GenAI gunslinging vs. architecting: when to use each approach

Executive overview

Most people prompt AI by firing off a quick request and refining through back-and-forth — gunslinging. For simple tasks this is fine, but for complex outputs it wastes time and rarely hits the mark.

Architecting replaces iteration with deliberate upfront collaboration: align on goals, establish best practice, agree on structure — then generate.

The core insight: design the house before you build it.

What gunslinging looks like

• Quick prompt, shoot from the hip, hope for a good result
• Output rarely hits the bullseye first time
• Leads to lengthy back-and-forth refinement
• Can feel like arguing with AI to get what you want
• Fine for simple, low-stakes tasks like emails

What architecting looks like

• Share goals explicitly and confirm AI understands them
• Work with AI on best practice and structure before any drafting
• Challenge AI's ideas; get it to challenge its own
• Only generate the final output once planning is solid
• For long documents: agree on the index first, then build section by section

When to use each

• Gunslinger: quick emails, basic tasks, speed over quality
• Architect: strategy papers, reports, workshop design, training sessions, anything needing excellent output
• Logic models (o3, o4) perform noticeably better under the architecting approach
• Architecting works across all major AI tools; especially effective in Copilot

Worked example: designing a team offsite

• Start with context and goals; confirm AI has understood them
• Ask AI to surface relevant best practices and science on team performance
• Explore structure and activity options before committing to any output
• Review and refine the plan until it's solid
• Only then build the workshop — segment by segment, not all at once