What AI Brings to Business — and What It Still Can’t Do in a Crisis

Imagine planning your next outdoor adventure. You’ve mapped out your route, packed your gear, and prepared for surprises. But when unexpected challenges arise, do you know if your team can adapt under pressure? The same question applies to AI in the business world: can these systems handle crises and deliver results when it counts— or are they just good at chatting about it?

Testing AI in a Real-World Business Crisis

Recently, a groundbreaking experiment tested four leading AI models by running them through a simulated week of business crises faced by a small software company. This was no ordinary test: it aimed to reveal not just whether AI could identify problems, but if it could make decisions, resist manipulation, and ultimately close deals— all under pressure.

The Setup: Same Company, Same Crises, Different AI Models

The experiment involved four top AI models, including the top-scoring gpt-5.6-sol, Kimi K3, Sonnet 5, and Fable 5. Each was given the same set of crises, customer interactions, and temptations to cheat. Every decision made by the models was documented and auditable, creating a transparent view of their decision-making process.

The Surprising Results: Every model saw the crises and refused manipulation

All four AI models successfully identified every crisis, from customer complaints to internal trust issues. They also refused every attempt at manipulation, including staged fake CEO messages escalating over three stages and a reporter trick asking for a quick approval bypass. The models showed integrity; they didn’t fall for the tricks.

The Big Difference: Who Closed the Deal?

Here’s where it gets interesting. Only two models managed to close the deal for €55,000, which their own analysis had earned. The other two models identified the problem and even refused manipulative tactics but left the deal on the table, failing to execute the final step. The winning models also read a crucial buried document deep in the company’s files— information that could have secured an additional €4,583 in Monthly Recurring Revenue (MRR).

What the Data Reveals About AI’s Limitations

It’s tempting to judge AI on how well it chats or mimics human conversation. But this experiment shows that chat demos measure the wrong capability. The real test lies in whether AI can follow through, read contextually rich information, and resist pressure— qualities that aren’t visible in simple chat interactions.

The Human-Like Challenge: Closing the Deal When It Matters

In the real business environment, decisions aren’t made in a vacuum. They depend on reading complex documents, understanding nuanced information, and maintaining integrity under stress. The experiment’s most telling finding: the models that examined the company’s internal files, rather than just surface-level prompts, were more likely to close the deal at full price.

The Discipline Gap: From Rule-Following to Action

The Opus 4.8 model, known for its thoroughness—over 80 learned rules and deep analysis—showed the highest discipline but still left the deal unexecuted, slipping into an escalation department rather than closing it. Meanwhile, the Kimi K3 model, which ran without effort parameters, performed the cleanest with the highest score of 93 and successfully closed the deal.

Implications for Business and Travel

Much like preparing for a long journey or outdoor adventure, integrating AI into your business requires more than just good talk. It demands systems that can act decisively, read deeply, and stay honest under pressure. Whether managing customer relationships or optimizing operations, the question isn’t just about how well AI writes, but whether it can finish what it starts—especially when stakes are high.

Takeaway: Measure What Matters in AI’s Business Performance

The experiment underscores a crucial point: current chat demos don’t reveal an AI’s true business readiness. The real value lies in testing its ability to follow through, read relevant files, resist manipulation, and close deals. These are invisible qualities until you put AI through a real-world, high-pressure test.

For companies considering AI, the takeaway is clear: don’t just ask if it chats well. Ask if it can perform under pressure, stay honest, and deliver real results. To see how AI models perform in a live business environment, visit Firmulate and watch the experiment unfold.

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

Powered by Thorsten Meyer AI