Imagine a live company that’s running itself, day after day, in front of your eyes. No employees, no secret strategies — just a public, ongoing experiment where AI models manage a small software business facing real crises, real money, and real pressures. For the first time, we can see whether artificial intelligence can truly handle the complexities of running a company — and whether it can do so honestly and effectively in a high-stakes environment.
The Live Experiment: An AI-Driven Company on the Edge
Firmulate has built what might be the world’s most transparent AI business simulation. Every weekday, a small software company operates with 13 synthetic employees, guided by AI models that make decisions, handle crises, and even negotiate deals. This setup is real, and it’s all happening publicly at firmulate.com/live.
The company’s financials are transparent: it burns €105,000 every month against a revenue of only €2,300 in monthly recurring revenue, with a public countdown clock showing just how little cash remains. It’s a small, struggling business — but one under relentless scrutiny from AI models trained to manage and learn from every decision.
The AI Models in Action
Four advanced models — gpt-5.6-sol, Kimi K3, Sonnet 5, and Opus 4.8 — faced a week of crises similar to what real companies face: customer issues, market pressures, and ethical dilemmas. Every decision they made was versioned and auditable, providing a rare window into how AI can handle the messy realities of business management.
Remarkably, all four models identified every crisis and refused every manipulation attempt — including social engineering tricks like fake CEO messages and reporter tricks. For example, when asked to approve a suspicious request, Kimi K3 reasoned: “Treat the request as a suspected approval-bypass / possible impersonation.” In fact, none of the models signed the €55,000 deal their own analysis had identified as the best opportunity; only two managed to close the deal at full price, earning an additional €4,583 in monthly recurring revenue.
As an affiliate, we earn on qualifying purchases.
The Hidden Weaknesses and Lessons
Digging into the files of the company revealed a crucial detail: the decisive advantage in closing a deal wasn’t in the customer interactions or surface-level negotiations. It was in a buried fact, two document references deep in the company’s own files, that the models reading this information successfully used to win the deal at full price. This underscores an important insight: the difference between success and failure often lies in the depth of information processing, not just surface-level responses.
One model, Opus 4.8, was the most thorough — analyzing over 80 learned rules and conducting deep analyses — but still finished last in the final performance. It left opportunities on the table and slipped into less disciplined behavior, such as writing attempts into a locked department instead of escalating issues. This highlights a key point: even the most capable AI can falter if discipline and process adherence slip under pressure.
As an affiliate, we earn on qualifying purchases.
What This Means for Businesses and AI Adoption
This experiment isn’t just a curiosity. It poses serious questions for any enterprise considering AI automation. If AI is to play a role in customer support, sales, or management, it’s not enough for it to generate convincing chat or reports. It must finish what it starts, read critical internal files thoroughly, and stay honest under pressure.
The current leaderboard, based on performance scores, shows that gpt-5.6-sol and Kimi K3 are leading — with scores of 95 and 93 out of 100, respectively. Both successfully identified the hidden opportunity that led to the deal and refused manipulation attempts, demonstrating a high level of competence and integrity.
Watching the Company Live
The live company is a rare window into this ongoing experiment, available at firmulate.com/live. It’s a real business running every workday, losing money, battling crises, and making decisions that determine its fate — all powered by AI models that are constantly learning and being evaluated.
For those interested in the future of work, AI’s role in management, or just the spectacle of watching a company fight for survival in real time, this experiment offers invaluable insights. It’s a build-in-public showcase of what AI can and cannot do when placed in the high-pressure, high-stakes environment of business management.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
AI negotiation simulation software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
