Tim Desoto, a 49-year-old founder and CEO based in San Francisco, has detailed the precise operational role artificial intelligence plays in his AI-powered shopping platform startup, while outlining the critical areas where human judgment remains paramount. His insights come from over a year of development since late 2024, navigating the rapidly evolving landscape of agentic workflows and multimodal models.
Desoto, who lacks a formal technical background, has developed a methodical approach to integrating AI, constantly evaluating when to switch between automated and human-driven processes. "I'm always trying to be flexible about when to switch from AI to human intervention, and vice versa," he stated, highlighting a key learning from his entrepreneurial journey.
The Evolving AI Toolbox
Desoto's operational stack includes paid business plans for leading large language models such as Claude Max, Gemini Ultra, and ChatGPT Business. He supplements these with AI-powered development and productivity tools including Cursor, Figma Make, Notion AI, Superhuman Ask AI, and the website builder Lovable, which he uses to create slide decks.
He actively monitors the San Francisco tech scene, attending meetups and developer conferences to identify effective new tools. Currently, he notes a significant industry focus on agentic workflows, with projects like OpenClaw and Moltbook generating buzz, and Claude Cowork gaining traction for enterprise-ready applications. "The focus is moving from 'what can agents do?' to 'how do we run them reliably and securely at scale?'" Desoto observed.
A Multi-Model, Multimodal Workflow
Desoto's creative process typically begins with a written prompt before shifting to multimodal interaction, talking aloud to the model. "I'll talk back and forth with it about my idea and try to get the agent to push back because I know that some AI models tend to be more agreeable," he explained. To ensure robustness, he frequently feeds the same document to multiple models simultaneously to gain diverse perspectives.
He assigns specific strengths to different models: Claude and Gemini for long-form analysis and structured insights, with Gemini's inline source linking being particularly valuable for verification. He uses ChatGPT and Claude for structured reasoning and formal writing, and both Gemini and ChatGPT for creative, multimodal concept generation. He praised recent updates to Gemini's image models for their faster performance, stable reasoning, and consistent image generation during modifications.
The Limits of Automation and the Human Edge
Despite sophisticated use, Desoto encountered clear limitations during the "vibe coding" of his product's alpha version, where 30-40% of the AI-generated code was erroneous. "I would have multiple screens running the code to figure it out, and I'd continue to use AI against AI until I could get to about 95% confidence," he recounted. This experience led him to contract several developers, accelerating development at a "much more robust, and scalable rate."
He starkly concluded, "As much as I can do with AI, it's amazing what technical people can do with AI tools that a non-technical person can't." To address strategic blind spots, he formalised an advisory board, drawing on their expertise and networks to connect with potential partners—a move he believes would be "more difficult as a solo founder."
Desoto's key conclusion underscores a fundamental division of labour in the AI era: "AI can generate possibilities, but choosing the right direction remains a human responsibility." While issues like hallucinations can be mitigated, he asserts that long-term strategic judgment, taste, and oversight are irreplaceably human functions.