Digital agents from your digital agency

Just a few weeks ago, both Google and OpenAI announced they would support MCP, a new protocol developed by a competitor in the AI space, and this milestone is being compared to the development of HTTP or APIs: a moment where several independent technologies come together and a new standard is born.

Hype abounds, and who knows what the future holds, but we share the excitement for this new standard. It enables teams like ours to connect tools to AI in ways that lets them act on their own: to search stuff, buy stuff, create stuff, DO stuff - with or without human intervention.

Why are we talking about it? Because we believe everything to do with software and the web is about to change. And we're getting in front of it (you should too).

Start planning for how AI will change and integrate with your business, like today.

Let’s get the one you might know out of the way: LLMs. That’s ‘Large Language Model,’ and they’re what ChatGPT, Co-Pilot and Gemini use to talk to you. LLMs interpret the relationships between words as sets of numbers.

But LLMs are trained on generic data rather than information specific to your business, and one downside is they famously 'hallucinate' answers when they lack information. You could train your own LLM, but that typically requires more power and compute than mid-sized cities. So how can they be useful to companies who need to get real information to customers, now? I bet you assume we have an answer? Nice. We actually do.

Your encounters with hallucinating chatbots have probably left you skeptical that AI can be trusted for YOUR business tasks demanding specificity and accuracy, but that's where these new techniques and protocols come in.

One is called RAG - Retrieval Augmented Generation. We use this technique to augment LLMs with your own data by storing it in a database, similar to the kind your company website might use today, and sending it along with a user's question to the LLM. Those databases are called Vector Databases and are efficient at taking the context of a user's question and finding the most correct answers to it - based on your real data.

This is the new one. MCP - Model Context Protocol - standardizes tools available to large language models (LLMs), allowing them to retrieve and utilize specific data, specifications, and integrations. For example, an LLM can use MCP to find product specifics to answer customer inquiries accurately. MCP integrates with APIs, web resources, software applications, and even desktop actions.

Almost anything you can do on a computer, MCP can do without being given explicit directions. This changes the game.

There’s a lot more to AI than what we've covered. This new 'agentic' web is a huge topic that's going to change everything we touch or do, and those changes are going to come faster than most people expect.

The less scary news is that we are built for this. We are ready to partner with you to harness these new possibilities into something that feels practical, useful and exciting.

Weirdly, we like change.

Contact us to talk about fit and we’ll find a place to start. Or go ahead and wait things out. Maybe we're wrong and it’s just a fad!