Navigating the AI Agent Landscape: A Strategic Imperative
Early Days of AI Adoption: A Simpler Choice
When generative AI, exemplified by OpenAI's ChatGPT, first emerged, the path for businesses seemed relatively straightforward. Companies eager to embrace this new technology often opted to license large language models from leading providers like OpenAI and Anthropic, or explored open-source alternatives from entities such as Meta. This approach allowed for rapid integration and minimized the risk of being left behind in the initial wave of AI innovation.
The Evolving Dilemma: Agentic AI's Impact on Business Strategy
However, with the increasing sophistication of generative AI and the rise of agentic AI, the decision to build or buy has become significantly more nuanced. This critical choice now demands a careful evaluation of multiple variables, including the scale of the business, the intended applications of AI, and what a company identifies as its unique value proposition that safeguards it from market competition.
Prioritizing Core Competencies and Strategic Flexibility
For large enterprises like Home Depot, certain customer-centric processes are deemed too vital to outsource. Ningyu Chen, their Senior Vice President of Technology, emphasizes that areas directly impacting customer experience are developed internally, while maintaining strategic partnerships with AI vendors for underlying technologies. He advises businesses to remain open and adaptable in their partnerships, recognizing the dynamic nature of the AI market where leadership can shift rapidly. Santhi Ramesh, CEO of Future Propel, reinforces that the build-or-buy decision often aligns with a business's core expertise. Organizations whose primary focus isn't technology are generally better served by strategic partnerships and acquisitions. Conversely, those with extensive, sensitive data or highly customized requirements may find building in-house to be the more suitable option.
Understanding the Financial and Operational Implications
Rachel Ibarra of Cardinal Group Companies points out that acquiring external solutions can sometimes lead to operational debt. Her organization employs a hybrid model, developing an in-house AI agent named Stan while also collaborating with external providers. She suggests that businesses should consider the overall coherence of their systems and organizational workflows when making this decision. Building internally can be advantageous when a company's data offers a distinct competitive edge or when only a small, internal component is needed, especially for non-technical businesses with limited engineering resources. Markus McKay-Fleisch from Smartsheet highlights the importance of robust infrastructure and governance, questioning whether companies possess the necessary framework to develop and maintain their own AI systems effectively. This includes ensuring strong hosting capabilities and the ability to manage or repair AI agents should they encounter issues.
Navigating AI Model Selection and Nuances
Developing an AI agent or system also involves meticulously selecting the most appropriate AI model for each specific application. Phenom, an HR technology firm, utilizes a diverse range of models, from open-source options like Mistral to cost-effective solutions from vendors such as Moonshot AI. CEO Mahe Bayireddi notes that organizations often need the flexibility to interchange models based on use case suitability, as different models excel in different domains, such as HR versus finance. This highlights the complexity of managing a multi-model environment and the need for systems that can adapt seamlessly.
Case Study: Building for Specific Needs and Cost Control
Steve Toy, the founder of Just a Bite Better, an AI-powered nutrition application, chose an internal development path for his AI agents. His app leverages multiple agents collaboratively to deliver comprehensive nutritional insights to users. Toy's decision was influenced by his technical background and the unique requirements of his application. He emphasizes that for his low-stakes business model, direct control over the technology allows for greater customization and cost efficiency. Building in-house also provides mechanisms to prevent unforeseen expenses, such as setting hard caps on monthly token usage for each model, thereby safeguarding against exorbitant bills. He advises businesses to thoroughly understand their specific problem domains when considering external solutions in a saturated AI vendor market.
