AI tools are supposed to make work easier, removing menial and time-intensive tasks your team members don’t want to do.
But for many businesses, expectations haven’t quite lived up to reality. Instead of eliminating this kind of work, AI simply changes it up.
It’s called human middleware, and it isn’t the way forward.
What is Human Middleware?
Human middleware describes any business workflow where the humans in the loop have to intervene between AI systems or steps. Humans operate in the middle, hence middleware. In these situations, humans end up shuffling information from system to system or doing other similar menial tasks…so that the AI can then do its job (which was, again, to eliminate menial tasks).
The Problem with Human Middleware
The problem here is that if AI isn’t improving work, saving time, or otherwise benefiting your company in ways that offset the costs, then it isn’t worth those costs. This is especially concerning if the “new” work is worse, more convoluted, or otherwise riskier than the old way of operating.
Here’s one example. Before generative AI, an employee might need to type up a meeting summary, attach it to an email, and then draft a short email message explaining high-level details.
That’s fine work to do. But it takes time, and AI can automate most of it — in theory at least.
But for many teams, that automation isn’t as real as it seems.
Instead of the work disappearing entirely, it just morphs into something different.
Consider the steps that your team member might have to take in this scenario:
- Read the AI meeting note summaries to verify details and clean up information that shouldn’t be included
- Copy that information over to another tool
- Verify that tool has selected the right recipients
- Make sure the email message hits the right tone
AI might have churned through the original time-consuming workflow in seconds. But if the new human workflow is more convoluted or just as time-consuming as the old one, did AI really accomplish anything?
Other Examples of Human Middleware
Other examples of AI gone wrong:
- Manually verifying data sets before AI can use them
- Vibe coding, then spending hours fixing what isn’t working
- Rewriting prompts and rerunning AI queries to get a better raw output
- Manually reworking generated outputs that come close but don’t quite hit the mark
In each of these, using AI to do the work ends up creating new, different work that may or may not save time. It also may be less secure, or more prone to overlooking mistakes.
For example, when AI writes an email for you, it always sounds good. There aren’t obvious grammatical problems. And if you’re skimming that output in a hurry, you might not notice that it isn’t actually saying what you need it to say. That is an entirely new problem: if you write the email yourself, you won’t ever end up spouting convincing-sounding nonsense (we hope!).
Why This Happens
AI tools and systems evolve quickly, and the underlying systems and processes don’t always keep up. Instead of reimagining business workflows to be AI-first or AI-native, teams simply bolt on small instances of AI to existing workflows built for human-only work.
There’s also a lot of duplication here: one tool looks great for a specific workflow step, and another seems better for something else. But getting these tools to work with each other (let alone your existing tech stack and workflows) is challenging.
So the end result: too many AI tools doing small slices of the work, trying to fit that work into systems that weren’t designed to take advantage of what AI can do.
A Better Approach
Your employees are capable of much more than serving as the un-artificial intelligence helping AI software communicate with other software. When your team isn’t busy doing that, they can solve creative problems, help customers in uniquely human ways, and create real value for your company.
A better approach is building workflows that use AI in smart, complementary ways, not just bolting on systems to replace similar pre-AI workflow steps.
We admit, this is a more complex way to set things up. If you need an experienced guide, that’s what we’re here to help with. Reach out anytime to discuss your needs.