The bookends of AI: Why data and business impact still define success

Prior to the broad release of LLMs, the rules of engagement for AI were very clear:

1. The quality of your data allowed your algorithms to make reliable predictions about the world
2. The details of your business case allowed your solution to maximize impact
3. The talent of your people allowed you to squeeze out every ounce of juice building the connective tissue between the two.

I spent many years as a Data Scientist, refining the skills to navigate between (1) and (2), with the measure of success being the ability to turn innovation into margin. With the advancement in Generative AI models and the technological ecosystem surrounding them, Data Scientists and AI/ML Engineers have an increasingly powerful tool in their toolbox. With more effective tools, the barrier to entry to achieve this value generation gets smaller and smaller.

Foundational Models already excel at navigating enormous amounts of data to surface relevant information. Each of us possesses an expert trivia teammate and knowledgeable brainstorm partner at our fingertips. In addition to the petabytes of internet data they are trained on, these models can now utilize Retrieval Augmented Generation (RAG), tools, and fine-tuning techniques to incorporate our own knowledge, best-practices, and ecosystems into their worldview.

One of the latest and greatest emerging advancements is Model Context Protocol, which allows Foundational Models to easily access an ever-growing list of data sources (Gmail, JIRA, Confluence, etc.) and tools (e.g., send an email, comment on a ticket, add to documentation, etc.). While this does provide a shared language for LLMs to query our data sources similar to RAG, what MCP allows us to do is move from asking questions about the world to acting upon it. MCP is a way for us to close the gap between knowing and doing. Now, not only do I have a thought partner, I have a digital intern who is always ready and waiting to go do research on my latest project or execute the next step of my task list. In addition, the Agent2Agent (A2A) Protocol provides a framework for teams of interns to communicate, plan and execute.

Combining MCP and A2A with a Foundational Model’s wealth of knowledge and company-specific guidelines has many organizations wondering just how much is within the art-of-the-possible. Could an AI agent become an expert across all of our knowledge systems, supercharge sales, or 10x every engineer?

I don’t think anyone truly knows the answer to this question. But one thing is clear - every AI company initiative remains bound by the bookends; the quality of your data and the impact of your solution. Foundational Models can help us quickly navigate across these bookends, but they cannot substitute for them.

These challenges are a human problem at their core, and can be broken down into simple tenents:

• An effective data source for AI is both governed and quality-controlled to ensure it can reliably deliver trusted information and insight, or plan and complete the right tasks while following the right rules
• An effective use-case for AI stems from a cross-functional team who deeply cares about their impact on the business and effectively using the technology at their disposal to create it

If your organization can create repeatable and constructive strategies for the bookends, they serve as tide-risers for your AI features, tooling and processes. You will:

• Put each and every AI initiative in a position to succeed, rather than fighting an uphill battle to deliver insight and action from information that has neither
• Start with value rather than starting with a solution you want to shoe-horn into value
• Ensure compliance and regulatory requirements are met to derisk AI initiatives
• Ensure any actions an AI Algorithm takes on behalf of a human are understood, approved and respect the context they operate in

If only there was some sort of software solution which could help you build these bookends scalably and effectively? Ok, I’m a technologist and not a salesperson, but I’ll see what my team of interns can help me do about it

• 

  Eric Warner

  Eric Warner

  Director, AI Engineering

  Collibra