Augment or Automate?

One thing I've noticed in almost every AI conversation is that people still tend to fall into 1 of 2 camps:

Replace the people.

Or...

Keep the people.

I don't think either position is completely right.

After years of building AI systems for businesses, I've come to believe the real question isn't whether to replace people with AI or go full-on Skynet.

It's - where should AI be part of the process.

Some work is repetitive, predictable, and well-defined. AI is already remarkably good at that, and it's getting better every month.

Other work depends on judgment, creativity, business context, or simply knowing when something doesn't feel right. That's where people still provide enormous value.

The mistake I see is treating AI as if it has to either do everything or do nothing.

There's a huge middle ground.

Take software development as an example.

I've been building what I call a "software factory". It reviews GitHub issues, determines whether they're bugs or feature requests, analyzes the codebase, writes an implementation spec, hands the work to a coding agent, reviews the results, requests revisions when necessary, commits the code to a new branch, and opens a pull request.

Could I let AI do all of that autonomously? Absolutely.

Do I have to? Not at all.

I can insert myself anywhere in the pipeline. Maybe I review the implementation plan before coding starts. Maybe I review the code before it's committed. Maybe I only review the final pull request.

The automation stays the same - the amount of human involvement changes based on the level of confidence and business risk.

That's why I think we also need to rethink what a "job" is.

People naturally accumulate responsibilities over time. We hire someone for one "job", and before you know it they're wearing 5 different hats.

AI doesn't have to work that way.

Instead of one AI doing everything, you can create specialized agents with narrowly defined responsibilities. Smaller contexts. Fewer conflicting instructions. Clearer separation of concerns. In practice, that often leads to better results than trying to build one giant "do everything" agent (or employee).

But the goal shouldn't be to remove humans from the loop. It should be to remove humans from the repetitive parts so they can focus on the work that actually benefits from human judgment.

Every business will draw that line differently, and that's ok.

The nice thing about AI is that your level of automation isn't permanent. Start by augmenting your team, learn what works (or doesn't), iterate, and automate more (or less) as your confidence grows.

Maybe the real AI debate isn't "will AI replace people?"

Maybe it's "where should people add the most value?"