Replacing people with AI is a bad bet. Augmenting them is another story. Here’s how I design Centaur workflows that pair human wisdom with AI speed.

“I fired my entire marketing team and replaced them with AI tools,” he told me. 

I was on a networking call with the founder and CEO of a health technology company. When he said those words, I tried my best to hide my shock. It was the first time I had heard someone in real life say something like this, rather than coming from the AI media hype machine.

I find this approach both unsettling and dubious.

Unsettling, because this “fire and replace with AI” mindset captures the fear that so many of us have, that our jobs are insecure because of AI.

Dubious, because I have really strong doubts that any company will get good results from relying purely on AI tools for their marketing (or any other company function, for that matter).

As the founder of a small communications company, I use AI tools every day to augment nearly every aspect of my business. But am I ready to fully hand off my marketing, writing, project management, or other tasks to AI?

No way. Not a chance.

In the debate between “AI will rule the world” hype and “AI can flush itself into the dankest sewer of the metaverse” opposition, I’m somewhere in the middle. AI tools like Claude work best as a value-add and not a substitute for existing teams’ knowledge, experience, and insight. 

I take the Centaur approach, with a human always in the loop and leveraging the power and speed of AI where most effective. It can take some reflection and experimentation to figure out the best places to make use of AI and when to stick with old-fashioned human effort.

Here are some approaches I’ve found helpful in designing practical and efficient AI systems for Helix River.

Assess the tasks on your teams’ plate

What are the right kinds of tasks to delegate to AI? The answer will vary for every person and team. Some power users will race towards offloading as much as possible from their plates, creating an army of AI agents to do their bidding. Others will tiptoe cautiously around this alien intelligence, dabbling with drafting an email or getting an answer to a technical question.

Here are some questions and considerations to help figure out the right balance for you or your team:

Is the task boring or tedious? 

We all have countless tedious tasks that must get done somehow that require long hours of repetitive work to complete. Or things that are simply boring and monotonous, and don’t require any high-level thinking. These tasks can be great candidates for AI delegation.

I’ve added two columns to my project management system in Asana to track task enjoyment and difficulty. With this data, I can export to a spreadsheet, which I hand off periodically to Claude to review and assess. Claude shares with me a synthesis report, identifying places where I should either delegate to someone else, automate with AI, or keep on my own task list at all costs. 

For example, through this analysis, I identified a way to streamline my audience analysis process. I’ve developed an AI agent to help with creating draft personas and synthesizing interview transcripts to validate personas. The agent cuts back on the long and tedious work, so that I can focus on strategy and storytelling.

Is the task related to your professional development? 

For any skills deeply connected to your success and growth, then delegating to AI is probably a mistake. For example, writing is central to my professional path, something I’ve spent many years practicing and cultivating. While it’s tempting to take the easy road and let AI write drafts of my articles or other content, I keep this task on my own plate. 

I use AI as a thought partner when considering ideas for articles or as a reviewer to provide feedback through different audience lenses. But I hold tightly to the effort of writing.

Like any important skill, writing is hard — and it should be. We cannot advance without struggle. So relying on AI to take away the hard work hinders my own growth. 

Does the task have a straightforward solution or process? 

In a recent article, I wrote about the distinction between kind and wicked learning environments, and how understanding the differences can help us work with AI effectively. A “kind” environment involves having clear rules, predictable patterns, and replicable solutions. In contrast, “wicked” environments have a complex mix of interconnected factors, unclear solutions, and long time horizons between action and result.

AI excels in kind environments. Give it a specific job to do with the right context, and you’ll have valuable results quickly. Wicked environments, however, pose far greater challenges to AI tools — and will almost always require a human at the helm.

Design your workflows

Once you’ve identified some tasks or activities that you’d like to automate, a next step is to map out a workflow — a sequence of tasks leading to a specific output. The workflow might just include a handful of tasks or could require breaking it down into multiple phases with several activities.

For an example of a simple workflow involving just a series of tasks, we could look at making a synthesis of meeting notes. This most likely just requires a series of tasks:

1. Capture notes and a transcript of the meeting. 
2. Draft some questions or criteria about what you want to learn from the meeting. 
3. Upload the transcript to your AI. 
4. Write the prompt with your questions and criteria. 
5. Provide other additional context as needed (e.g., previous meetings, related planning docs).
6. Complete the synthesis.
7. Upload to storage platform.

Note that any of these tasks could be completed by a human. But some of them, like getting a full transcript of the meeting and the synthesis, could be very time-consuming. Others are simply tedious, like uploading documents. Some tasks, like drafting questions or criteria to guide the synthesis, have reasonably straightforward solutions. All of these make great candidates for AI delegation.

Other workflows are far more complex. For example, a workflow for donor research might involve the following phases and activities:

• Prospect identification
  • Compile a list of potential donors from your existing supporter base
  • Research new prospects with mission alignment
  • Apply screening criteria (giving capacity, philanthropic history, mission fit)
• Prospect profiling
  • Gather public information on high potential prospects
  • Identify giving history, areas of interest, and known affiliations
  • Surface potential connections (mutual contacts, shared networks)
  • Draft profile briefs to inform cultivation
• Cultivation planning
  • Draft talking points connecting each prospect’s interests to your work
  • Outline initial outreach approach for each prospect
  • Set up a tracking system for cultivation activities

As with the simple workflow, I review the list of activities and tasks to identify those that are best suited for a human touch (wicked, tied to your experience and relationships), those that are best for AI to tackle (kind and tedious), and those that require more back-and-forth with the AI (AI as thinking partner). 

By breaking down large efforts like prospecting into phases, activities, and tasks, you can condense weeks of work into hours.

However, a workflow is only as good as the judgment behind it.

Using artificial intelligence with natural wisdom

For all its capabilities and despite the hype around AI, I’m a firm believer that we will always need humans at the helm. In particular, we need human wisdom to guide the how and why behind our workflows. Without human wisdom, AI only accelerates slop rather than strategic and compelling communications.

AI is very good at quickly generating intelligent responses to specific questions.

Humans have something AI does not, and perhaps cannot cultivate: wisdom. (Not to say that we always exercise this human capacity, but it’s in there!) 

Our capacity for wisdom can help us make better decisions.

We need human discernment to assess things like whether we’re using the right voice and tone in our messaging. It helps us determine when a story hits the right marks, or if it falls short. It helps us know when to delegate, when to automate, or when to hold onto our tasks. And it helps us know how all the pieces fit together.

Most crucially, it helps us to know whether firing your entire marketing team and replacing them with AI is a good decision (spoiler: it isn’t).

AI can’t replace humans. But when we combine the natural wisdom of humans with the speed and execution of AI in well-designed workflows, we can make some magic happen.