AI may have become mainstream when ChatGPT launched in November 2022, but for Exiger CEO Brandon Daniels, it’s been a decades-long journey.

“I have been in AI for 20 years,” Daniels, 45, tells Entrepreneur in a new interview. “I’ve built AI systems for my entire career.”

Daniels became CEO of Exiger in May 2022 after previously serving as president of its technology and growth business. He has also held leadership positions at intellectual property software developer CPA Global and business consulting firm Clutch Group.

Exiger helps businesses see and manage risks in their supply chains and third-party relationships. The company uses AI to show firms exactly what is in their supply chain and spot risks before they become disruptive. The company’s valuation is in the “multi-billion dollar range,” according to Daniels.

Related: AI Is Impacting Jobs. Here Are the Gigs Affected the Most, According to an Analysis of 5 Million Upwork Postings

Daniels tells Entrepreneur that AI is likely to replace jobs, predicting that the technology will go after call center jobs, administrative and support roles and marketing positions first. He also said that AI would lead to “a reduction” of roles in the software engineering field and impact the legal space. According to Precedence Research, the global legal services market size was close to $1 trillion in 2024 and is projected to reach $1.55 trillion by 2034.

“I think that’s a market that could see a huge deflation in volume of people necessary, from contracting to reviewing documents and litigation to giving sound written legal advice,” Daniels says.

According to Daniels, it takes more than a superficial understanding of AI to thrive as the technology takes over more tasks. He says workers need to develop three critical skills to adapt to AI: critical reasoning, deep domain expertise and design thinking.

Critical reasoning

Recent studies from MIT and other institutions show that relying heavily on AI tools can lead to a drop in critical reasoning skills, mostly because people start to offload thinking to the technology instead of engaging deeply with content on their own.

However, Daniels argues that if AI is used correctly, it actually demands deeper critical analysis, not less. To get the most out of AI, users need to understand its limits and actively verify the information provided. Good results come when people combine their own judgment with AI, rather than letting the technology do all the thinking, he says.

Related: Nearly 95% of Companies Saw Zero Return on In-House AI Investments, According to a New MIT Study

Daniels gave the example of writing a research paper with AI. Though it may be possible to copy-paste the paper from an AI chatbot like ChatGPT, the AI output could still have fundamental issues. For example, the paper could be focused on a specific sector of the technology market, and AI will pull facts about the overall technology space and assume they are true for that sector.

“If you don’t understand those assumptions that the AI is relying on to return you a result, then you’re going to get bad results,” he says.

Using AI correctly requires critical reasoning in the form of checking sources, questioning conclusions and ensuring that the information truly fits your topic and context. Otherwise, important distinctions get lost, and results can be off-base, Daniels explains.

For example, last month, Deloitte was caught using AI in a $290,000 research report commissioned by the Australian government, leading to a partial refund and a public apology. The AI model cited non-existent books, fake court cases and fabricated quotes in the report.

Deep domain expertise

Using AI effectively isn’t just about knowing the technology — it also requires deep domain expertise, or specialized, in-depth knowledge and practical experience in a field. Users have to understand the specific challenges, context and nuances of their field to get accurate results.

Daniels gives the example of building AI agents, or autonomous software systems that can operate independently without constant human input. Over the summer, he spent nights and weekends building AI agents that could perform different functions inside Exiger’s software. He used AI to help him code a lot of those agent flows.

Related: Perplexity CEO Says AI Coding Tools Cut Work Time From ‘Four Days to Literally One Hour’

As Daniels was building those agents, he realized that he was asking the AI to make too many decisions at once. For example, he would ask it to do three things and then compare them, or make multiple decisions and give him a score.

“I realized as I was building those agents that I had to subdivide the task into the individual decisions that I would make,” Daniels says.

Knowing each part required domain expertise, which Daniels has. He highlights that AI can process huge amounts of information, but unless you guide it with detailed knowledge, the technology may deliver off-target or misleading answers.

“Knowing what decisions need to be made in a process, knowing the data that you can utilize to make those decisions, and then being able to thinly slice them into a decision process or a decision tree, that is going to be a critical skill,” Daniels says.

Related: This Tech CEO Isn’t Using AI to Screen Resumes — Here’s Why

Design thinking

Daniels said AI requires design thinking, meaning users need to know what outcome they want to achieve, visualize the process and structure their approach before using the technology.

Instead of treating AI as a simple tool, design thinking means intentionally planning by mapping out the steps and imagining the end result. This mindset lets users build workflows that tap into AI effectively.

A research paper published in the International Journal of Innovation Studies last year found that design thinking “can work in harmony” with AI to yield “creative solutions” to common problems.

“Being able to develop design thinking skills is going to be absolutely and utterly critical for the future,” Daniels says.

Related: Microsoft AI CEO Warns That ‘Dangerous’ and ‘Seemingly Conscious’ AI Models Could Arrive in the Next 2 Years