{"body":"After more than a year of building on my own, I have understood a few things that have started to change the way I think.\n\nAI has greatly expanded what one person can do. It has created the possibility of doing work alone that used to require a developer, designer, analyst, editor and other specialists. At the same time, everyone has access to these capabilities. Products are easier to launch, competition has increased, and building a prototype is one thing while making a solution that really works is another.\n\nHuman attention remains limited. We have not upgraded it to match the tools.\n\nThere is a powerful narrative around AI: it does online work, solves mathematical problems, finds security flaws and will soon take people's jobs. It can seem as if all you have to do is press a button or give a voice command. The reality I encounter every day is a huge number of connections between systems that have to work together.\n\nAt each stage, AI can appear to avoid responsibility. Things look good inside one chat. But ask it to go somewhere else, take something and complete the next step, and you encounter permissions, requests for attention and integrations that do not work. After a year of intensive work, I still do not see an adequate solution to all of this.\n\nThe models are improving. They are taking on more complex tasks. In principle, I think almost any online task can be solved with AI, but some demonstrations require enormous resources: thousands of agents and millions of dollars in tokens. Ordinary people, and even companies, do not have those resources.\n\nModels are becoming smarter, faster and cheaper. That is a good trend. But in my experience, the number of points where things have to connect is staying the same or increasing. The further you go, the more integrations, approaches and compatibility issues multiply. Each can require attention, permission, coordination or more work.\n\nSome of this will probably improve. Yet orchestration problems I encountered before are still here. Security, personal data and restrictions also limit what agents can do on someone's behalf. CAPTCHA is one example. You can add another model that handles it differently, but that is another integration and another thing to support.\n\nThe systems can still do remarkable things. What I have built in Crash Brief would have looked like science fiction to me two years ago: many iterations and data from eight exchanges. At the same time, the larger and more interesting a project becomes, the more attention it needs.\n\nOne person can move from one chair to another, acting as designer, analyst or developer. A team can move in several directions at once.\n\nPeople often think that creating a product is already half the work. Today it can be relatively easy to create one. Convincing people to pay for it is another task, and the answer is not obvious.\n\nI do not think the answer is to send a million automated messages and hope people respond. In my view, that can put your mailboxes in spam, generate complaints and lose people who might otherwise have become customers. Building an audience and earning trust is not something I believe agents can do for you by themselves.\n\nBeyond the digital product there is market research, deciding how to work with a team, funding, integrations, partnerships and finding your customers. You have to show people what you have and speak to them in a human language. It is a lot of work for one person.\n\nI built Crash Brief as a product. My attempts to move it into the B2B market are difficult. Full automation has not worked for me there. Some work happens automatically under the hood, but the connections and conversations need to be personal. When outreach feels artificial or reaches the wrong person in the wrong way, you can end up blocked, in spam, or lose that potential customer.\n\nA team also allows professionals to contribute from different directions. Its strength appears when people's knowledge and experience are connected.\n\nIf everyone works in a separate chat, the others do not know what is happening: the problems they meet, the tools they use or how they solve them. One person's experience is not transferred to another.\n\nThis is where I see a strong opportunity for AI and for the studio we are opening: shared memory that makes one person's experience useful to others. A customer conversation can help the product and be used by a technical specialist. A technical solution can be taken into account in sales. Good solutions do not have to be reinvented.\n\nCommunication between departments often resembles a game of broken telephone. One person tells another, who retells it to a third, and details disappear. A model can synthesize and compact information and pass it to another model. I believe this can be faster, cheaper and more effective than relying on an HR specialist or coach to do the same work.\n\nIt is an ambitious task. I am interested in it myself, and I want to be the first customer of the service we are building.\n","content_language":"en","day_counter":{"label":"Day 413","number":413,"start_date":"2025-09-11"},"id":"2026-10-28-ai-human-attention-and-the-value-of-a-team","project_date":"2026-10-28","published_at":"2026-10-28T12:57:23+00:00","retrospective":false,"schema_version":"crash_brief_public_blog_post_v0_1","slug":"ai-human-attention-and-the-value-of-a-team","source_language":"ru","source_types":["founder_direct"],"status":"published","summary":"A year of building alone, the limits of AI autonomy, earning customer trust and the promise of shared team memory.","title":"AI, human attention and the value of a team","topic":"Founder experience with AI, trust and teams","type":"founder_post","updated_at":"2026-10-28T12:57:23+00:00","version":1}