AI Explored – "Building AI Agents: A Proven Process"

Host: Michael Stelzner
Guest: Sara Davison, Co-founder of AI Build Lab & Mavenly
Date: November 4, 2025

Episode Overview

In this episode, Michael Stelzner dives deep into the world of AI agents with Sara Davison, a leading practitioner and educator in building AI-powered workflows for businesses. The conversation covers the evolution from AI assistants to true AI agents, addresses common misconceptions, and provides a step-by-step methodology for implementing agentic systems that scale unique business processes. Practical examples, critical considerations about trust, and an open discussion on quality, tools, and human oversight make this a rich listen for marketers, creators, and business owners aiming to harness AI's transformative potential.

Key Discussion Points & Insights

1. Sara Davison’s Journey into AI (02:02–04:14)

• Sara's Entry into AI:
  • Self-taught, non-traditional background.
  • Entered AI via curiosity sparked by the launch of ChatGPT.
  • Formed a partnership with Tyler through shared interests and community engagement on platforms like TikTok.
• Building the Practice:
  • Gained a reputation for tackling complex real-world AI challenges.
  • Developed a structured, teachable system through repeated project implementation.

Quote:
"We got known for being the people that solved the really hard problems in AI... We know how to actually make things work in the real world, not just in theory."
—Sara (03:38)

2. Demystifying AI Agents versus Assistants and Automation (04:49–07:59)

• Common Misconceptions:
  • Many confuse chatbots, automations, and agents—they are not the same.
  • AI agents are about autonomy, not just automation or text responses.
• Autonomy Defined:
  • AI assistants (e.g., ChatGPT): User-driven, require explicit input for every step.
  • AI agents: Delegated actors, capable of reasoning, planning, and executing with minimal human intervention.

Quote:
"AI agents are essentially AI tools with autonomy. They have the ability to reason, plan and execute... almost like you're delegating to an AI team member."
—Sara (05:46)

3. The Business Value of AI Agents (07:10–09:54)

• Scalability:
  • AI agents allow for essentially infinite scaling—think virtual teams or departments.
• Customization and Intellectual Property:
  • True value emerges when businesses encode their 'secret sauce' into agents, protecting unique workflows and IP.

Quote:
"If everybody could access that same power... what is the moat anymore, right? ... The ability to codify... your process, your workflow, your business so unique... is something your competitors cannot copy."
—Sara (09:10)

4. Trust, Security & “Evolution Not Revolution” Mindset (10:05–13:27)

• Security Concerns:
  • Many fear sharing IP or “secret sauce” with AI systems.
• Sara’s Philosophy:
  • Gradual trust-building is key—don’t give full autonomy all at once.
  • It’s not about “revolutionizing” overnight, but about evolving processes with careful oversight.

Quote:
"Our whole philosophy... is what we call evolution, not revolution. Because you are essentially giving these tools autonomy to go and reason, plan and execute on your behalf."
—Sara (10:11)

• Market Disruption:
  • Traditional businesses risk being overtaken by startups leveraging agentic tools more quickly and flexibly.

Quote:
"You've got 12 to 18 months max and your business, as it looks like, will not exist... That could be both a good thing and... an opportunity for massive reinvention."
—Sara quoting Daniel Priestley (13:29)

5. Step-by-Step: How to Build Effective AI Agents

Step 1: Deep Dive Into Workflow Analysis (15:08–17:43)

• Don’t Start With Agents!
  • Begin by mapping current workflows, understanding each role, what breaks, and what makes key players effective.
  • Document both explicit standard operating procedures (SOPs) and the “magic” intuition of team members.

Quote:
"Without understanding that, you're just essentially dragging and dropping nodes in a tool... that doesn't capture what actually happens in real life."
—Sara (15:28)

Step 2: Engage Stakeholders & Uncover Operational Reality (21:44–24:31)

• Beyond SOPs:
  • Discover differences between documented procedures and what actually occurs (“operational reality”).
  • Involve key staff (i.e., “the Sues and the Toms”) in co-developing and testing prototypes to reduce resistance.

Quote:
"The SOP doesn't capture why Sue is so great at her job or why Tom can intuitively look at an application and make a gut check... Most people stop at the SOP and get bland results."
—Sara (21:54)

Step 3: Prototype & MVP Development (“Show Up with a Mockup”) (26:00–29:08)

• Rapid Iteration:
  • Build quick minimum viable prototypes, gather real feedback from users.
  • Use this cycle to refine, uncover missing steps or specialization needs.

Quote:
"These tools are so easy to pull together now that we go and provide like a mini MVP... Tell us what you think, what's missing?... That is a circular process that helps us define how we can make our agents better."
—Sara (26:00)

• Tools Mentioned:
  • Typing Mind: Unified playground for major LLMs.
  • Cassidy AI: Accessible agent-building platform, allows flipping between different models.
  • Advisor: Stay “tool and platform agnostic,” as tech is evolving rapidly (27:05).

Step 4: Iterate Toward Autonomy (The Assistant-to-Agent Spectrum) (29:22–31:08)

• Gradual Autonomy:
  • Don’t “flip a switch”—incrementally give systems more permissions.
  • Incorporate context, memory, multi-step planning, and team-like structures (including “orchestrator” or “boss” agents for coordination & QA).

Quote:
"It's an iterative process. It sounds like, oh, you just flick a switch and it becomes an agent. But it's actually a spectrum of how much autonomy that you, you give your agents."
—Sara (29:22)

Step 5: Example—Agentic Workflow for Document Generation (“DocGen”) (31:56–38:55)

• Practical Case:
  • Example from Tyler’s family greenhouse business.
  • Challenge: Custom market garden business plan generation—previously required weeks from top experts.
  • Solution: Orchestrator agent divides the task among domain-specialized AI agents, integrates transcripts, generates a comprehensive, customized document in minutes.

Quote:
"Our agentic team... infinitely scaled Roger’s expertise... to create this new line of business."
—Sara (35:32)

Step 6: Quality Control and Human-in-the-Loop (38:55–41:08)

• Critical Safeguards:
  • Always involve human review before final outputs (“never auto-send” to clients).
  • Use evaluation methodologies (Evals) to score and improve agent outputs over time—forms a learning feedback loop.

Quote:
"We are very strong in having what is actually called in AI, a human in the loop."
—Sara (39:14)

Step 7: Tool Flexibility and Continuous Maintenance (41:20–43:47)

• Tooling Is Not the Bottleneck:
  • Any tool (Typying Mind, Cassidy AI, n8n, Make, Zapier) can be used if you understand workflow and logic.
  • Requires ongoing maintenance to handle changing models, company procedures, and prompt drift.

Quote:
"Honestly, it's not even about the tools. It's about the mental models, frameworks, approach that you have... The tools are just piping."
—Sara (41:08)

Notable Quotes & Memorable Moments

• On Building Trust in AI Systems:
  "There isn't that gradual process of building trust and reliability in these agents to get to a point where you can then provide that autonomy."
  —Sara (10:50)

• On Staff Fears:
  "We really try to... engage people who are involved. We work alongside them developing this stuff."
  —Sara (23:37)

• On Maintenance:
  "There's things that happen called prompt drift... [and] you talked about hallucination, large language... There's a lot that actually goes into it... it's not a one and done thing."
  —Sara (43:13)

Timestamps for Key Segments

• Sara’s AI Journey & Background: 02:02–04:14
• Difference Between AI Agents & Assistants: 04:49–07:59
• Business Value & Competitive Advantage: 07:10–09:54
• Overcoming Trust & Security Fears: 10:05–13:27
• Where To Begin: Workflow Deep-Dive: 15:08–17:43
• Example—Document Generation (“DocGen”): 17:43–20:12; 31:56–38:55
• Prototype Development & Tooling: 26:00–29:08; 27:05
• The Autonomous Workflow/Orchestrator Agent: 31:56–38:12
• Quality Control & Human Oversight: 38:55–39:40
• Tool Flexibility and Frameworks: 41:08–42:38
• Ongoing Maintenance Needs: 43:03–43:47

Final Takeaways

• Effective AI agents aren’t off-the-shelf; their creation requires deep workflow understanding, stakeholder collaboration, iterative prototyping, and strong quality control loops.
• The true business advantage lies in encoding unique organizational know-how into agentic systems, not in generic automation.
• While the tools are now accessible, success depends on processes, methodology, and continuous involvement—AI agents deliver best when they augment (not replace) human strengths, with ongoing human oversight.

Connect with Sara: [LinkedIn – Sara Davison]
Mavenly Course: "How to Scale a Business with AI and Agentic Workflows"

For full show notes, visit: socialmediaexaminer.com/aipod