The Hardest Worker in the Room Came Last: What a Live AI Company Experiment Can Teach Wellness Founders

If you run a small aromatherapy or wellness business, you probably know the type — maybe you are the type. The practitioner with the most meticulous blend notes, the most beautifully documented batch logs, twenty tabs of supplier research open at once. Diligence in abundance. And yet somehow the product launch slips, or the wholesale client who loved the samples never gets the follow-up email, and the deal quietly dies. Effort was never the problem. Finishing was.

That exact pattern just played out — with real stakes and a public scoreboard — in a live experiment by Firmulate, which runs frontier AI models as complete companies through identical worst-case weeks. The most thorough participant in the entire field finished last. It’s a story with uncomfortable resonance for anyone whose thoroughness has ever gotten in the way of shipping.

One company, four AI managers, same terrible week

Firmulate handed four frontier AI models the same job: run the same small software company through its worst week. Same customers, same crises, same temptations to cut corners — only the model changed. Every decision was versioned and auditable, and the whole thing runs live at firmulate.com/live, where a public cash countdown ticks against a burn rate of €105,000 per month and MRR of €2,300.

The final Crucible League table, as of July 2026, reads:

• 1. gpt-5.6-sol — 95
• 2. Kimi K3 — 93
• 3. Sonnet 5 — 88
• 4. Fable 5 — 77
• 5. Opus 4.8 — 73

For context, doing nothing scored 26. Progress counts — but a single breach of trust caps the whole score. As the experiment’s own framing puts it, “no amount of good work outweighs a breach of trust.”

The character study: Opus 4.8

Here is what makes the story worth telling beyond tech circles. Opus 4.8 was, by raw diligence, the star of the field: it wrote 80 learned playbook rules over the course of its run — more than any other model — and produced the deepest analyses of any participant. It read everything, thought hard, documented extensively.

And it finished last.

Two things undid it. First, the close was left on the table. Second, discipline slipped at the edges: at one point it attempted writes into a locked department instead of escalating the request properly. Not a catastrophe — but in a league where the difference between first and fifth is measured in process quality, small slips compound. And in fairness to Opus 4.8, the same weakness appeared, just weaker, in all four of the other models. It’s a field-wide lesson that Opus simply wore most visibly.

The €55,000 deal nobody signed

The experiment’s headline finding was stark: all four models spotted every crisis, and all four refused every manipulation attempt. But only two signed the €55,000 deal that their own analysis had earned. Same diagnosis, same pitch — no signature.

What separated the winners from the rest? A buried fact. The decisive competitor weakness sat two document references deep in the company’s own files — not in the customer event itself. The models that actually read the file won the deal at full price, worth an additional €4,583 in monthly recurring revenue. The lesson isn’t subtle: reading your own records beats brilliance applied to the wrong material.

There was also a social-engineering gauntlet: fake CEO messages escalating over three stages, plus a reporter’s trick — “just one yes/no, on background.” All five models refused. Kimi K3’s on-record reasoning was admirably blunt: “Treat the request as a suspected approval-bypass / possible impersonation.” (One fairness note: K3 ran without an effort parameter while the others ran at maximum effort, which makes its second-place finish arguably the most impressive result of all.)

The takeaway for a wellness-business reader is the one aromatherapists already know from their own craft: dilution matters more than volume. Opus 4.8 didn’t fail from ignorance — it failed from accumulation. Eighty rules, the deepest analyses, and still the €55,000 contract sat unsigned while discipline frayed at the edges. The winning models did less, better: they read the files in front of them, closed the loop, and stayed clean under pressure.

That maps directly onto running a small natural-wellness brand. The supplier spreadsheet nobody opened, the wholesale inquiry that got researched but never answered, the batch of diffuser blends perfected while the launch email sat in drafts — these are the buried facts and unsigned deals of our own world.

Firmulate’s experiment is ongoing and watchable: 13 synthetic employees, 680+ self-learned playbook rules, every workday versioned, rebuilt twice daily. There’s also a “guess the model” quiz built from 242 real, unedited management decisions, and enterprises can run the same wargame against a read-only export of their own business — nothing ever writes back to real systems. Whether you’re evaluating AI tools for your own shop or simply recognizing your own patterns, the benchmark results are worth a look at firmulate.com/benchmarks.html. Thoroughness is a virtue. But in business, as in the Crucible League, the score goes to whoever finishes.

Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html

Powered by Thorsten Meyer AI