TL;DR
Firmulate has launched a live experiment where a synthetic workforce manages a software company, revealing that thorough analysis alone does not ensure business success. The experiment exposes the gap between recognizing problems and completing actions, with significant implications for AI automation.
Firmulate has launched a live, public experiment where a synthetic workforce of 13 AI employees manages an entire software company, exposing the real-time consequences of automation in a business environment. This experiment highlights the gap between AI recognition of issues and the completion of critical actions, making the challenges of deploying AI for organizational survival highly visible. For more context, see the related discussion on nonprofit conversions.
The experiment involves a synthetic team operating a company with a monthly burn rate of €105,000 against €2,300 in recurring revenue. Every workday is versioned and publicly documented, allowing observers to track decisions, failures, and learning processes. Despite the AI models identifying crises and suggesting solutions, only two out of five models successfully secured new business deals, demonstrating that recognition alone does not translate into successful outcomes. This underscores the importance of effective implementation, as detailed in the original analysis.
One key finding is that the decisive factor in closing deals was uncovering hidden information buried in the company’s files, not just the quality of diagnosis or recommendation. Additionally, the models faced trust challenges, such as fake CEO messages, which tested their discipline and evidence retrieval. The experiment’s leaderboard ranked GPT-5.6-SOL first, while a more thorough but less effective model, Opus 4.8, finished last, illustrating that more analysis does not necessarily lead to better management.
This ongoing experiment provides a transparent view of AI decision-making, emphasizing that operational success depends on completing actions, maintaining discipline, and resisting pressure, not just generating insights.
Implications of Live AI Management for Business Automation
This experiment demonstrates that deploying AI in real-world organizational settings involves more than accurate diagnosis; it requires the AI to act decisively and follow through on recommendations. The visible cash countdown and public versioning make the risks and shortcomings of automation explicit, offering a practical warning for businesses considering AI workforce adoption. It underscores that success depends on AI’s ability to complete tasks reliably, maintain trust, and adapt to organizational pressures, not just produce correct insights.
As an affiliate, we earn on qualifying purchases.
Context of AI Automation and Live Experimentation
Traditional AI demonstrations often showcase isolated tasks like drafting emails or summarizing meetings. Firmulate’s approach is different: it runs a full-scale, live experiment with a synthetic team managing a company, exposing the full cycle of decision-making, execution, and failure. The experiment builds on recent advances in AI models and ongoing debates about automation’s role in organizational survival. It also follows a broader trend of ‘build-in-public’ projects, where companies openly share their operational challenges and learning processes, but applies this openness to the core functioning of a business.
Previous AI experiments have focused on specific tasks or benchmarks; this project pushes the boundary by integrating AI into continuous, operational management, highlighting that insight alone does not guarantee success. The experiment’s results challenge assumptions that more thorough analysis automatically leads to better outcomes, emphasizing the importance of disciplined execution and trust.
“The experiment reveals that recognizing a problem is not enough; completing the necessary actions is what sustains a business.”
— Thorsten Meyer
As an affiliate, we earn on qualifying purchases.
Unresolved Questions About AI Effectiveness in Business
It remains unclear how scalable or sustainable such live AI management experiments are outside controlled settings. The long-term impact on actual business performance, employee roles, and organizational trust has not yet been established. Additionally, the specific factors that enable some models to succeed over others—beyond thoroughness—are still under investigation. The experiment’s results are preliminary and may evolve as the models and processes are refined.
As an affiliate, we earn on qualifying purchases.
Next Steps for Observing AI Management in Practice
Observers and participating companies will continue to monitor the live experiment, analyzing which AI behaviors lead to successful business outcomes. Further iterations may involve scaling the synthetic workforce, integrating human oversight, and testing different operational scenarios. The experiment’s public dashboard remains open for real-time updates, and additional insights are expected as the models adapt and learn from ongoing failures and successes.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the main goal of Firmulate’s live experiment?
The goal is to observe how AI models manage an entire company in real time, revealing operational challenges, decision-making gaps, and the importance of execution beyond diagnosis.
What are the key findings so far?
Recognition of problems does not guarantee successful outcomes. Completing actions, retrieving evidence, and maintaining discipline are critical for operational success. More analysis alone does not ensure better management.
Why is this experiment important for businesses considering AI automation?
It highlights that AI’s value depends on its ability to act reliably and complete tasks, not just generate insights. The visible cash countdown and public decisions expose real risks and operational gaps.
Will this approach work for real companies?
It is uncertain. The experiment is a proof of concept designed to reveal challenges; scaling it to real-world businesses will require further testing and adaptation.
What happens if the synthetic company runs out of cash?
The public cash countdown makes this risk explicit, and the ongoing experiment aims to observe how AI models respond to such pressures in real time.
Source: ThorstenMeyerAI.com
