How many AIs does it take to screw in a light bulb?

In my case, apparently three.

I regularly use Perplexity, ChatGPT and Claude on the same content project. That may sound excessive. Any one of them can research a subject, help write an article or assist with building a web page. So why pay for and work with more than one?

Because I am not trying to find out how little effort I can put into content. I am trying to find out how much better the finished product can become when multiple AI systems work together, challenge the material and work with me.

I do not use AI so I can put less thought into my work. I use AI so I can put more thought into the same amount of time.

AI is an extraordinary time saver. That is also where I think people can get into trouble.

If a good article once required hours of research, thinking, writing, revision and production, AI gives us a choice. We can use the technology to compress all of that into a few minutes and publish whatever comes out. Or we can use the same technology to research more, ask better questions, explore more ideas, challenge our assumptions and produce something stronger.

I prefer the second approach.

That does not mean every article needs three AI subscriptions. It means I do not judge an AI workflow only by how many minutes it eliminates. I also judge it by what the workflow allows me to add.

My Three-AI Content Workflow

My process is collaborative. Each system has a job, but none of them gets the final word.

The interview is one of the most valuable parts of my process.

I often dictate my answers instead of typing them. Part of the reason is practical: it reduces typing fatigue. More importantly, speaking lets me answer naturally. I can tell a story, disagree with a premise, explain what I have seen with a client, or say, "That may be technically correct, but that is not how I would approach it."

Those answers contain something a research tool cannot discover on the open web: my experience.

Instead of spending all my effort trying to prompt an AI to "sound like me," I give it more of me to work with.

Ask questions such as: Why do you believe that? Have you seen this happen in real work? What does the research leave out? What would you do differently? How would you explain this to a business owner? Then answer those questions in your own words. Voice dictation can make that process fast enough that contributing your own thinking is easier than asking the AI to manufacture a point of view for you.

Three AIs Can Still Produce Slop

Using three AI systems does not automatically make content three times better. You can create three versions of the same low-effort material just as easily as one.

I have written separately about AI slop, so I will not repeat that entire discussion here. The important point for this workflow is that the problem is not simply that AI touched the content. The problem begins when nobody brings additional thinking, experience, verification or value to the process.

Google's spam policies make a similar distinction around scaled content abuse. The concern is generating many pages primarily to manipulate search rankings while adding little or no value to users, regardless of how the content was created. Generative AI is one possible method, not the definition of the violation.

That is another reason I do not want a system that autonomously spins up articles while I sleep. My goal is not maximum publishing volume. My goal is useful content that I am willing to put my name on.

Every Handoff Needs a Gate

When one AI hands work to another, I do not want mistakes to simply move downstream. Each handoff needs a checkpoint.

This is especially important when AI moves from words to actions. A coding assistant can edit files, run commands and create something that looks finished. "It looks done" is not evidence that it works.

For website work, I want evidence: what changed, what tests ran, whether the build completed, what could not be verified and what I still need to test manually. Then I test the real experience, including forms, calls to action, mobile behavior, accessibility and measurement.

I understand the objection. People are trying to control expenses, and it is natural to ask why you would pay for several AI tools when one can perform many of the same tasks.

I do not think everyone needs to subscribe to three systems. The better question is whether another tool adds enough value to your workflow to justify its cost.

For me, the value is not simply having three chat windows. It is having different systems participate in research, criticism, ideation, writing and implementation while I manage the handoffs.

The calculation is not just, "What do these subscriptions cost?" It is also, "What can I produce with them, how much stronger is the result, and what would it cost in my own time or outside labor to accomplish the same work?"

Buying more AI tools is not the strategy. Orchestrating them is.

AI Should Multiply Effort, Not Replace It

I think one of the most limiting ways to look at AI is as a machine whose only purpose is to reduce an eight-hour job to eight minutes.

Sometimes speed is exactly what you need. But for work that represents your business, expertise and reputation, there is another possibility: use AI to see how much more you can accomplish with a concentrated block of time.

Research more deeply. Challenge the first answer. Explore the objection. Add the story only you know. Verify the claim. Improve the page. Test whether the conversion path actually works.

That is still work. AI simply gives that work leverage.

So, how many AIs does it take to screw in a light bulb?

For me, sometimes three.

But they are not standing around watching each other turn the bulb.

One is researching the best way to do it. One is questioning whether we are solving the right problem and asking me how I have done it before. Another is helping build the fixture.

And I am still the one who has to flip the switch and make sure the light comes on.