Less than two weeks after OpenAI introduced the GPT-6 Astra model in early September, Tóth Tamás, lead expert at AI Workshop, ran a practical experiment to see how far AI-driven work can be extended in business practice. Over a single weekend he coordinated nearly 70 hours of AI work while investing about 10 hours of his own time—mostly from a smartphone using dictation.
At the experiment's peak, up to 12 AI agents worked in parallel across different browser windows on sales, development and marketing tasks. The aim was not only to solve isolated tasks but to connect elements of a full business ecosystem: the website, Facebook Ads and Google Ads advertising systems, and Analytics were all part of the workflow.
What the system accomplished in practice
During the project the AI rapidly produced campaign creatives and settings: 70 final ad images were generated within one hour. After launching the campaign, the first purchase arrived within two hours, and within a few days 65 tickets were sold for a September event. According to the expert, preparing and launching a similar marketing campaign through a traditional agency would previously have taken 2–3 weeks; with the AI-driven processes many phases started immediately, resulting in savings of more than 1 million forints in graphic and ad-management costs.
Beyond marketing tasks, the AI analyzed the company's existing development projects: of 140 reviewed opportunities it identified 90 as marketable developments, and among these six products were prepared for immediate market testing and advertising. Additional outputs included an automated query application integrated into the enterprise management system and the start of development on an automated update module for a fitness software.
Human oversight and responsibility remain essential
Tóth Tamás emphasized that while generative AI relieves a significant portion of workload, strategic direction, review and final approval remain human responsibilities. Of the roughly 10 hours of human work, a substantial share was spent on professional control: giving instructions, checking results and catching errors. For example, a dictation mistake left an error in a generated video that was found during review, preventing a faulty ad from being published.
Throughout the project results were regularly tested, and in many cases materials were cross-checked with a different AI model to filter out possible mistakes. The expert views critical thinking and professional expertise as indispensable parts of AI use.
Organizing knowledge and the evolving leadership role
According to Tóth Tamás, the next step for company leaders is learning to think in processes and to organize the tacit knowledge in their heads into AI-based systems. In the multinational sector there are already meetings where AI agents participate alongside human managers, and leaders must account for the work of these agents just as they do for team members.
He gave an example where a recording or conversation made by a leader can be processed by AI and turned into structured, immediately usable training materials. These knowledge assets and automations can be embedded into existing company systems so employees can retrieve necessary information or verify tasks autonomously without continuous manager involvement. Companies can also create an internal AI assistant on their own server or in a closed system, populated with the firm's knowledge base, which staff can query about daily operating rules, training materials or internal procedures.
Data security and practical support for SMEs
The AI Workshop lead expert also called attention to data security: sensitive client, financial and banking data should only be handled in properly protected, closed systems.
Their experience shows that for many small and medium-sized enterprises the main challenge is not technology availability but how leaders transform existing workflows and hand tasks over effectively to AI systems. AI Workshop's training program and practical events aim to support companies in that transition.
On September 23 the organization will hold a 250-seat practical conference where participants can observe live how small-business automation and AI-based virtual teams operate in practice.
Conclusion
The weekend practical test in Hungary demonstrates that generative AI today can coordinate work across multiple domains and execute end-to-end business processes when guided and supervised by humans. The technology can accelerate processes and reduce costs, but managerial responsibility, professional oversight and data protection remain central concerns.
