Podcast Summary: Embracing Digital Transformation
Episode #348 — The Future of Automation: Agentic vs Deterministic Workflows
Host: Dr. Darren Pulsipher
Guest: Sid Brott, Founder of Refound
Date: May 5, 2026
Episode Overview
This episode dives deep into the evolving landscape of automation, contrasting traditional deterministic workflows (like RPA) with the rising wave of agentic, AI-driven processes. Host Dr. Darren Pulsipher welcomes Sid Brott, an entrepreneur and longtime AI practitioner, to demystify these terms, explore real-world implementation challenges, and look ahead at how enterprises can leverage hybrid or heterogeneous workforces blending people, deterministic automation, and reasoning AI agents. The conversation is practical, jargon-light, and focused on actionable insights for public sector and enterprise listeners grappling with digital transformation.
Sid Brott’s Origin Story and Evolving with AI [(01:22)-(05:07)]
- Sid’s background: Began as an engineer, moved into consulting, then entrepreneurship in tech, marketing, and product management.
- Early AI experience: Worked with OpenAI before mainstream adoption, built one of the first companies on the GPT-3 API in 2020.
- Timing and growth:
- Sid notes, “It was so early that the models weren't all that great… companies ended up saying, ‘we could probably create better content than this’” [(03:44)].
- The platform’s capabilities rapidly improved, with Sid observing the difference between the slow start and sudden leaps: “It's very hard to know. Like when you're in an exponential curve… at any point it always looks like a linear curve” [(05:48)].
- Transition: Leveraged early-mover status to build ventures as models matured, and now leads Refound, a consultancy focused on agentic automation.
Automation: Deterministic vs. Agentic Workflows [(08:55)-(13:20)]
Deterministic (RPA, Traditional Automation)
- Definition: “Your typical RPA and automation in general is [a] very predetermined workflow… if this happens, then do that… so you can automate a lot of processes like this.” — Sid Brott [(09:16)]
- Use cases: Best for processes with clear, invariant rules and few exceptions.
Agentic Workflows
- Definition:
- “The agentic workflow is where there's a lot of variance in what's happening that you can't just deterministically say if this happens, do that… when a workflow has more exceptions than rules.” — Sid Brott [(09:16)-(10:06)]
- Involves reasoning models able to evaluate unique, unscripted situations and determine the best course of action.
- Example: In customer support, standard requests (“where is my shoe?”) are deterministic; handling complex, ambiguous refund claims requires agentic reasoning.
Memorable Quote:
“I love how you said that—a workflow that has more exceptions than rules. Absolutely, I get it. That makes sense.” — Dr. Darren [(10:06)]
Blending Approaches: Hybrid/Heterogeneous Workforces [(13:20)-(16:33)]
- Systems In Practice:
- “You can just have questions that are very basic go straight through regular automations, and then questions that are more complex go to a reasoning model.” — Sid Brott [(13:00)]
- Trade-offs:
- Agentic workflows can be less predictable and possibly costlier per transaction, but dramatically reduce human workloads by automating previously “un-automatable” exceptions.
- “Maybe 20% of the questions are being automated and 80% handled by a human. For agentic systems it’s 80% handled by the AI and 20% by human—or maybe more.” — Sid Brott [(13:46)]
- New workplace paradigm:
- Dr. Darren suggests “heterogeneous workforce” as more accurate than “hybrid”: “I've got humans, I've got agentic flows, and I've got deterministic workflows... This is actually pretty cool.” [(14:29)]
Case Study:
Nike customer refund — traditional system sends all exceptions to a human. Agentic AI now evaluates customer history, order value, and can recommend, for instance, a 70% refund with a final human review only for confirmation.
Orchestrating The New Workforce [(17:39)-(18:29)]
- Workflow architecture:
- A system now must route each case to deterministic automation, agentic AI, or a human.
- Could be “an AI orchestrator, or...basic AI classification that just looks at what's coming in and classifies [the request]” — Sid Brott [(18:12)]
Data Privacy & Deployment Concerns [(18:29)-(21:07)]
- Enterprise & Public Data Security:
- Options include using enterprise versions of OpenAI, Anthropic, Gemini, etc., which offer compliance (HIPAA, regionally located data centers, no data used for training).
- Open source models (Meta, Google Gemma, etc.) can be deployed on-premises if maximum privacy is necessary.
- “Open source technology is nearly as good as the paid.” — Sid Brott [(21:00)]
- Listener Scenario: Real-world patient info example, confirming privacy-compliant AI usage is possible.
What Makes Agentic Different [(22:00)-(23:56)]
- Classic chatbots vs. agentic AI:
- Traditional ChatGPT workflow: The user provides context and executes actions; AI just advises.
- Agentic workflow: The AI gathers context, does research, makes recommendations, and can execute actions; humans review or approve.
- “You tell the AI what to do. The AI does the work. That's the easiest way of working.” — Sid Brott [(23:14)]
- Power shift: “We're in charge with the agentic flows. We're in charge, and we need to make sure that we stay in charge.” — Dr. Darren [(22:58)]
Implementation Challenges & Organizational Change [(24:33)-(26:23)]
- Integration hurdles:
- Major challenge is unifying data across silos and systems (“the biggest problem is ...so many different data silos...to first get all of that data into a common data context layer...”) [(24:46)]
- Requires deep process understanding, mapping tools/integrations before AI agents can be built.
- Data access/security:
- Solutions run on the client’s infrastructure: “We build the agents into their existing servers...so it sits within their [data], within their servers and their architecture.” [(25:54)-(26:23)]
Consultancy & Change Management [(27:13)-(32:38)]
- Practical engagement:
- Start by diagnosing bottlenecks and workflow pain points.
- Process is akin to classic management consulting (“good old fashioned process consultant,” jokes Dr. Darren [(28:53)])
- Sid distinguishes their model: Instead of costly headcount-based solutions, they deliver automated workflows (agentic flows) that reduce recurring labor costs.
- Training & hand-over:
- Clients are trained to use and maintain the systems, with technical support available for custom changes.
- “If we've understood the process well enough...there's really not much else to do.” [(30:56)]
- Business model future:
- AI agents as scalable, repeatable resources (potential “AI resource rental” idea), but with necessary customization per client due to unique business processes.
Getting Started and Practical First Steps [(33:10)-(34:49)]
- Sid’s exercise for listeners:
- Map out the entire workflow of a key process.
- Identify what must remain human-led and what parts could be handled by agentic AI.
- Focus automation on the most repetitive, time-consuming, or exception-heavy steps.
- “Everything else that could become AI agent...especially looking at the parts that take the longest or [are] most repetitive, those are the lowest-hanging fruit to automate.” — Sid Brott [(33:54)]
Notable Quotes & Memorable Moments
- Sid Brott (- on exponential growth):
“When you're in an exponential curve at any point it always looks like a linear curve” [(05:48)]
- Sid Brott (on agentic workflows):
“Workflow that has more exceptions than rules” [(10:06)]
- Dr. Darren (on hybrid workforce terminology):
“Heterogeneous workforce ... because I'm using more than two a lot of times” [(14:29)]
- Sid Brott (on paradigm shift):
“You tell the AI what to do. The AI does the work.” [(23:14)]
- Sid Brott (on real deployment):
“The biggest problem is … so many different data silos … to first get all of that data into a common data context layer” [(24:46)]
- Sid Brott (best first exercise):
“Map out the whole process ... every step of the way, map it out ... decide after you map out the full workflow what should be human, what can go into AI ...” [(33:19)-(33:54)]
Key Timestamps
- [01:22] — Sid’s AI & entrepreneurial origin story
- [08:55] — Deterministic automation vs. agentic flows
- [10:06] — “Workflow with more exceptions than rules”
- [13:20] — Blending deterministic and agentic approaches
- [15:06] — “Heterogeneous workforce” concept
- [18:12] — AI orchestrators for workflow routing
- [19:09] — Data privacy & open source LLMs
- [22:00] — What makes agentic AI workflows truly different
- [24:33] — Deploying agentic flows: real barriers
- [28:53] — Consulting parallels
- [33:10] — First steps for automating your own processes
How to Get Started / Learn More
- RefoundAI: Visit refoundai.com to engage with Sid and his team directly.
- Action item from Sid: Map your process, identify automation candidates, and start with high-friction repetitive steps.
Tone & Takeaways
The conversation is candid, approachable, packed with practical scenarios and analogies to ground complex new technology in reality. Listeners will walk away understanding:
- The clear distinction between rules-based and reasoning-based automation.
- Why agentic workflows are just now becoming viable—and why most organizations aren’t operationalizing them (yet).
- How blending deterministic, agentic, and human-driven work is forming the backbone of the future enterprise.
- The practical path to start their own transformation.
“You tell the AI what to do. The AI does the work.” (23:14) — That’s the future of digital transformation.