[November 11, 2024]
Our team went to INBOUND in Boston in September, HubSpot's four-day conference dedicated to the latest trends and tactics in marketing, sales and Artificial Intelligence. It’s always a highly useful event, and we want to share the highlights of what we learned here because the three days we were there was very much like drinking from a fire hose.
Spoiler alert: we’ve got so much to say about what we learned that we broke up our post-INBOUND info dump into two parts. Go here to read Part 1.
See AI as a creative partner, not an intern
There was a lot of buzz at Inbound around AI, just like everywhere else right now. One of the sessions we found highly useful was about moving beyond platforms like ChatGPT to focus more on building and, more importantly, onboarding AI.
Even though AI can operate without human intervention, humans have to train it to do so. (An “AI agent,” is an autonomous intelligent system performing specific tasks without human intervention.) One of our favorite lines from the event regarding this new technology might be, “AI won’t take your job, but someone who knows how to use AI better will.”
Before you overthink this and get tech-timidated, consider “agents” you encounter all the time already — Alexa, Siri, the robot receptionist who frustrates you when you call Mediacom.
There are different types of AI agents that perform different functions. Let’s start there:
Simple reflex agents - These are the most basic type of AI. They don’t modify behavior based on learning and use condition-action rules. Your Roomba or robot vacuum is a simple reflex agent. It notices a dirty floor, and takes action to clean it up. But it will not learn where the dirty spots are regularly on your floor and modify to clean those with increased frequency, or even learn to stop the messes from being made in the first place.
Model-based reflex agents - Model-based agents are intelligent and use memory and percept history (things it has perceived) to make decisions. Autocorrect is a model-based reflex agent. It notices which ducking words you use often and adjusts recommendations based on the user’s typing habits.
Goal-based agents - Well, obviously this one is an agent designed to meet specific goals. A GPS navigation system evaluates routes and recommends the fastest or shortest path to a destination. The goal is to get from point A to point B.
Utility-based agents - Utility-based agents compare the utilities of all possible outcomes and choose the one with the highest utility. In plain English, they are like goal-based agents, but they function in situations where there are multiple options to choose from. The “recommended for you” sections on Netflix and Spotify are utility-based AI agents. They assess what you like with the goal of giving you more watch options. In these scenarios, the AI has considered what else you’ve clicked on to present choices. Dynamic pricing systems that can adjust prices in real time would also be an example of a utility-based agent.
Learning agents - Learning agents live up to their name, meaning they learn from experience to improve performance. Speech recognition software is a type of learning agent. As you speak to Google or Alexa, it learns how users speak and adjusts. Spam filters in your email would also be an example of a learning agent. The filters are always learning what new spam looks like to identify it.
Hierarchical agents - Hierarchical agents are an organized group of agents arranged in tiers. These are systems made up of other types of AI agents. This is one of the more complex ways to deploy an AI system because it’s basically a supervisor over other agents. You might have a hierarchical agent making decisions after a goal oriented agent has sent information to a learning agent, for example. Hierarchical agents orchestrate the production line; high-level agents plan and allocate tasks, while lower-level agents control specific machinery like robotic arms for assembly tasks.
Multi-agent systems - This one is exactly what it sounds like. A multi-agent system is built of multiple types of AI agents. AI multi-agent systems bring together capabilities across artificial intelligence, distributed systems, robotics, control systems and human-computer interaction.
AI is helpful at automating rudimentary tasks and helping with ideation, but you have to show AI what success looks like. If you’re preparing to use AI in your marketing efforts, think about it like hiring a new employee, not like talking to Tony Stark’s J.A.R.V.I.S. You can’t tell an AI agent to get to work without showing it what you want it to do.
How do I train my AI agent?
Now that we’re clear on what you can train AI to do, what do you want AI to do? At AKC, we’re experimenting with having AI write posts for social media through the beta version of Hubspot’s Breeze Copilot AI.
As you get started, remember that context for your AI is crucial. Try beginning the process with a unique point of view. If you’re trying to generate content with an AI, you’ll want it to create content that is authentic and digestible. AKC does a lot of work with agriculture, so we’re focused on training our AI to make sure it understands the context of “harvest,” “planting,” “commodity crops” and so forth.
Think about what your trigger words will be. “Trigger words” are words that trigger AI to take action. For example, most of us are familiar with “Hey Siri,” or “Hey Alexa.” Those are trigger phrases to get those AI agents to wake up and pay attention.
Here are some tips we got at Inbound about training an AI:
- Be simple
- Be direct
- Don’t overthink
- Gather and clean essential data you want your AI to understand
- Use delimiters to separate thoughts (symbols like commas and semi-colons)
- Document what you’ve taught your AI, so that you can replicate it if you want/need and also update it.
- Define “good.” Also define “great.”
After training your AI, consider making the prompt a template that the AI can replicate in the future as a building block when you want to run the same prompt for new data or information.
The biggest thing we’re watching as we start to train our AI is the voice. A brand has a unique human voice and personality. To use some slang from Gen Z, AI can't replicate the vibe of a real person, so we’re making sure we read and edit all the text written by our AI.
Stay tuned for more updates! If you want to talk about custom AI agents for your marketing efforts, contact us at content@akcmarketing.com!