Highlights
Startups are experimenting with AI agents and humanoid robots like Dictador’s “Mika” for executive tasks and decision-making. Platforms such as Altan and Artisan AI show how autonomous agents create software or functionality as “employees” with almost human supervision. It emphasizes that fully autonomous “AI CEOs” are still in the experimental stage and need clear surveillance and accountability.
As artificial intelligence (AI) systems become more capable, some startups and companies are experimenting with ideas that allow machines to take on more responsibilities, including executive roles. From autonomous agents creating software to robots acting as “experimental CEOs,” these efforts examine whether AI can handle strategic or administrative decisions.
In this article, we consider recent experiments in autonomous decision-making between start-ups. What has been tried, how effective it is, the challenges that still exist, and the ability to learn.
What does a machine do to run a company?
To be clear, “machines that run the company” usually do not mean completely replacing all human leaders. This refers to delegating some executive, strategy, or management roles to an AI agent, robot, or algorithmic system. These AI systems take inputs such as goals and data, and generate outputs such as decision-making and actions at different levels of human monitoring. Startups exploring this will begin by testing narrow or well-defined areas, such as product development, software maintenance, marketing campaigns, and client interaction, then move towards broader operational and strategic decision-making.
Real-world experiments
Mika, the robot CEO of Dictador
One of the most notable experiments is Dictador, a Polish rum company that appointed an AI-powered humanoid robot named Mika as its experimental CEO in 2022.
Mika’s tasks include choosing an artist to design a company’s bottle label, interacting with the company’s DAO (decentralized autonomous organization) community, and managing specific communication duties.
However, important decisions such as employment and firing still depend on people.
Mika argues that decisions rely on data and aim to be free from personal bias in line with strategic goals.
This situation remains more symbolic and experimental than fully functional. It refers to future possibilities and raises questions about which tasks can be safely assigned to AI.
Altan: Autonomous Agent Building Software
More closely to startup operations than ritual roles, the Barcelona-based startup is an AI agent platform that uses minimal human help to create, launch and manage software, raising around US$2.5 million in the previous seed round.
Users explain their ideas through text and voice, and teams of AI agents (UX designers, full stack developers, product managers, etc.) work together to provide software solutions such as back-end automation, infrastructure, and databases.
Altan claims that it already has over 25,000 users, many of whom use it to launch items such as reservation systems and inventory software.
One example is Julius Kopp, a non-technical founder. He generated around $10,000 in monthly recurring revenues in about 60 days using Altan’s platform.
Altan represents a practical step towards a machine-driven business where part of the creation, maintenance and operation of a product is automated or led by agents.
Artisan AI: AI Employees
Another startup in this field is craftsman AI. This designs AI as employees rather than tools or assistants.
Their first AI employee is the Business Development Officer (BDR) AVA. AVA leads, writes and sends emails with client tones, manages outgoing sequences, and improves performance.
Artisans argue that the agent can learn, adjust and improve over time.
This does not mean that the artisan has an AI CEO, but it shows how companies delegate important autonomous decisions to AI agents.
What works and what doesn’t work
Strength
Speed, scalability, cost savings: AI agents work continuously, perform daily tasks quickly, and grow faster than human teams due to many simple functions. Altan’s ability to quickly provide functional software to non-technical founders is an example.
Bias reduction (theoretical), consistency: Tools like MIKA claim to reduce personal bias. Agents like AVA can maintain a consistent tone and approach. However, “in theory” is still important. AI can take bias from training data.
Enabling Non-Technical Actors: One major advantage is lowering the barriers for non-technical founders who have access to tools that needed a skilled development team. Altan users reflect this change.
Testing new governance or management models: including, for example, AI and robots in visible leadership roles – even symbolically, forcing them to think about what decisions really mean. The Dictador/Mika case is useful, even if it has not yet been replaced by human strategic thinking.
limit
Scope of decision: AI is often trusted in clear, low-risk, everyday tasks or symbolic roles. Complex strategic choices such as mergers, large-scale financial commitments, and human resource changes remain human because they are difficult to automate context, ethics, long-term vision, moral judgments, and stakeholder subtleties. Mika doesn’t fire people.
Reliability, Unexpected Behavior, and Monitoring: AI Systems can Fail. Agents may misunderstand prompts, draw false conclusions, or overlook interdependence. These drawbacks require human surveillance.
Ethical, legal, and trust concerns: stakeholders such as employees, customers, and regulators can oppose machine leadership. Accountability can be unclear. Who is responsible for AI making harmful decisions?
Cost, Maintenance, Data Dependencies: Creating, Training, Maintenance, and Update Autonomous Agents requires data, computing power, and system design. Continuous costs and efforts exist, especially for new businesses.
Performance and Functional Roles: Some “AI CEOs” positions are primarily symbolic or PR focused, like Dictador’s Mika.
Technical and Organizational Enablers
For businesses experimenting with machine-driven leadership and autonomous decision-making, certain enablers are emerging.
Modular Agent Framework: Systems created from subagents with specific roles such as design, coding, testing, deployment, etc. perform better than a single large AI system. Specialization helps you manage complexity. Altan is a good example.
Clear Goals and Metrics: Success requires a clear definition of what AI should focus on, including speed, uptime, revenue, design quality, and how to measure performance.
Human Surveillance and Control Points: Even if machines make decisions, humans usually remain involved for high risk or strategic choices.
Trust, transparency, and explainability: Stakeholders need to understand how decisions are made. You also need to be clear about the behavior of the model, data sources, and errors.
Step-by-step implementation: Many experiments start by assigning narrow tasks before expanding.
Risk, ethical implications, and human elements
The idea that AI or robots could be “CEOs” raises many ethical, legal and social questions.
Accountability: If a decision causes harm (financial, reputation, or safety-related), who is responsible? Was it the person who founded the system, the company, the developer, or the AI itself?
Bias and fairness: AI systems learn from data. This means you can either pick up or even strengthen the bias. Decisions about design and artists may support a particular style or demographic.
Jobs and roles: Automation can take away certain jobs, especially in everyday leadership and management areas. However, there are also opportunities to redefine work. Humans can focus more on surveillance, strategy, culture and values.
Trust and legitimacy: Stakeholders such as employees, customers, and industry partners may not trust decisions made by non-human agents or avatars. They could view these decisions as lacking empathy, morality, or human insight.
Legal Restrictions: Legal Law, contracts, liability, and employment rules are based on human behavior. It is unclear how governing bodies will handle non-human CEOs or robots that make legally binding decisions.
Conclusion
Currently, the answer is: It involves partial, experimental and human surveillance. The machine can already assume leadership-related roles. CEOs can typically automate tasks that delegate or handle specific decisions, such as product design selection, artist selection, and code development through agents. Startups like Altan are pushing the boundaries even further towards machine-driven operations. Still, fully autonomous leadership – strengthening all strategic, ethical, legal and interpersonal decisions is not yet here.
Experiments are valuable to what is revealed. They need automation-friendly tasks, tasks that require human judgment, and those that require organizational and legal changes to move more responsibility safely into the machine. For founders and stakeholders, the key is to test in low-risk areas, build trust and transparency, maintain monitoring, and continually monitor performance. The idea of AI “CEO” is no longer science fiction, but it is still a very ongoing work.
