Why Smart Builders Need Better AI Workflows, Not Just More AI Tools
In this Tinker Club session, Leigh-Anne Nugent and Micah Adler dig into a tension many builders are feeling right now: AI tools are evolving at a wild pace, but that does not automatically make business systems better. From GPT-5 and Claude to Copilot and Salesforce-style guardrails, this conversation is a practical look at what happens when hype collides with workflow reality, and why thoughtful implementation still matters most.
LESSONS YOU CAN TAKE FROM THIS:
1. New AI releases can improve outputs and still disrupt workflows
One of the clearest themes in this conversation is that model upgrades are not always a clean win. A new version may be more capable in some areas, but less predictable in others. That creates a real operational challenge for businesses relying on consistent outputs, especially when prompts, tone, or workflow behavior start changing without warning.
2. Structured automation and open-ended AI are not the same thing
Leigh-Anne makes an important distinction here: many businesses say they want AI, but what they actually need is predictable automation. If the goal is a repeatable outcome, a well-structured workflow may be more useful than an agentic system that changes its behavior over time. That difference matters when accuracy, governance, and customer-facing consistency are on the line.
3. AI tools still need active monitoring and real governance
A major takeaway from this session is that no serious AI workflow can just be “set and forget.” Models change. Features get deprecated. Responses shift. If a business is using AI in production, especially in customer-facing scenarios, someone needs to monitor it, test it, and keep refining the guardrails, directives, and risk controls around it.
4. The best tool depends on the job, not the branding
Micah’s comparison of GPT-5, Claude, and Copilot highlights a useful reality: no single tool is winning every use case. Some tools are stronger for structured project work, some are better for coding, some handle memory better, and some still feel too fragmented. Builders need to evaluate tools based on how they actually perform inside real workflows, not just based on launch buzz.
KEY TAKEAWAYS:
AI model updates can improve quality while still breaking workflow consistency.
Businesses often need structured automation more than unpredictable agent behavior.
AI in production requires ongoing monitoring, testing, and governance.
Tool selection should be based on workflow fit, not hype.
Human oversight is still essential, especially in customer-facing AI use cases.
