This told essay is based on a conversation with Tim DeSoto, a 49-year-old founder and CEO based in San Francisco. The following has been edited for length and clarity.
I’ve been working on an AI-powered shopping platform startup since late 2024. With all the advances in the technology industry, people may think of AI like a hammer and everything else like a nail, but in my experience, that doesn’t work.
I don’t have a technology background, but since starting my business I’ve learned a lot about where to leverage AI and where not to leverage it. I try to be flexible about when to switch from AI to human intervention and vice versa.
Over time, I have become clearer about where to use AI in my business and where not to use it.
I rely on my network and social media to decide which AI tools to use
I’m lucky to live in San Francisco. I went to meetups to hear what people were using and attended several developer conferences. My ears always perk up when I hear that a new tool or version of something is working for someone else in my network.
The current focus is on agent workflows. OpenClaw has gained traction as an open source autonomous agent project, and Moltbook has taken it even further by visualizing interactions between agents in a social environment. Claude Cowork is also gaining traction, especially among teams looking for enterprise-ready agent workflows with clear guardrails.
Desoto’s startup Goodlife is an AI-powered shopping platform. Mr. Zachary Feinberg (BI)
Beyond the agents themselves, the focus is shifting to “What can agents do?” “How can we do this reliably and securely at scale?”
Whether you look at X, LinkedIn or other platforms, there’s a lot of really great work being done to share these updates.
We use AI for all common tasks that save you time
My paid stack includes business plans for models like Claude Max, Gemini Ultra, and ChatGPT Business, as well as AI-powered development and productivity tools like Cursor, Figma Make, Notion AI, Superhuman Ask AI, and Lovable.
The image model of Gemini has become incredible. Everything has been significantly improved with the latest updates for the new model. We noticed faster performance, stability of inference, and enhanced multimodal capabilities, especially in image generation. I was impressed with how consistent the images remained during the retouching process. I even noticed some improved responses regarding real-time information.
Desoto uses AI models like ChatGPT and Gemini every day to help run its business. Mr. Zachary Feinberg (BI)
I divide the tools I use based on the latest developments and use them as they were designed or what I think will work in my current work flow. For example, I use Lovable, an AI website builder, to create slide decks.
My “AI conveyor belt” exercise helps you test different AI models
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’re happy with the output, use another model to get another view. For long-form analysis and structured insights, I prefer Claude and Gemini. Gemini’s inline source links are especially useful for verification and deeper research. I primarily use ChatGPT and Claude for structured reasoning and formal writing.
In some cases, you might want to push documents to multiple models at the same time and see what is returned at the same time. For creative exploration and multimodal work, use both Gemini and ChatGPT to generate early concepts, mockups, and visual input. Some models are better at certain tasks than others, but you can always get a more comprehensive view by feeding content to multiple models.
This process can take as little as 15 minutes, but more impactful decisions can take hours or even days depending on their complexity.
Eventually we needed a human developer
Desoto Vibe coded an alpha version of the product before bringing in developers to speed up production. Mr. Zachary Feinberg (BI)
When we were vibecoding the alpha version of our product, we came across a spot where we were 30% or 40% wrong. I couldn’t figure out what exactly the problem was. Run your code on multiple screens to understand it and continue to use AI against AI until you reach approximately 95% confidence.
I have signed on several developers to move the product forward. We are now developing products at a faster, more robust, and scalable rate.
Just like what I can do with AI, engineers can do things with AI tools that non-technical people can’t do.
A human perspective is still invaluable in my business.
I reached out to a number of informal mentors and friends in this area who could have helped me early in the process, but I recently decided to formally have an advisor to help me through the problem. It was a huge feather in my cap.
These advisors have unique expertise to draw on and know many smart people in the field who are working on projects that can help identify potential blind spots. This helped me connect with potential partners in a way that I thought would be more difficult as a solo founder.
AI can do a lot for DeSoto, but it can’t replace human judgment. Mr. Zachary Feinberg (BI)
I feel like the extent to which I can trust AI is now clearer than when I first started my journey. Issues such as hallucinations and synchrony can be alleviated, but long-term strategic decisions and preferences still require human oversight.
AI can create possibilities, but it is up to humans to choose the right direction.
Do you have a story to share about running your business with AI? Contact this reporter, Agnes Applegate: aapplegate@businessinsider.com.
