For Non-Technical People, Using AI Well Isn't Really About Technology

How can people without a coding or engineering background really bring AI into their work?

Yesterday I talked about AI with a group of HR leaders, and we came to a question I think matters a lot:

How can people who can't code, and have no engineering background, really bring AI into their work?

Over the past few years I've used AI to build recruiting automation, AI interviews and an HR command center, and I've started letting agents take part in daily work.

From first asking AI to help draft documents, to now having it read data, organize information and carry out tasks, I've gradually arrived at one idea:

For non-technical people, using AI well isn't about how many tools you learn. It comes down to three things: a knowledge base, workflows, and AI governance.

01 | Knowledge base: build your own AI knowledge base

AI doesn't know your company.

It doesn't know your leave rules, pay structure or performance policies, and it doesn't know the unusual cases you've handled before.

Without that information, it can only answer from general knowledge. When information is missing, it may even produce answers that sound reasonable but are actually wrong.

So when a company adopts AI, tools alone aren't enough. It also needs its own knowledge base.

I think of a knowledge base in two parts.

Organizational knowledge

This covers regulations, company policies, operating procedures, forms and past cases.

Most of this already exists. It's just scattered across documents, folders and systems.

The work is to organize, categorize and update it so AI can search and cite it.

For example, with RAG (retrieval-augmented generation), AI first finds the relevant material in the company's knowledge base before answering, and answers from that content instead of relying only on the model's general knowledge.

Personal knowledge

This part is even more worth building up.

Think of the reasoning you use when handling an employee relations case, the tone of your emails, what you habitually confirm before going to a manager, and why you make a particular call when something unusual comes up.

That experience is rarely written into company policy documents, yet it's where a professional's real value lies.

I've started turning my own HR working principles, judgment calls and communication style into instructions AI can read.

It isn't only about telling AI what to do, but helping it understand why. That way, when AI meets a situation it hasn't seen before, it has a chance to work out an appropriate response from those principles.

A knowledge base is more than a pile of documents. It turns organizational experience and personal expertise into assets AI can use.

02 | Workflow and methodology: redesign how work gets done

Once the knowledge is in place, the next step is to think about how AI can take part in the work.

One idea I care about a lot is AI First.

AI First: let AI do the first draft

When you get a task, don't rush to do it yourself. First ask whether AI can produce the first version, and then have a person judge, revise and sign off.

But this isn't just handing your existing work to AI.

What matters more is breaking the whole process down again.

Workflow design

Take hiring. Traditionally, HR might receive résumés, read them, send invitations, call candidates to confirm interest, and then write up interview notes.

If you simply add AI to that, you might just gain a tool that helps write invitation emails.

But if you redesign the process, AI can help organize résumé information, do an initial analysis, send automatic notifications, and even handle part of the initial screening through AI interviews, with HR then confirming the results and next steps.

The difference between the two is huge.

The first uses AI to speed up the old work. The second rethinks how the work should be done.

Human-AI collaboration

Of course, not everything should be fully automated.

We still need to decide which work can go to AI, which needs human professional judgment, and which steps need a person to check.

The worker's role changes too.

We used to carry out every step ourselves. Now we need to be good at breaking down tasks, defining what good output looks like, designing processes, and reviewing what AI produces.

The value of AI isn't just doing the same work faster. It gives us the chance to redesign the work itself.

03 | AI governance: set the boundaries for what AI does

When AI moves from a chat tool to an agent, the risks are completely different.

Chat-based AI mainly provides text and information. An agent can connect to tools, read email, change files, and even operate internal company systems.

At that point you have to ask:

What can AI do on its own? What must always be confirmed by a person?

I see three important areas.

Access control

Don't give AI access to every system just because you want it to do more.

Follow the principle of least privilege: grant only what's needed, and manage read and write permissions separately.

Human approval

AI can help sort email, draft documents and summarize meetings.

But sending email, deleting data and changing sensitive information should keep a human approval step.

Guardrails

Agents also carry a risk that's easy to overlook: prompt injection.

When AI reads outside emails, documents or web pages, it may run into malicious instructions written specifically for AI. Without proper protection, it may mistake instructions in outside content for commands it should follow.

That's why important safety rules can't live only in the prompt. They also need to be enforced in code, checking permissions and conditions before anything runs.

For example: require approval before sending email, stop and ask the user before changing sensitive files, and keep a log of every action.

AI can help carry out work, but that doesn't mean it should have unlimited decision-making power.

Looking back at the three

A knowledge base gives AI knowledge to work with.

Workflows show AI how to take part in the work.

Governance keeps AI acting within the right boundaries.

All three look like technical topics, but what they really test is management ability.

You need to understand your own expertise, know how your company runs, know what good work looks like, and clearly define permissions and responsibilities.

So I don't believe non-technical people need to become engineers before they can create value in the AI era.

If anything, the people who truly understand the business, know the processes, can organize knowledge and know how to manage risk are the ones best placed to bring AI into a company's everyday operations.

AI tools will keep changing. Your understanding of your craft, and your ability to redesign work, are the competitive edge that lasts.

How much of your expertise have you already turned into knowledge that AI can read, understand and use?