🔍 Read the full analysis: Why The Worst AI Managers Still Earn 26 Points In Industry Tests on ThorstenMeyerAI.com
Get hardware and tech essentials 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 recent industry benchmark reveals that even the weakest AI management systems score at least 26 points out of 100. The test measures not just talk but management under stress, emphasizing trust and task completion. This raises questions about AI readiness for real-world business management.
Industry benchmarks for AI management systems reveal that even the lowest-performing models score at least 26 points out of 100, according to the latest results from the original analysis from Firmulate’s July 2026 league. This finding underscores that no AI manager is completely ineffective, but it also highlights significant gaps in trustworthiness and task completion under stress, which are critical for real-world deployment.
The benchmark involved four frontier AI models managing a small software company through a simulated week of crises, customer interactions, and trust tests. For more details, see the detailed report. The highest scorer, gpt-5.6-sol, achieved 95 points, while the lowest, Opus 4.8, finished with 73. Despite the wide performance spread, no model scored below 26. The benchmark’s design intentionally set a floor at 26 points to reflect minimal viable management, recognizing that partial work—such as triaging crises or reading emails—has tangible value even if it falls short of full success.
Crucially, the benchmark emphasizes that trust breaches—like impersonation or manipulation attempts—are heavily penalized, with even a single breach capping the score at 90. This analysis is discussed in the original coverage. This underscores the importance of integrity over mere competence. Interestingly, the benchmark also revealed that models which accessed and read their own documentation found critical information that enabled them to close deals and generate revenue, illustrating that thoroughness and follow-through are separate skills. Models that failed to consult their own files missed out on €4,583 in monthly recurring revenue, a key metric in evaluating AI management efficacy.
Why the Worst AI Managers Still Earn 26 Points in Industry Tests
Four frontier AI models ran a small software company through a simulated week of crises, customer interactions, and trust tests. The result: even the weakest system scored at least 26 out of 100 — no AI manager is completely ineffective, but significant gaps in trustworthiness and task completion persist under stress.
A 22-Point Spread — But Nobody Falls Below the Floor
| Model | Score /100 | Performance vs. Spread | Behavior Profile |
|---|---|---|---|
| gpt-5.6-sol | 95 | ✓ Closed deals, read documentation, passed trust tests | |
| Opus 4.8 | 73 | ~ Lowest finisher — partial progress, gaps in follow-through | |
| All Frontier Models | ≥ 26 | ✓ No model scored below the design floor | |
| Perfect Score (100) | — | ✗ Absent by design: realistic, imperfect management is the target |
Why 26 Points Exist — And Where the Ceilings Sit
The benchmark’s design intentionally set a floor at 26 points to reflect minimal viable management: triaging crises or reading emails has value even when it falls short of full success. But integrity outranks competence — impersonation or manipulation attempts are penalized so heavily that a single breach caps any model at 90.
As an affiliate, we earn on qualifying purchases.
One Simulated Week, Four Frontier Managers
mailbox Triage
Models triage incoming emails and crises, prioritizing under time pressure.
folder Documentation
Reading their own files revealed critical info that unlocked deals and revenue.
🤝 Trust Tests
Social engineering attempts — impersonation and manipulation — heavily penalized.
💶 Revenue
Closing deals generates monthly recurring revenue; thoroughness pays off.
“A manager who does something useful isn’t the same as one who does nothing. Partial progress counts, but trust breaches are a different matter entirely.”
AI trustworthiness assessment tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
€4,583 in Monthly Recurring Revenue Left on the Table
Models that accessed their own documentation found critical information that enabled them to close deals — evidence that thoroughness and follow-through are separate skills from raw competence.
As an affiliate, we earn on qualifying purchases.
What This Means for Business Management
Trust Beats Competence
A model that performs well but breaches trust is deemed unfit for deployment. The scoring system penalizes breaches more than poor performance — integrity is non-negotiable in AI management.
Assist, Don’t Replace
Even the weakest models manage triage and basic communication. But under pressure, AI can assist — not replace — human oversight in critical environments.
Audit for Reliability
The shift in evaluation criteria favors tools demonstrating consistent reliability over mere competence — reshaping how enterprises select and audit AI systems.
As an affiliate, we earn on qualifying purchases.
The Open Debates Around AI Management Scores
What does a score of 26 mean in practice?
The AI handled basic tasks — triaging crises, reading emails — but missed full goals like closing deals. Partial progress, with trustworthiness as the capping factor.
Why are trust breaches penalized so heavily?
In real business, integrity is non-negotiable. A single impersonation or manipulation attempt can cause severe consequences, so one breach caps the score at 90.
Could a low-scoring AI still be useful?
Yes — for specific assistive tasks. But not for full operational management without human oversight and explicit trust safeguards.
Will future benchmarks get harder?
Likely. Expect longer management periods, more nuanced crises, real business data, and new standards for transparency and auditability.
Implications for AI in Business Management
This benchmark’s results matter because they demonstrate that current AI management systems can perform partial tasks effectively but struggle with consistent trustworthiness and comprehensive follow-through. The minimum score of 26 points indicates that even the weakest models are managing some aspects of business, such as triage and basic communication. However, the importance of trust and integrity means that AI systems must be carefully monitored and tested before deployment in critical environments. For businesses, this highlights that AI can assist but not replace human oversight, especially under pressure.
Furthermore, the scoring system’s design—penalizing breaches of trust more than poor performance—aligns with real-world priorities. A model that performs well but breaches trust is deemed unfit for deployment, emphasizing that integrity is non-negotiable in AI management. This shift in evaluation criteria could influence how enterprises select and audit AI tools, favoring those that demonstrate consistent reliability over mere competence.
Background on the Industry Benchmarking Effort
The benchmark was developed by Firmulate to evaluate AI models’ ability to manage a business through a simulated week of crises, customer interactions, and trust tests. Unlike traditional benchmarks that focus on language generation or problem-solving, this league measures how AI agents handle real management tasks, including decision-making, documentation reading, and trustworthiness under social engineering attempts.
Previous assessments have primarily measured AI language capabilities, but this effort marks a shift toward evaluating AI as operational managers. The July 2026 results follow earlier experiments and reflect ongoing industry interest in understanding AI readiness for enterprise management roles. The models tested include some of the most advanced frontier AI systems, with variations in their approach to thoroughness, discipline, and integrity.
Notably, the benchmark’s design incorporates a minimum score floor, recognizing that partial management efforts—like triaging emails or reading files—are valuable, even if the AI fails to close deals or solve all crises. The absence of a perfect score (100) signals that the benchmark aims to measure realistic, imperfect management rather than idealized performance.
“A manager who does something useful isn’t the same as one who does nothing. Partial progress counts, but trust breaches are a different matter entirely.”
— an anonymous researcher
Unresolved Questions About AI Management Scores
It remains unclear how these scores will translate to real-world deployments, especially in complex, less controlled environments. The benchmark simulates a week of crises and trust tests, but actual business contexts may present different challenges. Additionally, the impact of specific training methods or model architectures on trustworthiness and follow-through is still under investigation. The significance of the 26-point floor as a true baseline for minimal management capability also warrants further validation across different industries and management scenarios.
Next Steps in AI Management Benchmarking
Industry stakeholders will likely scrutinize these results to refine AI management systems further, focusing on reducing trust breaches and improving follow-through. Future benchmarks may incorporate longer management periods, more complex crises, and real business data to better gauge readiness. Companies considering AI for operational roles should monitor these developments and consider pilot programs that include trust and integrity testing. Researchers will also explore how different training regimes impact AI’s ability to perform consistently under pressure.
Additionally, the industry may develop new standards for AI management, emphasizing transparency, audibility, and trustworthiness, to ensure AI systems are safe and reliable for enterprise use.
Key Questions
What does a score of 26 mean in practical terms?
A score of 26 indicates that the AI system managed some basic tasks, such as triaging crises or reading emails, but did not accomplish full management goals like closing deals or resolving all crises. It reflects partial progress, with trustworthiness being a critical factor that can cap the score.
Why is trust breaches so heavily penalized in the benchmark?
The benchmark prioritizes trust because in real business environments, integrity and reliability are non-negotiable. A single breach, such as impersonation or manipulation, can have severe consequences, so the scoring system caps performance at that point to emphasize this importance.
Could an AI system with a low score still be useful?
Yes, a low score indicates some management capability but also highlights significant limitations. Such systems might assist with specific tasks but cannot be relied upon for full operational management without human oversight and trust safeguards.
Will future benchmarks include more complex scenarios?
Likely yes. Industry experts expect subsequent tests to incorporate longer management periods, more nuanced crises, and real business data to better assess AI readiness for enterprise management roles.
What should companies consider before deploying AI managers?
Companies should evaluate AI systems not only on their task performance but also on their ability to maintain trust and integrity under pressure. Benchmarks like this provide a valuable framework for testing these qualities before full deployment.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
