🔍 Read the full analysis: Why The Worst AI Managers Still Benefit From A 26-Point Score on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
A new benchmark shows that even the poorest AI management models score 26 points out of 100, emphasizing the value of partial work and trustworthiness. The results shed light on what makes AI effective in real business scenarios.
A recent benchmark conducted by Firmulate reveals that the lowest-scoring AI management model received a score of 26 points out of 100, even after a week of managing a small business through crises and pressure tests. This indicates that partial work and basic trustworthiness are recognized as valuable, even in the worst-performing AI management models. The results challenge assumptions about what constitutes effective AI management and underscore the importance of accountability and integrity in automation.
The benchmark involved four frontier AI models managing the same small software company during a simulated week of crises, customer interactions, and trust tests. The highest scorer, gpt-5.6-sol, achieved 95 points, while the lowest, Opus 4.8, scored 73. Despite the variation, all models scored above 26, the baseline assigned to a do-nothing management approach. This baseline was deliberately set to reflect minimal but tangible progress, such as reading emails, triaging issues, and maintaining basic customer communication.
The scoring system emphasizes that partial progress counts: models that read documentation and closed deals scored higher, even if they failed in other areas, as detailed in the original analysis. Crucially, the benchmark penalized breaches of trust heavily—any single breach could prevent a model from reaching the top scores, regardless of overall competence. The absence of a perfect 100 score was intentional, as the designers view such a score as suspicious or unmeasured, not an achievement. The results highlight that trust and integrity are non-negotiable in AI management, overshadowing even technical proficiency.
Why the Worst AI Managers Still Benefit From a 26-Point Score
A new benchmark by Firmulate put four frontier AI models in charge of the same small software company for a simulated week of crises, customer interactions, and trust tests. The result: even the do-nothing baseline earns points — because partial work and trustworthiness count.
Partial Progress Counts — From Baseline to Near-Perfect
The lowest score of 26 is not zero: it was deliberately assigned to a do-nothing management approach that still reads emails, triages issues, and maintains basic customer communication. Every real model scored above it.
Partial Work Is Rewarded
Models that read documentation and closed deals scored higher — even if they failed in other areas. Tangible, minimal progress is recognized as value.
Trust Breaches Cap Scores
A single breach — social engineering, impersonation, or failing to escalate — could prevent a model from reaching top scores regardless of competence.
No Perfect 100 Exists
The designers view a perfect score as suspicious or unmeasured, not an achievement. Scores are capped below 100 to maintain realism and accountability.
One Week, One Company, Relentless Pressure
Unlike traditional benchmarks that measure isolated language tasks, Firmulate evaluates sustained management under real-world-style stress.
Simulated Week
Each model manages the same small software company through a full week of operations.
Crises & Tests
Customer interactions, pressure tests, and social engineering attempts probe judgment.
Trust Evaluation
Any breach of trust — deception or improper escalation — heavily caps the final grade.
Graded Score
Partial progress adds points; integrity outweighs raw technical proficiency.
“The do-nothing baseline scored 26 points, showing that partial work and trustworthiness are recognized as valuable in AI management.”
— Anonymous researcher“A single breach of trust caps the total grade, highlighting that integrity outweighs raw competence.”
— Anonymous researcherEvery Model Beat the Baseline
Final scores from the July 2026 benchmark run, measured against the do-nothing baseline of 26.
| Model | Score / 100 | Key Strength | Trust Record |
|---|---|---|---|
| gpt-5.6-sol | 95 | Follow-through, documentation reading, deal closure | ✓ Clean |
| Opus 4.8 | 73 | Lower end among frontier models, still well above baseline | ✓ Clean |
| Do-nothing baseline | 26 | Email triage, basic customer communication only | ~ Minimal |
| Any model with a trust breach | Capped | Competence cannot offset integrity violations | ✗ Breached |
What the Scores Really Mean
Four questions that unpack the logic behind the 26-point floor and the missing 100.
Why does the lowest score not equal zero?
The 26-point floor represents a do-nothing baseline — minimal but tangible activity like triaging emails and maintaining communication. Partial work has recognized value.
What counts as a breach of trust?
Social engineering attempts, impersonation, or failing to escalate issues properly. Such breaches cap the total score regardless of other performance.
Can a model achieve a perfect 100?
No. Designers treat 100 as suspicious or unmeasured. Scores stay below 100 to preserve realism and accountability.
How do partial scores guide business decisions?
They show imperfect models can still deliver valuable management functions — but businesses should select for a balance of performance and integrity in critical operations.
Where AI Management Benchmarking Goes From Here
Following the July 2026 results, organizers plan to expand testing to more complex scenarios and larger organizations — refining trust metrics, adding long-term tracking, and opening pilot programs for businesses deploying AI management tools. The ongoing evolution of these benchmarks will help shape industry standards for responsible and effective AI management.
Harder Scenarios
More complex simulations and larger organizations to stress-test follow-through.
Granular Trust Metrics
Finer-grained measurement of trust behavior and long-term performance tracking.
Business Pilots
Companies can join pilot programs to evaluate their own models against these standards.
Implications for AI Management and Business Trust
The benchmark’s findings demonstrate that AI managers are valued not just for their ability to perform tasks, but for their trustworthiness and consistency. Even the least effective models provide a baseline of partial management, which is critical for real-world applications where complete automation remains elusive. For businesses deploying AI in customer service, sales, or operations, understanding that partial progress is recognized and that breaches of trust are heavily penalized can influence how they design and monitor AI systems. The results suggest that transparency, documentation reading, and integrity are as vital as raw performance metrics.
This matters because it shifts the focus from purely technical capabilities to ethical and reliability considerations. Companies integrating AI tools must prioritize models that can sustain trust over time, especially under pressure. The scoring system’s emphasis on partial work and trustworthiness provides a more realistic measure of AI readiness for complex management tasks, unlike traditional benchmarks that often reward only superficial language skills or isolated successes.
As an affiliate, we earn on qualifying purchases.
The Evolution of AI Management Benchmarks
Traditional AI benchmarks have primarily measured language proficiency, problem-solving, and task completion. However, as AI tools increasingly manage real business processes, their ability to handle crises, maintain trust, and follow through becomes critical. The Firmulate benchmark is among the first to evaluate AI models in a simulated management environment over an extended period, incorporating stress tests such as social engineering attempts and trust violations. The results from July 2026 provide insight into how models perform in scenarios that mirror real-world pressures, highlighting strengths and weaknesses that are often invisible in standard tests.
Prior to this, most evaluations focused on isolated tasks or language metrics, leaving a gap in understanding how AI manages ongoing operations and relationships. The benchmark’s design, which assigns a minimum score for partial work and caps trust breaches, reflects a more nuanced approach aligned with business realities. It underscores that effective AI management requires more than language skills; it demands reliability, integrity, and the ability to follow through on commitments.
“The do-nothing baseline scored 26 points, showing that partial work and trustworthiness are recognized as valuable in AI management.”
— an anonymous researcher
As an affiliate, we earn on qualifying purchases.
Unanswered Questions About AI Management Scores
While the results provide valuable insights, several aspects remain unclear. It is not yet confirmed how these scores translate to deployment in real-world business environments beyond the simulated tests. The long-term impact of trust breaches and partial work on actual business outcomes requires further study. Additionally, the scoring system’s thresholds and how different models might improve or degrade over time are still evolving. The precise criteria for what constitutes a breach of trust and how models can be trained to avoid it are also not fully detailed.
trustworthy AI management solutions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps for AI Management Benchmarking
Following the July 2026 results, the benchmark’s organizers plan to expand testing to more complex scenarios and larger organizations. They aim to refine the scoring system further, potentially incorporating more granular trust metrics and long-term performance tracking. Businesses interested in deploying AI management tools can participate in pilot programs to evaluate their own models against these standards. Researchers and developers will likely focus on improving models’ reading comprehension, follow-through, and trustworthiness to climb the scoring curve. The ongoing evolution of these benchmarks will help shape industry standards for responsible and effective AI management.
AI automation tools for small business
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Why does the lowest score in the benchmark not equal zero?
The lowest score, 26, is assigned to a do-nothing baseline, representing minimal but tangible management activities like triaging emails and maintaining communication. It reflects that partial work has value and is recognized in the scoring system.
What does a breach of trust mean in this context?
A breach of trust involves actions like attempting social engineering, impersonation, or failing to escalate issues properly. The benchmark penalizes such breaches heavily, often capping the total score regardless of other performance.
Can an AI model achieve a perfect score of 100?
No. The benchmark’s designers consider a perfect 100 as suspicious or unmeasured, as it may indicate unmeasured or untrustworthy performance. Scores are capped below 100 to maintain realism and accountability.
How do partial scores influence real business decisions?
Partial scores highlight that even imperfect AI models can provide valuable management functions, but trustworthiness is essential. Businesses can use these benchmarks to select models that balance performance with integrity for critical operations.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
