A cooperative agentic framework is presented in this paper, which enhances function calling capabilities in Large Language Model (LLM) powered agents while optimizing resource utilization. Weaker and stronger agents are combined through a consensus-based method in this approach, addressing the challenge of balancing performance with computational costs. The framework is evaluated across four diverse test suites: Open Weather, The Cat API, Home Search, and Booking, with the number of calls to weaker agents, consensus thresholds, and fallback strategies being manipulated. It is shown by the results that success rates are significantly improved by increasing calls to weaker agents and implementing strong fallback mechanisms, particularly for initially lower-performing agents. Notably, high success rates comparable to GPT-4 at significantly lower costs were achieved by models like Claude-3-Haiku and DeepSeek-Coder. Particular efficacy is demonstrated by the framework in complex scenarios, where performance was substantially boosted by strategic fallbacks. A practical solution for deploying LLM-powered agents in resource-constrained environments without compromising performance is offered by this approach, potentially increasing the accessibility and adoption of powerful agents in real-world applications ranging from API interactions to complex systems.