our brains over how to help people cope with this new life. Because convenience doesn’t answer every question. It can take part of the load off, but it won’t necessarily explain what you need the freed-up time for. And that ties into another piece of advice I liked: find the minimum necessary level of comfort.
Is the Valley advising people to live worse?
I see it differently. You need enough stability to have the strength to act. But when everything is so comfortable that you don’t want to change anything, it’s easy to stop moving forward. For me, it’s a question of how much comfort helps you stay effective, and how much is already lulling you to sleep.
So what would you not want to entrust to artificial intelligence?
Choosing what to do, how to do it, who to do it with, where to go. You can delegate drafting an email or analyzing documents. But I want to keep the ability to get to the bottom of a complicated issue, come up with my own business, and get through an uncomfortable conversation with someone close to me. If a machine frees up my time, I’ll still have to decide what to spend it on. For now, in the Valley, freed-up time is mostly spent running forward even faster. The question is – where to? Toward what future? I think that in about 3 years, the outlines of that future will emerge from the technological fog clearly enough…
