I've helped lead AI adoption at both PwC and Freshworks. Here are 3 common AI mistakes I see workers making.

• Geetha Rajan is a Freshworks strategy leader, and previously led upskilling at PwC.
• She says employees should avoid three mistakes when adopting AI tools.
• Rajan said it's crucial for workers to learn how to think strategically and not outsource thinking.

This as-told-to essay is based on a conversation with Geetha Rajan, a director on the global strategy team at Freshworks, a SaaS company. She's based in the San Francisco Bay Area. Her identity and employment have been verified by Business Insider. The following has been edited for length and clarity.

I'm a director on the global strategy team at Freshworks, where I drive high-priority strategic initiatives that shape the company's growth, investment decisions, and execution, including on AI adoption.

Previously, I spent nearly a decade at PwC advising Fortune 500 companies across healthcare, financial services, and technology on growth strategy and digital transformation. As part of my role, I led the upskilling of over 50,000 employees on automation tools.

Technological transformation has always been happening, but the cloud or mobile transformation took at least 10 years to fully adopt. ChatGPT hit millions of users in the first few months.

A lot of employers will keep expecting that you use AI every day without really understanding the consequences. That's the pressure that actually leads you to make more mistakes rather than use it thoughtfully.

These are some of the mistakes I see that make employees making when adopting AI:

1. Going from 0-100% overnight

A lot of people try to jump straight into becoming Iron Man and fully automate their workflow. It should be a process. The first step is treating AI or the technology as a super intern, so you have the most control over things while giving it low agency.

For example, if you start with giving structured data, but you verify every output. AI can hallucinate outputs that are beautifully formatted.

2. Outsourcing strategy and thinking

I'm a strategy consultant and advisor. So, in terms of the ideation and thinking, that's the one part I don't usually outsource to AI.

This has come through a lot of experience in consulting and being in the workforce itself. I first want to mentally write down my model and first principles. I definitely verify numbers and even try to extract unstructured data from AI, but I still write my first draft very rigorously, keeping my first-principles hat on.

After you've completed a draft, you can ask it to poke holes and say, "Hey, you are the most skeptical board member, or the CFO, poke holes in my strategy."

A lot of AI outputs are really polished. But if you don't have that acumen, if you haven't seen this enough number of times, you actually can't tell if an AI is actually making a mistake or not. This is where a lot of the workslop comes in: You just take the AI output and throw it into an email or an analysis.

I've made this mistake myself, where I had five or 10 minutes, and I asked AI to quickly write down some design principles for me and throw them on a slide. When I was presenting, I was like, "Wait, I don't think this makes sense, and this is not what I was actually trying to say." I actually embarrassed myself.

You can also easily get caught up in a situation where the language AI uses is not something you would use colloquially or even in a professional setting.

Sometimes my biggest worry is what happens five years from now — when nobody actually did that initial job, and we're burning the ladder as we try to climb it. As much as AI can do things, I think it's more about the commitment to yourself that you still learn problem-solving skills and how to use Excel.

3. Approaching AI without context

You need to know exactly who you're solving for and what the purpose of solving that exercise is.

For example, if you're building an AI model to understand your business's customer segments, you still need to know your segments at a high level. That's the part I would never outsource. If you don't have that context yourself, you could just go in a million directions.

The fundamental things about taste, process, architecture, how you build things — those don't come from any tool, irrespective of whether you're using ChatGPT or the latest model. If AI throws 50 ideas at you, you need to know which one of those is going to stick. As an employee, it is your responsibility to pick the right one, so you need the acumen and expertise to do so.