Artificial intelligence (AI) agents that operate autonomously within enterprise workflows have become a prominent topic among businesses. However, concerns have arisen regarding the tangible benefits of these technologies, with many suggesting that the current discourse may be more hype than reality. Gartner has indicated that enterprises may be at the “peak of inflated expectations,” suggesting potential disillusionment if vendors fail to demonstrate real-world applications.
Despite these concerns, some global companies, including Block and GlaxoSmithKline (GSK), are actively experimenting with AI agents and reporting early returns on investment. Block has developed an interoperable AI agent framework known as “goose,” aimed initially at software engineering tasks. The platform is now utilized by approximately 4,000 engineers and claims to handle about 90% of code generation, reportedly saving engineers around 10 hours of work weekly by automating various tasks. The framework facilitates communication by compressing Slack and email interactions and is designed to feel like collaborating with a single colleague.
In parallel, GSK is integrating multi-agent architectures within its drug discovery processes to enhance product development. The pharmaceutical company utilizes domain-specific large language models and various ontologies to manage expansive scientific datasets and streamline experimentation. GSK emphasizes the importance of rigorous testing, often rerunning multiple agents in parallel to ensure reliability and correctness.
Both firms highlight the necessity of human expertise in conjunction with AI. While there are opportunities to enhance operational efficiency through these technologies, domain knowledge remains crucial to ensure compliance, security, and reliability in outcomes. The ongoing evolution of AI agents brings both promising advancements and challenges that require careful consideration and integration into existing workflows.
Source: https://venturebeat.com/ai/enterprise-leaders-say-recipe-for-ai-agents-is-matching-them-to-existing-processes-not-the-other-way-around/