your AI building blocks: LLMs, workflows, & agents

AI tools are everywhere, and with all the buzz, it can be hard to understand the categories of these tools, let alone how to deploy them in your business.

But it doesn't have to be this complex.

Here’s my simplified breakdown (written as I continue to learn myself):

1/ Large Language Models (LLM)

This is what most people picture when they think of AI, thanks to tools like ChatGPT, Claude, and Gemini. These are smart tools that read and write like people, capable of answering questions, writing summaries, or helping with email and code when prompted.

Simply put, given a question to answer or instructions, LLMs assemble a response by predicting the words to include in their response. It seems like magic, and in a way, it is. But ultimately, LLMs are sophisticated word prediction machines (a thought that might help you sleep better if you're worried about the robots taking over).

LLMs are the "brains" of AI. The big ones are trained on vast amounts of public internet data. But, crucially, they don’t know anything specific about your company’s customers, products, or internal data unless you connect them.

2/ Workflows and Automation

Think of a workflow as a step-by-step checklist for a repeatable task, such as filing a claim, onboarding a new customer, or handling a support case. These processes involve multiple steps done the same way, repeatedly. Software tools like HubSpot, Salesforce, or ServiceNow usually sequence these workflow steps.

You can even build custom workflows using "no code" tools like Airtable or Monday, replacing the hand-coded (Java, C++, C#, etc.) internal applications I used to build 25 years ago when I graduated from college.

To make different software solutions work together, platforms like Zapier, Workato, Tray.ai, and Microsoft Power Automate come in. These workflow, or “orchestration” platforms, connect to various software via APIs. They capture data, process it (even interacting with an LLM for tasks previously requiring human intervention), and send outputs to the next system. We used to hand-code much of this software integration stuff, too.

3/ AI Agents

An agent does something on behalf of someone else. A travel agent, for instance, plans a trip on your behalf by following a standard process: gathering your preferences, researching lodging and flights, drafting an itinerary, gathering feedback, booking reservations, and communicating the details.

AI agents are like little digital helpers. Give them a specific task within a larger process, and they can either complete it or provide assistance. They can connect to external systems, utilize LLMs for human-like intelligence, and integrate results back into the workflow.

If you're planning a single-family vacation, ChatGPT might suffice. But if you have to do these tasks over and over, like a travel agent does for all their clients, you would need a swarm of AI agents to automate repetitive tasks.

Consider training an intern (or an AI agent) to file an auto insurance claim: they'd need to verify identity, look up account info, check premium payments, gather loss details, determine coverage, recommend next steps, find a repair contractor, and schedule repairs.

You can see how AI agents could automate many individual tasks within this claims process, with the workflow stitching it all together.

This same approach applies to any business process — onboarding customers, managing executive sponsor changes, or handling upsell opportunities.

4/ Workflows vs. Agents

When I asked ChatGPT about the difference between agents and workflows, it supported my understanding:

However, an OpenAI paper explains how agents can, themselves, orchestrate workflows, especially for dynamic orchestrations like checking across multiple systems (field service, order management, sales, etc.) to determine where an order is stuck and taking the following appropriate action.

AI-driven workflows will eventually automate decision-making across complex processes, such as claims management, vacation travel planning, or those customer success cases mentioned above.

The technology already exists. But this is why consulting will play a role in AI adoption. We must understand, dissect, and map processes to determine where AI can replace predetermined sequences and tasks that rely on human intervention.

What’s next?

Phew… that was a lot.

As you can see, there's nuance to the AI buzzwords we’re all throwing around. In the coming months, we’ll spend more time dissecting these concepts and tools.

Not only do we want to automate mundane manual work; we want to use AI to do things we’ve never been able to do before because we couldn’t imagine scaling them.

🤘

We’re grateful you choose to read each week. When you’re ready for more, there are a couple ways we can help:

» Cover Your SaaS is a financial literacy course for go-to-market leaders. Grab your copy here.

» Promote your product and services to over 5,500+ senior SaaS Customer Success pros by sponsoring our weekly newsletter and podcast.

Was this email forwarded to you? Sign up at ChiefCustomerOfficer.io.