We define AI pretty simply at Teammate AI: computers doing things humans used to do shouldn’t do anymore.
An AI teammate is pretty much that same idea. It’s an AI system that helps you get work done. It might increase throughput, save time, reduce errors, remove a bottleneck, or simply take a repetitive task off someone’s plate.
I think you can pretty safely rebrand “AI teammate” or even “AI agent” to “smart automation” and you’re in the right working mindset.
For the remainder of this essay, if you prefer the term “automation” or “agent” instead of “teammate,” feel free to use it interchangeably.
An AI teammate could be ChatGPT or Claude. It could be an off-the-shelf piece of software. It could be a relatively small automation running quietly in the background. Or it could be a custom system that touches five different pieces of software and completes an entire workflow.
At the end of the day, the important question to ask yourself is, “Did this technology create any of the following quantifiable impacts?”
- Save me time
- Improve throughput
- Reduce errors
- Alleviate bottlenecks
Think about a human teammate
At the risk of stating the obvious, you usually hire a human because there are things that need to get done.
Especially junior employees. You’re primarily getting repetitive tasks done and training them to someday become senior employees.
With the former, an AI teammate can be great. You give it responsibility for work that is appropriate for it, and it helps the rest of the team accomplish its goals faster.
That’s also why “AI teammate” is broader than “chatbot.”
A chatbot is, essentially, language-in, language-out. It’s very helpful, and it’s a useful tool on the toolbelt. Helpful chatbot examples include having AI do a Google search for you, edit your email, or help you think about your thinking. That’s all chatbot work, which is one kind of teammate.
But an AI teammate (or agent or automation or fill in the blank) can also take action.
A teammate can go into another system. It can pull information from different sources. It can execute a multi-step process. It can update software. It can prepare something for a human to review. In the right environment, it can take a task from beginning to end.
When we share case studies, we’re sharing AI teammates.
If I were making the analogy to a traditional workforce, a chatbot is often acting like a researcher, scribe, or ghostwriter. An AI teammate can act more like an assistant or a colleague.
Where should you use one?
This is the question companies tend to struggle with.
They know AI is important. They buy some licenses. They encourage everyone to experiment. Then six months later they’re wondering why the business doesn’t feel dramatically different.
The problem is usually that they started with the AI instead of starting with the work. With this method, it’s a solution in search of a problem.
Before automating a process, understand the process.
Who is involved? What happens at each step? What software is being used? What comes into the process? What is supposed to come out?
You should leave with an SOP or a process capture that looks sorta like this:
Those are usually the best places to start.
An example from our own company
At Teammate AI, we use an internal workspace called Raven.
When we’re designing software for a customer, there can be a huge amount of relevant information.
What did the customer tell us in meetings? What requirements did they send us? What have we built for similar customers? Where should a human remain in the loop?
How do we design the automation so that AI token costs don’t balloon?
Raven can pull from our Asana, Microsoft Office, Teams conversations, customer meetings, company-owned IP, and other internal information.
It then produces a first-pass requirements document that our team validates and improves.
Raven saves our team many hours of doing the same work over and over again. More importantly, it saves our customers from human error—missed details or nuances—in the hunt for the critical-path items leading to transformation.
AI teammates do not need to be fully autonomous
One of the difficult lessons we’ve learned in the last year is to reset expectations about AI teammates. A lot of times, they can save meaningful time (say, 70+% of the task’s time), but the work that the human does is different from the work before. We’ve had adoption issues from this exact scenario:
- A human does a manual, boring task. It takes 5 hours.
- We bring in AI to automate it. Now the task takes 90 minutes and is more accurate.
- The 90 minutes the human spends working are different from before. Say, instead of doing Excel formulas and moving data around, they’re validating that the data that was moved around is accurate.
- The human says, “I don’t want to use it because it doesn’t work,” when, in reality, the AI still saves 3.5 hours; it’s just that the value proposition to retrain wasn’t good enough.
These scenarios arise because the expectations were improperly set. Most AI agents, especially nowadays, are human-in-the-loop. There’s a human either kicking off or validating that the AI teammate did what it’s supposed to do. But the more I think about it, the more sense it makes. Humans validate the work of other humans. We don’t blindly trust human outputs. Why would we do that with AI?
How to start
The approach we use is straightforward:
Map the process. Identify the repetitive or slow parts. Delegate the appropriate work. Review the output. Measure what changed.
And I’d particularly measure time. How much time was saved by having an AI teammate do this task? How much time will be saved over the course of one, two, or three years? And is that time saved worth the investment in getting the task automated?
Most companies think they need to build an overwhelming AI strategy that completely transforms the way the company operates. In reality, I think that the opposite is probably closer to true: humans push the company forward while AI teammates do the dirty, repetitive work.