The 16-Hour OpenClaw Obsession: Why You Need to Start Today

In this week's Proc & Roll, Conrad Smith and Zachary Bachir sit down with Kasey Smith for a special jam session on OpenClaw and agentic AI. If you think this is just another buzzword, think again. This technology is moving so fast that it might fundamentally squash the legacy software layers procurement has relied on for decades.

Here is a breakdown of the episode's most critical insights and why you need to start experimenting with AI agents today.

Meet "Gavin": The Sourcing Agent Built in 90 Minutes

• No Coding Required: Kasey Smith, who openly admits he is not highly technical, successfully set up an OpenClaw agent on a Mac Mini using natural language prompts, or "vibe coding".

• End-to-End Automation: In just 60 to 90 minutes, Kasey built a contract management agent named "Gavin".

• The Workflow: Gavin operates on a "heartbeat" schedule, constantly checking a designated drive folder for new contracts. When a contract is dropped in, Gavin analyzes it against corporate policy, identifies exceptions (like payment terms exceeding Net 30), and automatically drafts a remediation email to the supplier.

• Seamless Handoffs: If the contract is fully compliant, Gavin bypasses the supplier and routes the contract directly to legal for signature via a Slack message. All of this was constructed through semantic text conversations rather than traditional drag-and-drop workflow builders.

The Death of the Traditional CLM

• Replacing Legacy Tools: Zachary Bachir made a bold prediction during the episode: traditional contract lifecycle systems like Ironclad might be finished.

• The ROI Shift: Conrad noted that the ROI of spending hundreds of thousands of dollars on a CLM platform disappears when your team can build an AI agent in a few days that successfully handles 80% to 90% of your edge cases.

• Squashing the Software Layer: Instead of forcing business users to navigate complex databases like SAP or Coupa, agents will execute these workflows in the background. Users will simply state what they need in a Slack or Teams channel, and the agent will handle the orchestration.

Context Engineering: Boss Agents vs. Sub-Agents

• The Cost of Complexity: While it is tempting to give an AI agent access to everything, Zachary warned that poorly managed context windows can lead to mistakes and high processing costs, noting he once blew through $200 in a single day.

• The Sub-Agent Strategy: The most efficient operating model is to design a "boss" agent that delegates specific tasks to specialized "sub-agents".

• Optimizing Models: Sub-agents can use smaller, more cost-effective models for simple administrative tasks, only escalating to expensive, high-powered models (like Opus) for highly complex reasoning.

Don't Wait for IT—Start Now

• The Security Bubble: Corporate IT departments are not going to let you drop a rogue Mac Mini into their enterprise security bubble anytime soon.

• Build Your Sandbox: However, Kasey and Conrad strongly urge procurement professionals to set up a sandbox environment at home and start learning.

• Lead the Charge: By learning to build and guide these agents on your own time, you will position yourself to lead the enterprise conversation when the organization is finally ready to deploy them.

Transcript: Proc-N-Roll | The 16-Hour OpenClaw Obsession: Why You Need to Start Today

Conrad: Hello everybody, and welcome to Proc & Roll. Today we have a super special jam session on OpenClaw and agentic AI. I'm joined by Kasey Smith, our head of operations, and Zachary Bachir from KPMG. This technology is completely different from what we discussed a year ago, and it's going to change everything about procurement and your tech stack. Kasey, tell us about your journey into OpenClaw.

Kasey: I ran to the Apple store, bought a Mac Mini, and stood it up despite not being technical at all. The most coding I do is "vibe coding" using natural language. You can literally open ChatGPT or Claude, tell it you want to set up an OpenClaw assistant, and it will guide you step-by-step through the terminal commands.

Zachary: The most interesting thing is that these are computer-use agents that can operate your PC and web browser. Will we have control agents managing systems, or just deploy agents that do anything a human can?.

Kasey: Let's jump into a demo. I built an agent named "Gavin" in about an hour. Gavin runs on a heartbeat schedule, constantly monitoring a Google Drive folder for new contracts. He analyzes the contract against our rules, finds policy exceptions, notifies me (the CPO) in Slack, prepares an amendment, and even drafts an email to the supplier requesting changes.

Conrad: If you took two days to build this out, you could create a very strong Contract Lifecycle Management (CLM) platform. The idea that you have to pay hundreds of thousands of dollars for a legacy CLM system won't hold ROI anymore.

Zachary: The ironclads of the world are finished; agents can do this out of the box. However, as you cater to more edge cases and build complex workflows, you hit context limits. Feeding an agent too many policies at once causes mistakes.

Kasey: Exactly. That's why you use a "boss" agent that delegates tasks to specialized "sub-agents". It is much faster, less token-intensive, and more accurate than having one massive brain try to search everything at once. Sub-agents can even use cheaper models for simple tasks and only escalate to complex models like Opus when needed.

Conrad: Show us how easy it is to update Gavin's rules. People get intimidated by this, but it's just chat.

Kasey: I'll feed Gavin our rules document and ask, "Based on our rules, what do you think we are missing?". He analyzed it and suggested adding a rule for Intellectual Property (IP) ownership. I simply replied, "Please go add number two in the rules document," and he updated the Google Doc automatically. Even cooler, when we told Gavin to notify Zach about IP violations, the agent checked our org chart, realized Zach wasn't in the directory, and asked for clarification on who he is and whether to add him to our Slack channel.

Zachary: That is the intelligence of the underlying models at work. Ultimately, I believe agents must replace the entire legacy tech stack, like the Coupa stack. Users won't need to interact with databases; the agents will build and query them in the background.

Conrad: Let's wrap up with practical advice. How do procurement leaders get started?.

Kasey: Start experimenting in a safe sandbox environment. Your enterprise IT is not going to let you drop an OpenClaw bot into their security bubble right now. Do it on your own time, build faux scenarios, and prepare to demonstrate its power so you can lead the conversation when your organization is ready.

Zachary: Get into it in your personal life. You don't need to do a lot of research; just ask the LLM to guide you. You will pick it up very quickly.

Conrad: Don't be intimidated, folks, you can do this. We want to help you join the revolution of automation. Find us on LinkedIn, engage with us, and let's create a better future for procurement together.

This transcript has been edited for clarity while maintaining all substantive content