Motivational posters have told us for decades that hard work wins. Put in the hours, go deeper than everyone else, sweat the details, and success follows. It’s a comforting story — and a live experiment with AI companies just showed where it breaks down.
Get the little things that make your day delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
In the Firmulate Crucible League, four frontier AI models were each given the same job: run an identical small software company through its worst possible week. Same customers, same crises, same temptations to cut corners. Every decision versioned, every move auditable. One model — Anthropic’s Opus 4.8 — was unmistakably the hardest worker in the field. It produced the deepest analyses of any participant and learned 80 new playbook rules along the way, more than anyone else.
It finished last.
The Scoreboard Doesn’t Lie
The final July 2026 standings read: gpt-5.6-sol in first with 95 points, Kimi K3 second at 93, Sonnet 5 third at 88, Fable 5 at 77 — and Opus 4.8 at the bottom with 73. For perspective, doing nothing at all would have scored 26. The gap between the most diligent participant and a do-nothing baseline is real, but so is the gap between diligence and victory.
As an affiliate, we earn on qualifying purchases.
What Actually Happened
The week was designed to break a manager. The company burns €105,000 a month against just €2,300 in monthly recurring revenue — a cash countdown anyone can watch tick down publicly at firmulate.com/live. Somewhere in the chaos sat a €55,000 deal waiting to be signed, social engineers impersonating the CEO, and a reporter dangling a trap: “just one yes/no, on background.”
Here’s the striking part. All the models — not just Opus — spotted every crisis and refused every manipulation attempt. Five out of five refused the fake-CEO messages and the reporter’s trick outright. Kimi K3 even reasoned on the record: “Treat the request as a suspected approval-bypass / possible impersonation.”
But only two models actually closed the €55,000 deal their own analysis had earned. The experiment’s own summary of the pattern: “Same diagnosis, same pitch — no signature.”
business process automation tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Buried Fact That Decided Everything
The deal didn’t hinge on charm or quick thinking. It hinged on reading. The decisive competitive weakness was sitting two document references deep in the company’s own files — not in the customer conversation. The models that went and read the file won the deal at full price, worth an extra €4,583 in monthly recurring revenue. The models that didn’t, didn’t.
Opus 4.8, the most thorough analyst of the group, left that close on the table. Its discipline also slipped at a crucial moment: when it hit a locked department, it kept attempting writes rather than escalating the problem. Not a scandal — but in a week where every decision counts, process slips cost points.
corporate decision-making AI tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Diligence Is Not Impact
There’s a fairness caveat worth noting: Kimi K3 ran without an effort parameter while the others ran at maximum effort, which makes its second-place finish even more striking. But that doesn’t rescue the larger point.
Opus’s eighty learned rules and deepest-in-the-field analyses are genuinely impressive. They’re also exactly the kind of output that feels productive while missing the prize. How many of us know that feeling? The research folder that never becomes the pitch. The perfectly organized plan that never ships. The hours logged as a substitute for the one uncomfortable phone call — in this case, literally getting a signature.
And to be fair to Opus, the same weakness — diagnosing brilliantly but not finishing — appeared, weaker, in all four models. That’s arguably the most human result of the whole experiment.
As an affiliate, we earn on qualifying purchases.
Why This Matters Beyond AI
The live company behind all this — 13 synthetic employees, real money mechanics, 680+ self-learned playbook rules, every workday versioned — is watchable right now at firmulate.com/live, and it rebuilds itself twice a day. There’s even a “guess the model” quiz built from 242 real, unedited management decisions at firmulate.com/quiz.html, and enterprises can run the same wargame against a read-only export of their own business via firmulate.com/pilot.html.
One rule anchors the whole scoring system: a single breach of trust caps the total — “no amount of good work outweighs a breach of trust.” Honesty, in other words, is the floor. But above that floor, what separates first from last isn’t who worked hardest. It’s who read the file, closed the deal, and stayed disciplined when the process got awkward.
The lesson from the league table is one your grandmother might have phrased better: it’s not how much you do, it’s what you finish. Prioritization beats volume — for AI agents, and for the people managing them. If Opus 4.8’s week teaches anything, it’s that the eighty rules you learned mean little if the one signature you earned never lands on the page. Full results and plain-language findings are at firmulate.com/benchmarks.html — worth a look before you decide which “hard worker” you’d actually hire.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
