HR should automate last.

I use AI across almost every part of Wellhana, from lead generation and sales to content, product, fulfillment and strategy.

The more I use it, the more convinced I am that one of the easiest mistakes is trying to automate too early. I see the same temptation in HR. The conversation quickly moves to agents, workflows and automation: What can AI take over? What can run without us?

Elon Musk has pushed the opposite sequence for years. Tesla went heavily into factory automation, then pulled back after finding that some parts of the process should have been simplified or removed before they were automated.

His rule is simple: heavily question the assumptions, delete what you can, simplify and improve what remains, speed it up, then automate last.¹

I use a simpler version of that same idea with AI:

Audit the work first

My starting point is surprisingly manual.

I have Claude projects around the main areas I own inside Wellhana. Growth breaks into lead generation, sales and content. Product covers product development and the Academy. Operations covers onboarding and fulfillment. Strategy cuts across everything.

Each project gets the context for that responsibility. My sales project knows our offer, target customer, positioning, sales materials, common objections, past conversations and examples of work I like. Content gets a different set of context. So does fulfillment.

When I sit down to work, I never open a blank chatbot and explain Wellhana again. I open the project for that area and continue the work from there.

This semi-manual stage teaches me a lot about the work itself. I see what I keep asking AI to do, what information it repeatedly needs, where I keep correcting it, which steps add nothing and which decisions still need my judgment.

As Musk argues, many times that is the absolute best improvement. There is no reason to make a weak or unnecessary step faster.

For everything that survives, I keep working with AI semi-manually for a while. I change instructions, add better context, remove steps and improve the sequence. Eventually, some workflows become clear and repeatable enough that I know the inputs, the steps, the expected result and where I still want human judgment.

That is when I start turning them into skills, agents and automations.

How I would apply this in HR

I recommend the same approach for HR leaders.

Start with the major areas of work you repeatedly own. Create separate AI projects around the ones that deserve their own context, then feed them the policies, past work, examples, goals, templates and decision rules you repeatedly need.

Work there for a while. You will see what should disappear, what needs a better process, where AI needs better context and where your judgment still matters.

Once a workflow becomes clear and repeatable, automate it.

There is enormous leverage in AI agents and automation. I just would not start there.

Best,
Henry Kaerki
Founder, Wellhana