Key Takeaways

  • AI agents aren't killing SaaS; they're revitalizing it by interacting with existing software tools. Claire Vo admitted a "trough of despair" about SaaS, only to find agents excel at "pressing buttons" within applications.
  • Yana Welander's fashion startup, Yana Banana, uses AI agents like Codex with a tool called 'Computer Use' to operate specialized 3D design (CAD) and fashion software (like Clo) she hasn't personally mastered.
  • This agent-software partnership allows AI to perform complex tasks, like generating and fitting patterns on 3D models within Clo, which the AI alone couldn't do.
  • The approach unlocks previously impossible or practically infeasible tasks by automating the tedious parts, moving beyond simple code generation to context-driven execution.

The Method

Yana Welander's core insight reframes the AI-SaaS relationship: instead of AI replacing software, it becomes a power-user. Welander demonstrates this with her fashion startup, Yana Banana. She uses AI agents, specifically Codex, combined with a tool called 'Computer Use.' This combination allows the AI agent to operate complex, specialized software that Welander herself hasn't learned.

Think of it this way: the AI isn't writing new software or reinventing the wheel. It's interacting with existing, feature-rich SaaS products. Welander points to her use of a specialized fashion design software called Clo. This tool is designed to generate patterns and fit them onto 3D models. An AI alone might struggle with the precise geometric and fashion-specific logic required. However, with 'Computer Use' acting as the interface, Welander has “Codex go and use the software that I haven't learned how to use myself. And it does such a great job at something that it itself couldn't do.”

This "software-is-back" moment, as Claire Vo puts it, means agents aren't just coding. They're becoming expert operators of existing tools. Vo notes, "Do you know what? Like agents are really good at pressing buttons. So like a software is back for but agents are going to use it." This tactical partnership allows founders to tap into deep software functionality without the steep learning curve or custom AI development. It shifts the burden from a human learning complex UI to an AI agent executing tasks within that UI.

Where This Breaks Down

While powerful, this agent-driven approach isn't a silver bullet. The conversation hints at limitations. Precise pattern generation, for example, is still an area where AI alone might struggle. The AI agents are good at "pressing buttons," but the underlying logic and creative judgment often still need human oversight or input. Claire Vo also touches on the hidden complexity of "context." She says, “The hardest part of shipping with AI isn't the code, it's the context.” Agents need more than just software access; they need to understand “What's the right ticket? What did the spec say? What got decided in Slack?” Without this rich context, even an agent proficient in operating software can make irrelevant or incorrect moves. Simply giving an AI agent access to a tool without clear, well-defined objectives and contextual information will lead to bad outcomes. The system relies heavily on the quality of the prompt engineering and the initial setup to direct the agent effectively.

What to Do With This

This week, identify one tedious, complex task in a SaaS tool you use daily but haven't fully mastered. Instead of hiring an expert or spending hours learning the software yourself, explore an AI agent service that integrates "Computer Use" or similar capabilities. Give the agent a detailed, context-rich prompt and observe its ability to execute within your existing software. This is how Yana Welander gets things done "today" rather than waiting.