During an onboarding session for a project involving AI training and evaluation, I looked at the screen and saw 115 people connected at the same time. It was only one session, but it changed how I thought about what we now call a business plan with AI—and about the role left for us as founders when these tools can do far more than write an attractive document.
Around me were consultants, strategists, finance professionals, and subject-matter experts from across the world. We usually talk about AI through the products we see: ChatGPT, Claude, Grok, and others. We ask questions, request research, a business plan, a financial model, or a presentation, and judge whether the answer is good enough. But beneath those products, something much larger is happening. People with professional experience are teaching AI how to do the work itself.
The people teaching AI to become a consultant
Over the past year, I have worked on and participated in processes involving model training and evaluation companies, including Mindrift and Alignerr. I want to be clear: I cannot and do not claim that every project I joined was performed directly for a particular company such as OpenAI or Anthropic. In some projects, the end client is not disclosed, and confidentiality restrictions may apply. Leading companies of this kind are at the forefront of the industry, but that does not make them my direct clients.
What is clear is that companies in this field are building large networks of experts to support training and evaluation for the advanced AI industry. Mindrift, for example, describes a global community of more than 20,000 AI experts. Its consulting page discusses projects for senior consultants, including people with backgrounds at McKinsey, BCG, Bain, Deloitte Strategy & Operations, and comparable roles. Their work includes evaluating and improving AI-generated business analysis.
Alignerr is recruiting MBAs and business specialists to build complex cases, financial models, market-entry scenarios, and benchmark solutions, and to assess AI-generated business plans, forecasts, and analyses. To me, the implications are substantial. The industry is building the professional capability behind the answers we receive. A business plan with AI is no longer just an interesting chat exercise.
What is AI learning to do when we ask for a business plan?
A business plan sometimes sounds like a single document. Anyone who has developed one knows it is a collection of professional tasks: market research, competitor analysis, understanding target customers, assessing an opportunity, building a business model, revenue forecasts, cost structures, financial modeling, scenarios, risk analysis, presentations, and recommendations. Much of this once required a human team and many hours. Models can now assist with some of these tasks, but the quality depends on the task, the information supplied, and human review.
OpenAI is also evaluating models on real-world professional tasks. GDPval covers 44 knowledge-work occupations, including financial analysts, financial managers, and financial advisers. That does not mean models already perform every job better than people. It does indicate where the industry is heading and what kinds of work it is trying to transfer to these systems.
A business plan with AI is only the beginning
Suppose AI can produce useful market research within hours. Then it builds a competitor and alternatives table, a financial model, and a presentation. Next, it checks assumptions, finds more sources, and revises the model. So far, it is helping a founder do their work—a substantial change in itself. But what happens when the same system can also act, accept a task, and keep working without someone sitting beside it and specifying every next step?
When AI no longer just answers, but takes on work
August 2026 saw the launch of Grok Bot. xAI describes it as a team of agents you can assign real work to. The agents have cloud computers and can work inside apps and websites, carry out tasks from start to finish, and continue without instructions at every step, subject to permissions and approval checkpoints. In September, the company also held a three-day event showcasing live Grok Bot use cases for engineers, product managers, salespeople, marketers, and founders.
This is not yet AGI independently creating a startup, and we should not make that leap prematurely. Still, it is no longer the AI we knew a few years ago that waited for a prompt, wrote a paragraph, and stopped. We are beginning to move from a model that answers to an agent that receives a goal. That distinction may matter especially to people building and running businesses.
What happens when the agent identifies the opportunity itself?
Here we move from facts to a future scenario, and I want that distinction to be clear. Imagine an AI agent with access to market data, news, research databases, financial information, and user behavior. It identifies an inefficiency or an underserved customer segment, analyzes market size, maps competitors, and builds a business model and forecast. It then creates an initial product, activates other agents for research, content, development, and operations, and begins executing.
This is no longer pure science fiction. In May 2026, CoinDesk reported on an experiment involving an AI agent called Manfred. According to its developers, it established a company, obtained a US employer identification number, and gained access to financial infrastructure. The report relies on the project team’s claims, not proof that an agent has complete business or legal independence. It remains an early experiment from which broad conclusions would be premature. Even so, it illustrates where the discussion is moving and what might happen as tools gain more ability to act in the world.
Suddenly, the question is broader than whether AI can write a business plan. If it can help identify opportunities, plan, and eventually execute, we should also think about our own place in the process. When AI agents become entrepreneurs, where will you be?
To me, AI’s greatest advantage is its pace
A human founder deliberates, reads, talks to a partner, waits for a meeting, revises the presentation, and asks for another opinion. Then they build a spreadsheet, fix it, take a vacation, return, gather their courage, and finally act. We are human, and entrepreneurship involves much more than thinking of a good idea. AI agents can run tasks in parallel and process large volumes of information without human breaks. They too face resource, permission, and reliability limits. A speed advantage does not remove the need for oversight.
The article’s main illustration compares one day with 100 days. This is not a research benchmark or a claim that every entrepreneurial task can be performed 100 times faster. It illustrates a possible difference in pace. To me, it points to the central issue: as the cost of thinking, research, and production falls, execution time becomes a competitive weapon. Using that time well may matter more than one more polished version of the presentation.
This is also where the problems begin
I do not think the conclusion is to let AI do everything. As tools become more powerful, we need to understand where people must remain involved, which decisions they are responsible for, and what information they supply. Before celebrating the next document AI produces, founders should take four issues seriously.
1. Intellectual property, privacy, and ideas
Before putting a business plan, customer information, contract, financial model, or sensitive idea into an AI system, understand which system you are using and under what terms. Avoid slogans here too: it is incorrect to say that everything uploaded to AI is automatically used for training. For example, OpenAI states that inputs and outputs from its business products and API are not used for training by default, while Anthropic states a similar principle for its commercial products.
That does not remove the need to think before uploading material. Are you entitled to share it? Does it belong to a client? Is it covered by an NDA? What information is actually necessary, and could you use a reduced or redacted version? Answer these questions first, even when pasting a file into a chat and getting a response takes only moments.
2. The bias toward the average
If a hundred founders ask for the best market-entry strategy, the right price, what to include in a pitch deck, and what an investor wants to see, they may all receive similar recommendations. AI is good at recognizing patterns, but business innovation often happens when a founder breaks one. As more founders use the same intelligence, I believe it becomes more important to evaluate the advice and decide when not to follow it.
3. The founder who cannot explain their own plan
You can now arrive at a meeting with a remarkable document: market research, TAM, SAM, SOM, a financial model, a three-year forecast, a go-to-market strategy, and competitor analysis. Then the investor asks why you assumed customers would pay that price, what supports the conversion rate, why you chose that market, and which assumption carries the greatest risk. Suddenly, the impressive document is not enough to sustain the conversation.
If the founder cannot answer, the plan has been built, but their understanding has not grown with it. To me, that is one of the biggest risks of an AI-generated business plan: the document may be more persuasive than its presenter’s understanding. Even if AI does most of the work, you need to explain and challenge the assumptions and stand behind your decisions.
From an AI answer to research you understand
The challenge is not simply getting a good answer from AI, but understanding what it is based on.
Take competitor research. A model can produce a list of competitors in moments. You still need to determine who solves the same problem, what alternatives exist, what the sources say, and what remains unknown.
That is also the approach behind Mikud.AI, a product I am developing for structured competitor research for founders and early-stage teams.
Rather than stopping at a list of names, the process is designed to help you examine alternatives, save sources, and separate AI assessment from your own judgment.
The aim is not for AI to decide whether an idea is good. It is for you to understand the competition better and choose what to investigate next.
4. Execution
To me, this is the heart of the matter. You can spend an entire night with AI and emerge with a better plan, a better model, a better presentation, deeper research, and twenty more ideas. In the morning, you may still have no new customer, no sale, no user conversation, no tested product, and no contract. You spent another night planning, but nothing happened in the world to teach you whether the plan works.
Getting things done is better than perfect. As AI makes planning cheaper and faster, the ability to move things forward in the real world matters more. Someone still needs to turn the research, spreadsheet, and presentation into a step that can be tested with customers, suppliers, or partners. Otherwise, it is easy to keep refining the plan indefinitely and feel as though you are progressing.
What will the human founder’s advantage be?
I do not think the answer is creativity, or at least not creativity alone, because models can suggest surprising ideas too. I am also not sure it will be knowledge, since AI can hold more knowledge than a person. The human advantage will need to combine judgment, understanding people, context, responsibility, courage, leadership, choosing under uncertainty, and execution in the real world.
The founder of the future will not compete with AI on reading speed or document production. They will need to use that speed to understand more and act faster while retaining the ability to decide what is right for their venture. To me, that means beginning to change our mindset now, before we know exactly how far these agents will go.
A new entrepreneurial mindset: what should you ask AI?
Think of AI as a way to explore more possibilities in less time. Instead of simply asking it to write a business plan, ask which business decision you need to make and what could prove you wrong. Rather than request one answer and follow it, ask for three conflicting answers and the strongest argument against each. The goal is to deepen your understanding and use the responses to test your direction.
Then, before spending another evening polishing the plan, test something in the world. In a world of AI agents, founders may compete most intensely on turning thought into action. The easier plans become to produce, the more important it is to understand what deserves execution and what needs to change after contact with reality.
When AI agents become entrepreneurs, where will you be?
When I ask that question, I am mostly thinking about us. What happens when AI can research and build faster, examine more options, and run more tasks in parallel? Will we keep deliberating for another 100 days, or use those capabilities to become faster, more critical, more aware founders who actually execute? That future has not fully arrived, but it no longer seems far away.
Frequently asked questions
Yes. AI can assist with market research, competitor analysis, initial models, forecasts, and drafting. Source verification, assumption checks, and human judgment are still required.
Advanced models can help build financial models, calculations, and scenarios. Users remain responsible for checking assumptions, formulas, and conclusions.
It depends on the product, account type, data policy, and information involved. Confidential material, client information, and content covered by agreements require review before uploading.
Early experiments show agents carrying out parts of business and operational processes. There is not yet a basis for claiming that an AI agent operates as a fully independent entrepreneur and legal entity without people.
Some analytical and technical work is already performed using AI. Human value is likely to shift toward judgment, problem definition, critical review, context, leadership, and execution.