Will AI agents be employees of the year in 2026?
Only time will tell. One category that AI absolutely dominated last year was highly congenial colleagues. While this may sound like a good thing, it can be a problem for founders who rely on AI as their only teammate.
When legal, human resources, and supply operations are all led by AI agents, unsubstantiated flattery can create costly blind spots. This is one of the reasons OpenAI said goodbye to the “yes man” version of ChatGPT, And why are some AI-powered solo founders training their tools to fight back?
Three executives who rely on AI every day shared with Business Insider how they built workflows and prompts to challenge their AI employees with their ideas. Quotes have been edited for length and clarity.
I avoid asking the AI for its opinion and instead have it rate my ideas on a scale of 1 to 10.
Business owner Yeshim Saidan. Yesim Saidan
Yesim Saidan She is a branding and communications professional in her early 50s based in the Netherlands.
Everything changed for me when OpenAI launched custom GPT. We’ve created over 17 custom GPTs and built our team using this feature. Next, I thought of my ideal mentor and created a custom GPT for that mentor.
When you’re stuck on a business decision or need to come up with a creative idea or strategy, brainstorming usually starts with Steve Jobs’ Custom GPT. When encouraging questions, avoid questions like, “What do you think about this idea?” Because AI usually agrees with me and wants to please me. Instead, ask, “On a scale of 1 to 10, how good is this idea?”
I’m not saying this idea is bad, but you might say it’s a 5 now. Then ask, “Okay, what would make a 10?”
Usually that’s when I start leveraging the Steve Jobs experience I’ve been training with. Iterate over and over until you get the most helpful and honest feedback possible. Depending on the task, we typically do 3-5 rounds of adjustments to get a more strategic deliverable.
Created a governance structure and a roundtable with AI agents to get a better response
business owner Aaron Snead; BI’s Michelle Bruzzese
Aaron Sneed He is a 40-year-old defense technology solo founder based in Florida.
When I started my business, I realized I didn’t have the money to pay lawyers, HR people, and many other companies. So I used AI to create what I call a “council.”
My AI council consists of:
- acting chief of staff
- human resources agent
- financial agency
- accounting agent
- Legal, communications and PR agency
- Security and compliance agent
- engineering agent
- quality agent
- supply chain agent
- training agent
- manufacturing agent
- business system agent
- facility agent
- field operations agent
- IT and data agents
You don’t need to have a lot of yes agents. I intentionally trained them to push me back. I want them to test my theory to help me with what I’m trying to accomplish. Training never really stops.
The models are better, and so are my prompts. We now have a better understanding of what information should be included in agents, including the priority governance structure. I have a series of files setting out these requirements to reduce the risk of illusions and false or fraudulent information.
I told the lead agent which model would be prioritized when making decisions. For example, legal, compliance, or security-related things are given higher priority. So I tell my chief of staff to listen to these models more than anyone else.
I have a roundtable set up in the project section of ChatGPT with all the AI agents, including the lead staff. There, you can put something like a request for proposal into the chat and all agents will review it at the same time. I use this roundtable as a precaution against hallucinations and lack of knowledge.
One way to reduce hearing only positive feedback is by using an “AI conveyor belt.”
Business owner Tim DeSoto. Tim DeSoto
Tim DeSoto is a 49-year-old founder and CEO based in San Francisco.
I have been working on AI since the end of 2024 to launch a startup. I don’t have a technical background, but since starting my business I’ve learned a lot about where to leverage AI and where not to leverage it.
Having the AI push what I’m building or ideating, both while I’m thinking of the idea and afterwards, is part of an exercise I call the AI conveyor belt. I usually start with a written prompt and go multimodal by talking to the model out loud. I know that some AI models tend to be more favorable, so I discuss my ideas a lot and try to get the agents to disagree.
Once you get a satisfactory output. Use another model to get a different view. I’ll throw in some documents and try going back and forth with the new model. In some cases, push documents to both models at the same time to see what is returned. Some people are better at providing feedback, researching, and annotating than others, but you can always get a better, more comprehensive view by feeding your content into multiple models at once.
Do you have a story to share about running your business with AI? Contact this reporter, Agnes Applegate: aapplegate@businessinsider.com.
