Startup Founder Shares Where to Leverage AI, and What It Can’t Replace

This as-told-to essay is predicated on a dialog with Tim Desoto, a 49-year-old founder and CEO, based mostly in San Francisco. The next has been edited for size and readability.

I have been engaged on my startup, an AI-powered purchasing platform, since late 2024. With all of the developments within the tech trade, folks may consider AI as a hammer, treating every little thing else as a nail, and in my expertise, that simply would not work.

I haven’t got a tech background, and since beginning my enterprise, I’ve realized lots about the place to leverage AI, and the place to not. I am at all times making an attempt to be versatile about when to modify from AI to human intervention, and vice versa.

Over time, it is change into clearer to me the place and the place to not use AI for my enterprise.

I depend on my community and social media to assist me resolve which AI instruments to make use of

I am lucky to dwell in San Francisco. I am going to meetups to listen to what persons are utilizing, and I’ve attended some developer conferences. My ears at all times perk up once I hear {that a} new instrument or model of one thing is working for anyone else in my community.

Proper now, there is a robust concentrate on agentic workflows. OpenClaw generated buzz as an open-source autonomous agent challenge, and Moltbook amplified that consideration by making agent-to-agent interplay seen in a social atmosphere. Claude Cowork can also be gaining traction, significantly amongst groups on the lookout for enterprise-ready agent workflows with clearer guardrails.

Past the brokers themselves, the main focus is shifting from “what can brokers do?” to “how will we run them reliably and securely at scale?”

Whether or not I am taking a look at X, LinkedIn, or different platforms, there’s plenty of actually nice work being executed to share these updates.

I take advantage of AI for all the standard time-saving duties

My paid stack consists of enterprise plans for models such as Claude Max, Gemini Extremely, and ChatGPT Enterprise, together with AI-powered improvement and productiveness instruments reminiscent of Cursor, Figma Make, Notion AI, Superhuman Ask AI, and Lovable.

Gemini’s image models have change into unimaginable. The newest updates to the brand new mannequin have actually improved every little thing. I seen quicker efficiency, extra steady reasoning, and stronger multimodal capabilities, particularly in picture era. I used to be impressed by how constant the pictures remained throughout modifications. I’ve even seen some enhancements within the responses it offers about real-time data.

I divvy up which instruments I am utilizing based mostly on the place the most recent developments are, and use them both as they have been designed or as I feel I can use them in my present operational move. For instance, I take advantage of Lovable, an AI website builder, to make slide decks.

My ‘AI conveyor belt’ train helps me take a look at totally different AI fashions

Normally, I begin with a written immediate, then go multimodal, speaking out loud to the mannequin. I am going to discuss backwards and forwards with it about my concept and attempt to get the agent to push again as a result of I do know that some AI models are typically extra agreeable.

As soon as I get an output that I am proud of, I take advantage of a special mannequin to get a special view. For long-form evaluation and structured insights, I lean towards Claude and Gemini. Gemini’s inline supply linking is especially helpful for verification and deeper analysis. For structured reasoning and formal writing, I primarily use ChatGPT and Claude.

Generally, I am going to push a doc out to a number of fashions on the identical time and see what comes again concurrently. For artistic exploration and multimodal work, I take advantage of each Gemini and ChatGPT to generate early-stage ideas, mockups, and visible inputs. Some fashions are higher than others at sure duties, however I am at all times getting a extra well-rounded perspective by feeding content material to a number of fashions.

The method can take as little as quarter-hour, or higher-impact selections can span a number of hours to a few days, relying on complexity.

I ultimately wanted human builders

When I was vibe coding the alpha model of my product, I might hit spots the place 30 or 40% can be improper. I did not know what precisely the issue was. I might have a number of screens working the code to determine it out, and I might proceed to make use of AI against AI till I may get to about 95% confidence.

I contracted a number of builders to assist transfer my product ahead. Now I’ve a product creating at a quicker, way more strong, and scalable fee.

As a lot as I can do with AI, it is wonderful what technical folks can do with AI instruments {that a} non-technical particular person cannot.

The human perspective remains to be irreplaceable in my enterprise

I reached out to plenty of casual mentors and buddies within the area who could possibly be useful early on in my course of, however lately I formalized having advisors to bounce issues off of. That has been an enormous feather in my cap.

These advisors have their very own experience to attract on, and so they know many good folks on this area engaged on initiatives which have helped us establish potential blind spots. It is helped me join with potential companions in ways in which I feel can be tougher as a solo founder.

I really feel like I’ve a clearer view now of what I can belief AI for, in comparison with what I assumed originally of my journey. Whereas issues like hallucinations and agreeability could be mitigated, long-term strategic judgment and style nonetheless require human oversight.

AI can generate prospects, however selecting the best route stays a human duty.

Do you might have a narrative to share about working an AI-powered enterprise? Contact this reporter, Agnes Applegate, at aapplegate@businessinsider.com.