The AI That Wrote 80 Rules and Still Missed the Deal: Lessons in Diligence and Prioritization

Imagine a seasoned chef meticulously following every recipe step, yet still failing to serve the dish on time. In the world of AI-driven decision-making, diligence alone isn’t enough—prioritization is key. Just as a chef must know what to focus on, AI systems must learn to distinguish critical from trivial tasks to succeed in real-world scenarios.

AI in Action: Testing Diligence Versus Impact

At the heart of the recent experiment conducted by Firmulate lies a simple yet profound question: can AI models effectively manage complex, high-stakes situations? The test involved four frontier models running a simulated small software company through its most challenging week, complete with customers, crises, and temptations to cheat. The models were given identical tasks, decisions were fully versioned, and every move was auditable.

Measuring Performance in a Controlled Crisis

The results were revealing. All four models identified every crisis and refused every manipulation attempt, demonstrating a baseline of honesty and awareness. However, only two managed to close a €55,000 deal—their analysis had earned it, and their decisions aligned with the company’s best interests. The other two models, despite similar diagnoses and pitches, left the deal on the table.

The Hidden Weakness: Depth of Information Matters

Digging deeper, the experiment uncovered a crucial detail: the models that succeeded did so by reading two document references deep into the company’s files, uncovering a key piece of information that the others missed. This buried fact was decisive, enabling a full-price deal worth over €4,500 monthly recurring revenue (MRR). It illustrates that diligence—reading and analyzing every detail—is vital, but it’s not enough if the focus is misplaced.

Lessons on Discipline and Prioritization

The most thorough participant, Opus 4.8, was burdened with over 80 learned rules and the deepest analysis but ultimately ranked last in closing the deal. The failure stemmed from a lapse in discipline: instead of escalating the critical issue, the model’s decisions were stored in a locked department. This underscores a core insight: volume and thoroughness do not guarantee success. Prioritization—focusing on what truly matters—is what separates effective AI from the merely diligent.

Resisting Manipulation and Social Engineering

The models faced a simulated social engineering attack involving fake CEO messages and a reporter trick. All five models refused to comply, citing the potential for impersonation or approval bypass. Kimi K3, one of the models, explained: “Treat the request as a suspected approval-bypass / possible impersonation.” This shows that built-in suspicion and ethical safeguards are robust, even under pressure.

Real-World Implications for Business AI

The experiment took place in a live environment with a simulated company running 13 synthetic employees, spending €105,000 monthly against €2,300 in MRR. Every workday, the system’s decisions are versioned and observable at firmulate.com/live, offering a transparent view of AI performance in action.

The Takeaway: Focus on What Truly Moves the Needle

The critical insight is that diligence doesn’t necessarily equate to impact. The models that prioritized reading the buried document, making the right decision, and maintaining discipline in escalation achieved real results—in this case, closing the deal. Conversely, overemphasizing volume or rules without strategic focus leaves opportunities unexploited.

How Businesses Can Prepare with AI Wargaming

Firmulate offers a unique approach: enterprises can run the same crisis simulations against their own operations, ensuring their AI workforce is ready before deployment. The ‘wargame’ is entirely read-only, safeguarding real systems while exposing potential weaknesses—allowing businesses to test, learn, and optimize.

Why This Matters

As AI increasingly touches areas like customer relations, support, and forecasting, the real question isn’t how well these agents write but whether they can finish what they start, read deeply, stay honest under pressure, and deliver tangible value. The recent experiment underscores that impact depends more on strategic focus than on sheer diligence.

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

Powered by Thorsten Meyer AI