You Can't Prompt the Room: The Last Skill AI Won't Replace - Balázs Horváth, VisualLabs
AI has shifted the software development bottleneck from coding to defining *what* to build. This talk emphasizes the critical role of human skills in eliciting true requirements and understanding business value, detailing practical tools like story mapping, a 4-question value framework, and the VAD thinking path to ensure valuable AI-driven solutions are built, focusing on impact over mere feature delivery.
The central premise is that with AI rapidly advancing, writing code is no longer the bottleneck in software development. Instead, the critical and most expensive part of the lifecycle has shifted to the “people work”: understanding what truly needs to be built by eliciting real requirements from stakeholders. This requires human skills that AI cannot replicate, as “you can’t prompt the room.”
The speaker, Balázs Horváth, illustrates this with VisualLabs’ internal hackathon, where 21 agent ideas were generated, but 17 were abandoned. These ideas failed not due to technological limitations, but because they lacked clear business value, data access, or a defined owner. The four successful agents, now in production, highlight the importance of building what’s truly “worth building.” Drawing an analogy to Henry Ford’s “faster horse” versus a “car,” the speaker emphasizes that AI, by its nature, provides the most common answers. True innovation, a “magnitude shift better,” requires human insight to identify unarticulated needs and problems.
This shift means that the ability to write good code is no longer the paramount skill. The focus has moved to the “analysis toolkit,” encompassing skills traditionally associated with functional consultants, business analysts, and design thinking. These “old skills with new economics” are crucial for navigating the AI era.
Three practical tools are presented to facilitate this re-focus:
- Story Mapping: This technique helps capture the process backbone and user stories at a high level. By mapping out stages (e.g., in a support system: contacting, triaging, resolving, closing), teams can identify user stories beneath each stage. It’s intended to remain high-level to provide the big picture, allowing for the definition of Minimum Viable Products (MVPs) and subsequent backlog items. User stories, following the structure “As a [persona], I need to [what] so that [why],” are particularly effective as AI is trained on such patterns and can generate better results when provided with structured input, including acceptance criteria. Daisy-chaining these stories creates coherent system specifications.
- The 4-Question Value Framework: To ensure that the right problems are being solved and to elicit better responses from AI, these questions guide the requirements process:
- Whose problem is this? (Identify a direct, quantifiable persona.)
- What does winning look like for them? (Define successful outcomes and how they can be achieved.)
- What would make them refuse to use it? (Identify barriers like platform unavailability, usability issues, or data security concerns.)
- Would it change a decision? (Focus on impacting user decisions for better outcomes.)
- VAD Thinking Path (Value to Architecture to Design): This disciplined approach ensures that projects are grounded in real needs.
- Value: Begin by understanding how value is created, what constitutes it, and the current process flow.
- Architecture: Investigate the underlying systems that support this process.
- Design: Only then, design the system to best support the identified value and process, considering necessary process changes.
The speaker highlights the importance of shifting away from anti-patterns like high feature velocity with low adoption, or the “demo is the deliverable” mindset. Instead, organizations should prioritize “features used more than twice” over simply “features shipped.” This requires moving “smartest people towards our customers, towards the business problems,” involving subject matter experts in decision-making, and conducting mapping sessions (like user story maps or business model canvases) before any building commences. The ultimate goal is to “build the right thing, and not just the next thing.”