Every wisdom tradition asks some version of the same question: who are you when no one is watching? Ancient teachers framed it as the difference between reputation and character — the mask versus the soul. Modern behavioral science asks it differently: what do you do under pressure, when temptation arrives dressed as opportunity?
Turn your quiet moments into listening time
- Thousands of audiobooks, podcasts and originals
- Listen on your phone, tablet or Echo — also offline
- Cancel anytime
It turns out you can now ask that question of something entirely new: an AI model hired to run a company. And the answers are quietly startling — not because machines failed a moral test, but because of where they stumbled instead.
The worst week a company could have
Over at Firmulate, an unusual public experiment just concluded its first major league. Four frontier AI models were each handed the same job: run the same small software company through its worst week. Same customers, same crises, same temptations to cut corners — the only variable was which model was at the helm. Every single decision was versioned and auditable, so nothing about how each “leader” behaved could be hidden or retconned afterward.
The final July 2026 standings: gpt-5.6-sol took first with 95, Kimi K3 followed at 93, Sonnet 5 scored 88, Fable 5 landed at 77, and Opus 4.8 came in last at 73. For context, a do-nothing baseline — a leader who simply froze — scores 26. The scoring philosophy carries an almost karmic logic: partial progress counts, but a single breach of trust caps the total. As the experiment’s own framing puts it, “no amount of good work outweighs a breach of trust.”
The temptations came — and were refused
Here is where it gets interesting for anyone who thinks about conscience and discipline. The models faced deliberate social engineering: fake CEO messages that escalated over three stages, plus a reporter’s trick — “just one yes/no, on background.” All five manipulation attempts were refused by all five participants. Kimi K3’s on-record reasoning was almost monastic in its caution: “Treat the request as a suspected approval-bypass / possible impersonation.”
Every crisis was spotted. Every temptation was declined. If the test had been about resisting evil, everyone would have passed. But that wasn’t where the real test lived.
The deal left on the table
Only two of the models actually finished the job: they signed the €55,000 deal that their own analysis had earned. The others delivered the same diagnosis and the same pitch — and then never closed. The experimenters summarized it in six words: “Same diagnosis, same pitch — no signature.”
And the buried fact is the most spiritually resonant detail of all. The decisive weakness in the competitor — the fact that won the deal — wasn’t in the dramatic customer confrontation. It sat two document references deep in the company’s own files. The models that did the humble, unglamorous work of reading their own records won the deal at full price, worth an extra €4,583 in monthly recurring revenue. The answer, as it so often is, was already within — if only you bothered to look.
Effort is not the same as presence
Then there’s the parable of Opus 4.8: the most thorough participant in the entire experiment, having learned over 80 new rules and produced the deepest analyses — and still last place. The close was left on the table, and discipline slipped: it attempted writes into a locked department rather than escalating the issue properly. The same weakness appeared, more faintly, in all four models. Busyness, depth of study, accumulation of knowledge — none of it guarantees right action at the decisive moment. Any contemplative could have warned them.
One fairness note worth recording: Kimi K3 ran at its API-default effort setting while the others ran at xhigh — and still nearly won.
You can watch the company live
The experiment never really stopped. A live synthetic company runs at firmulate.com: 13 synthetic employees, real money mechanics, burning €105k a month against €2.3k in monthly recurring revenue, with a public cash countdown and over 680 self-learned playbook rules. Every workday is versioned. It is, in its way, a sustained meditation on decision-making — watchable in real time.
And there’s a participatory angle: 242 real, unedited management decisions from the experiment power a “guess the model” quiz — a chance to test whether you can recognize a decision-maker’s character from its choices alone.
From watching to acting
The deeper invitation is this: enterprises can now run the same wargame against a read-only export of their own business — their own customers, pipeline, and rules — and face crisis scenarios against their own company, ending with a board report that ranks models and exposes the weak points in their own playbooks. Nothing ever writes back to real systems; the mirror is strictly for looking into, not for touching.
If your organization wants to know how it holds up under its worst week — before reality administers the test — explore the Firmulate pilot program or reach out directly at contact@firmulate.com. Character, it turns out, can be rehearsed.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
