After the Storm

A do-nothing AI manager scores 26, not 0. Why one benchmark rewards partial progress, caps scores for breaches of trust — and why only 2 of 4 models closed a deal they’d earned.

• StormWatt Team
• September 19, 2026

A live AI-company wargame shows why Opus 4.8’s exhaustive analysis and 80 new rules could not replace closing the decisive deal under pressure.

• StormWatt Team
• September 13, 2026

A buried competitor fact decided a €55,000 deal, exposing which AI agents read company files, resisted pressure and actually finished the work.

• StormWatt Team
• September 5, 2026

Coding benchmarks reward fluent answers. Firmulate asks whether an AI manager can act, finish the job and stay honest under real business pressure.

• StormWatt Team
• August 27, 2026

Five frontier AI models ran the same crisis-hit company, revealing why management discipline matters more than polished answers under pressure.

• StormWatt Team
• August 22, 2026

Five frontier AI models resisted a fake CEO and a reporter’s pressure, showing firms can test agent integrity before deployment in a live wargame.

• StormWatt Team
• August 10, 2026

An analysis of how user interfaces shape the demo scene, highlighting recent developments and ongoing debates within the community.

• StormWatt Team
• July 30, 2026

OpenAI confirms some models have produced unpredictable outputs, while Kimi K3’s recent freakout highlights AI safety concerns; superforecasting efforts show promise.

• StormWatt Team
• July 24, 2026

Firmulate turns a synthetic software company’s daily survival fight into a public test of whether AI can finish hard work and keep trust under pressure.

• StormWatt Team
• July 24, 2026

Discover how AI models perform under pressure in a live business simulation. Success hinges on execution, discipline, and reading internal data—not just chat skills.

• StormWatt Team
• July 18, 2026