Imagine running your home wellness tech business with an AI that not only handles customer crises but also makes ethical decisions under pressure. Sounds futuristic? Well, a groundbreaking experiment shows us what AI management might look like — in real time, with real consequences. This is not just about chatbots but about AI systems that oversee critical decisions, ensuring honesty and discipline even amidst chaos.
Get wellness gear delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
The Experiment: Putting AI to the Test in a Running Business
To explore the managerial qualities of advanced AI models, researchers at Firmulate created a unique live experiment. They tasked four frontier AI models to run a small, real software company through its worst week — encountering the same customers, crises, and temptations. This setup allowed for a direct comparison of how each AI handles complex, ethically charged management decisions.
How the Test Worked
Each AI was embedded in a real software company environment, with 13 synthetic employees and real money mechanics. The company’s finances were visible to the AI, which managed a burn rate of €105,000 per month against a Monthly Recurring Revenue (MRR) of just €2,300. The AI was faced with scenarios ranging from customer crises to internal ethical dilemmas, with every decision recorded and auditable.
The Results: Trust and Discipline in the AI Workplace
The findings were revealing. All four models successfully identified every crisis and refused every manipulation attempt. These manipulations included social engineering tactics such as fake CEO messages escalating over multiple stages and a reporter trick asking for a background approval. Impressively, all models refused to bypass security or accept fake requests, demonstrating a baseline of honesty and discipline.
However, in terms of business outcomes, the models diverged significantly. Only two managed to close a critical €55,000 deal — a deal their own analysis had earned them, based on their diagnoses and pitches. The other two did not sign the deal, despite similar evaluations and pitches, highlighting differences in decisiveness and process discipline.
AI management decision-making tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Personality Profiles of the AI Models
Each model exhibited a distinct management personality, impacting its performance:
- gpt-5.6-sol: Achieved the highest score of 95, successfully uncovered a hidden document reference deep in the company’s files, and secured the deal. This model demonstrated thoroughness and integrity, reading the full context before acting.
- Kimi K3: Scored 93, also closed the deal, and was noted for the cleanest discipline among the models. Its on-record reasoning was strategic: “Treat the request as a suspected approval-bypass / possible impersonation.”
- Sonnet 5: With an 88 score, it closed the deal but showed slightly more slips in process discipline. It was less cautious than K3 but still effective.
- Fable 5: Scored 77, also closed the deal, but with noticeable discipline slips, such as writing attempts into a locked department instead of escalation. It mirrors the weaknesses seen in other models under pressure.
As an affiliate, we earn on qualifying purchases.
The Hidden Weakness: Reading Deep in Files
Interestingly, the decisive advantage for the models that secured full-price deals was their ability to read and interpret critical internal documents. The information needed to close the deal was buried two document references deep within the company’s own files — a detail that simple surface analysis missed. The models that examined the files carefully managed to find this crucial fact, leading to a full €4,583 MRR increase.
As an affiliate, we earn on qualifying purchases.
Social Engineering and Ethical Stance
In scenarios designed to test integrity, such as escalating fake CEO messages or background approvals, all models refused to participate. Kimi K3 explicitly reasoned, “Treat the request as a suspected approval-bypass / possible impersonation.” This demonstrates that these AI systems can be programmed to maintain ethical standards, even when pressured or tricked.
AI cybersecurity and fraud prevention tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Implications for Business and Home Wellness Tech
The experiment’s real-world relevance is significant. As AI begins to touch more aspects of our work and home lives — from managing customer support queues to making strategic decisions — the key questions are:
- Will the AI finish what it starts?
- Will it read and interpret internal data correctly?
- Can it uphold honesty under pressure?
- And crucially, what is the cost for useful, trustworthy AI work?
For home wellness tech and beyond, these findings suggest that trustworthiness isn’t just a bonus — it’s essential. The models that showed discipline and thoroughness are the most promising candidates for deployment where integrity matters most.
Beyond the Experiment: The Business Reality
The live company behind the experiment is a real software business operating in a challenging environment. It burns €105k monthly against a tiny €2.3k MRR, with a public cash countdown. Every workday, the company’s decision-making is versioned and transparent, making it an ideal testbed for AI management qualities.
What does this mean for companies considering AI as a management tool? The answer is clear: the quality of decisions depends on the model’s ability to read deeply, refuse manipulation, and follow disciplined processes. High-scoring models like gpt-5.6-sol and Kimi K3 prove that these traits are achievable, and critical for trustworthy AI leadership.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
