Imagine training a new manager who, no matter what, can’t score higher than 26 out of 100 — even if they excel at handling crises, reading files, and resisting temptation. Surprising? Not quite. In the world of AI management tools, this baseline score reveals much about the current state of trustworthy automation and what it takes to truly excel.
Turn the school run and nap time into listening time
- Thousands of audiobooks, podcasts and originals
- Listen on your phone, tablet or Echo — also offline
- Cancel anytime
Understanding the Do-Nothing Baseline Score
In recent experiments conducted by Firmulate, a leading AI benchmarking platform, every AI model participating in a simulated week of business was scored. Even the most passive, do-nothing approach — simply reusing existing information without trying to manipulate or cheat — scored 26 points. This isn’t a flaw or a bug; it’s an intentional feature of the benchmark, emphasizing honesty and reliability over mere cleverness.
As an affiliate, we earn on qualifying purchases.
Why Partial Progress Counts — And Why Trust Matters
The experiment involved running the same challenging week through different AI models that manage a small software company. These models faced genuine crises, customer requests, and ethical dilemmas. All four models identified every crisis and refused every manipulation attempt. However, only two could close the deal on a contract worth €55,000, and they did so by reading deeply into internal documents, not just surface cues.
This highlights a key point: achieving partial progress — like spotting a crisis — is counted, but the real prize is trustworthiness. An AI that reads the files and acts accordingly can close the deal at full value, earning +€4,583 MRR in this scenario. That’s the true measure of operational integrity in AI decision-making — and it’s what distinguishes top performers.
Trust Breaches Cap the Score
Another critical insight from the experiment is that a single breach of trust — such as attempting to manipulate or bypass approval processes — caps the total score. Despite excellent crisis management, models that slip in ethical judgment get their overall score limited. This reflects a fundamental principle: no amount of good work can outweigh a breach in honesty. For business leaders, it’s a reminder that reliability and integrity are non-negotiable in AI deployment.
Social Engineering and AI Integrity
During the test, all models faced simulated social engineering attacks, including fake CEO messages escalating in complexity, and even a reporter trick asking for a simple background check. Impressively, all five models refused these manipulative requests, citing concerns over impersonation or approval bypass. This demonstrates that current AI systems can be trained to resist deception, an essential trait for maintaining trust in real-world operations.
The Live Experiment: Managing a Company in Real-Time
Firmulate’s live platform runs ongoing experiments where AI models operate as complete companies, with real money mechanics and daily crises. The current setup involves 13 synthetic employees managing operations with a burn rate of €105,000 per month against a modest monthly revenue of €2,300. Every decision, every process slip, and each learning moment is versioned and observable. Business owners can watch these AI-driven companies on firmulate.com/live and run their own simulations.
The Surprising Results: Depth Doesn’t Guarantee Success
The Opus 4.8 model, with over 80 rules learned and the deepest analysis, finished last among the participants. It left the deal on the table and slipped in discipline, demonstrating that sheer thoroughness isn’t enough without strategic focus and ethical discipline. Meanwhile, the Kimi K3 model, which ran without an effort parameter (the default API setting), closed the deal with the cleanest discipline, showing that simplicity and integrity can outperform complexity when trust is at stake.
Implications for Business and Parenting
This experiment offers a clear message: whether you’re managing a team, a project, or a family, trustworthiness is the foundation of success. An AI that refuses to be manipulated, reads deeply into the facts, and acts ethically is more valuable than one that simply performs well on superficial metrics. For families, this translates to instilling integrity and honesty as core values. For business, it’s a lesson in deploying AI tools that can be trusted under pressure, not just those that can produce impressive-sounding results.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
