New research involving more than 900 games finds human teams achieved an 84% win rate versus just 14% for AI-only teams, revealing how leadership and collaboration strategies influence performance.

SAN FRANCISCO, CA / ACCESS Newswire / November 10, 2026 / Skillprint today released new findings from Signal, its human-AI evaluation platform, revealing a critical challenge as artificial intelligence becomes increasingly integrated into workplaces and everyday life: even the most advanced AI systems still struggle with the communication, coordination and adaptability required to work effectively alongside humans.

The experiment tested popular AI models, including those from OpenAI, Anthropic and Google, alongside human participants in a cooperative platform game where teammates were connected by a virtual rope and had to work together to complete challenges. Rather than measuring how well AI could solve problems independently, the experiment examined a more practical question: Can AI effectively collaborate with people, adapt to their behavior and help teams achieve better outcomes?

The experiment surfaced three key findings about human-AI collaboration:

  • Humans dramatically outperform AI at collaboration. Human-only teams achieved an 84% win rate, compared with just 14% when different AI models worked together. This sixfold difference highlights a significant gap in AI’s ability to coordinate actions, plan together and achieve shared objectives.
  • Human leadership and collaboration styles make a measurable difference. Teams consisting of two humans and one AI agent performed better when a human directed the team’s strategy than when the AI agent led. The findings also suggest that how teammates communicate, coordinate and divide responsibilities can influence outcomes, underscoring the importance of collaboration strategies, not just individual intelligence.
  • More AI agents don’t necessarily create better teams. AI win rates declined as teams expanded from two agents to three and four, highlighting how additional agents can introduce communication and coordination challenges rather than automatically improving results.

Chethan Ramachandran, co-founder and CEO of Skillprint, said: “AI is getting remarkably good at solving problems on its own, but that’s not the same as being a good teammate. What we found is that collaboration is a fundamentally different capability. Who leads, how teammates communicate and how they adapt to one another can dramatically change the outcome. As companies introduce more AI agents into the workplace, the question shouldn’t just be how intelligent those systems are, but whether they actually help people work better together.”

Methodology

Skillprint conducted over 900 runs of the experiment in an eight-level cooperative platform game designed to measure communication, planning and coordination. The experiment included 800 AI-only games, 128 human-AI games and 25 human-only games, evaluating popular AI models including Claude Fable 5.1, GPT-6 Astra, Gemini 3.8 Flash and Kimi K3.

Skillprint compared completion rates across different team configurations, finding an 84% win rate for human-only teams versus 14% for AI-only teams. The experiment also examined how different leadership approaches and collaboration styles influenced team performance.

Players navigated obstacles, solved puzzles and worked together to reach the finish, with teammates connected by a virtual rope that required coordinated movement. Skillprint tracked game progress, resets, falls, communication and overall scores, comparing performance across different team configurations and leadership approaches.

Beyond completion rates, the experiment found that humans collaborating with AI waited longer, communicated more frequently and experienced more falls than human-only teams. These behavioral differences offer additional insight into how AI teammates can influence human performance, even when teams are ultimately able to complete the same objectives.

Why Human-AI Collaboration Needs a New Benchmark

As AI evolves from answering questions to taking action alongside employees, evaluating individual model intelligence is no longer enough. Organizations increasingly need to understand whether AI agents can communicate effectively, coordinate with colleagues, adapt to human behavior and contribute to better collective outcomes.

Traditional AI benchmarks typically measure how accurately or efficiently a model completes a defined task. Skillprint Signal extends that approach by examining what happens when AI must collaborate with people in dynamic environments where success depends on shared decisions, communication and adaptability.

Since 2020, Skillprint has conducted more than 800 game-based cognitive studies across more than 500,000 player sessions and more than 1,000 games. Its platform maps more than 150 game mechanics to human signals spanning cognition, mood and personality, translating gameplay behavior into measurable intelligence about how people think, feel and perform.

Read the full experiment findings and methodology at https://experiments.skillprint.co/posts/signal/.

Learn more about Skillprint Signal at https://skillprint.co/ai-labs.

About Skillprint

Skillprint is a human-performance intelligence company using gameplay to understand and enhance how people think, feel and perform. Since 2020, Skillprint has combined cognitive science, game mechanics and AI to translate real gameplay into measurable human-performance signals. Skillprint’s mission is to help build a future where AI augments human capability. Learn more at skillprint.co.

MEDIA CONTACT

Kate Pietrelli
+1 760.518.2633
kate@pathadvisors.com

SOURCE: Skillprint Inc.

View the original press release on ACCESS Newswire

Media gallery

About The Author
Skillprint Inc.