Vibe Coding 101: How to Build Your Own Procurement Tech

In this week's Proc and Roll, Conrad sits down with Zach for a jam session on the rapid evolution of Artificial Intelligence and how it is reshaping the procurement landscape. They discuss how AI has transitioned from a theoretical concept to a practical tool that empowers procurement professionals to build their own custom software solutions.  

Here is a breakdown of the main topics and takeaways from this exciting episode.

1. English is the New Coding Language

The most profound shift in technology today is that the traditional barrier to entry for software development has vanished.  

• No More Syntax: Zach emphasizes that you no longer need to learn traditional programming languages to build an application. English is now your coding language.  
• Rapid Prototyping: Conrad shares his own journey, noting that while he struggled to build AI applications a year ago, the models have become so smart that he could now build a functioning Contract Lifecycle Management (CLM) system in just a few days.  
• The AI Unblocker: If you run into a bug or a broken feature, you don't need to be an engineer to fix it; the AI agent will explain the issue in plain English and propose the necessary fixes for you.  

2. Design Before You "Vibe Code"

While it is tempting to jump straight into "vibe coding" and let the AI generate screens immediately, Zach warns that this often leads to scrapped projects.  

• Avoid Over-Engineering: If you give an AI a generic prompt like "build a contract database," it will likely over-engineer the solution with features you don't need yet, like native e-signature integrations.  
• Have a Conversation First: Spend significant time upfront chatting with your AI co-pilot to define the exact scope, resources, and phased approach of your build.  
• Focus on Taste: Now that anyone can generate code, the true differentiator is your "taste"—your ability to design a user interface and experience that your team will actually want to adopt.  

3. The New "Build vs. Buy" Reality

Procurement teams are constantly drowning in administrative tasks, but getting IT to build a custom solution or securing the budget to buy one can be a massive hurdle.  

• Take Control of the Tech Stack: Instead of paying $50,000 to $150,000 for an off-the-shelf CLM system, procurement professionals now have the power to automate and build these tools themselves.  
• Assess the Cost of Failure: The hosts advise evaluating the risk before building. You should never try to vibe code a core ERP system due to the massive financial and regulatory stakes. However, building a custom RFP tracker or an internal intake portal is a perfect, low-risk opportunity for self-serve automation.  

4. Accelerated Learning and the Human Advantage

Despite the incredible capabilities of AI, Conrad and Zach remain highly optimistic about the future of human workers in procurement.  

• A Masterclass in Systems: Building applications alongside an AI agent offers a profound learning experience, giving you an advanced understanding of systems and architecture in a matter of months.  
• Imagination is Irreplaceable: AI cannot magically whip up the perfect tool without clear direction. Humans are still required to bring imagination, process knowledge, and ultimate decision-making to the table.

Transcript: Proc-N-Roll | Vibe Coding 101: How to Build Your Own Procurement Tech

Conrad: Welcome back to Proc and Roll, your guide to practical procurement! Today, we have another jam session with the one, the only, Zach from London. This man has been playing with AI longer than most people I know in procurement. I continue to be dumbfounded by the amount of automation I can bring to repetitive tasks. It takes a little more time up front to set up an agent, but once it is running, it saves loads of time. Looking back 18 to 24 months, Zach, is AI moving faster or slower than you anticipated, and is it over or under delivering?  

Zach: It is moving incredibly fast, and most people are struggling to keep up. While everyone writes opinions about AI, the frontier labs keep moving forward, and competitive open-source models are coming out of places like China. Enterprises now have to figure out whether to deploy internal models, open source, or frontier models. The biggest shift is that the traditional coding language barrier is completely gone. English is your coding language now. You can build anything you want, and wherever you get stuck, the AI agent unblocks it for you.  

Conrad: A year ago, you inspired me with a Zapier workflow to analyze documents, but when I tried vibe coding last summer, I got stuck and frustrated. Agentic AI felt like a buzzword. But earlier this year, Casey on my team set me up with OpenClaw, and it blew my mind. I used to struggle with programming languages in college trying to map code to ones and zeros, but now I just paste an error screen into the agent, and it fixes it. In fact, I have Claude Code open right now. If you asked me to build an enterprise-ready Contract Lifecycle Management (CLM) system with AI reading metadata and running analysis across all documents, I could have that up and running by Monday.  

Zach: Absolutely, because the underlying technology is so much smarter now. To build a contract tool today, all you really need to understand from an engineering standpoint is basic concepts—like the fact that a system has a database with fields. AI carries you through the rest. However, the key lesson I have learned is that you must spend significant time up front on requirements and design. If you give AI a generic prompt to build a contract database, it will over-engineer the whole solution and add things you do not need yet, like native DocuSign or Adobe Sign integrations. You have to chat with your AI co-pilot, define a narrow Phase 1 slice, and get clear on your exact constraints before coding.  

Conrad: That makes total sense. When I prompt Claude Code using frontier models like Fable or Opus, it asks me discovery questions up front about scope, users, and repository architecture. This completely shifts the "build vs. buy" conversation for procurement leaders. We are at a pivot point where instead of dropping $50,000 to $150,000 on an off-the-shelf CLM system, a procurement team with basic skills can build and own that automation themselves.  

Zach: Exactly, but you have to evaluate the cost of failure. I would never advise someone to vibe code their own core ERP system because the regulatory and financial reporting stakes are way too high. But for core, upstream procurement activities—like intake requests, RFP generation, contract drafting, and project tracking—procurement professionals are in the best position to build custom tools because that is their bread and butter. It all comes down to user experience, judgment, and context.  

Conrad: Plus, look at how work actually gets done in the real world. People build fancy redlining features inside CLM portals, but legal teams and suppliers always download documents into Word because Word is the killer app for redlining. The same thing happens with RFPs—people end up downloading data into Excel because portal UIs do not allow fast matching. AI portals and Model Context Protocols (MCPs) will bridge that gap by connecting directly to our data libraries without forcing us into clunky web screens.  

Zach: That is why design and product decisions matter so much when building with agents. When I built version one of my autonomous procurement agent, I suffered through so many failures and rebuilds that the application became unorganized. When I started version two from a blank slate, I gave the agent access to all my notes from version one. Version two ended up thinner, faster, cheaper, and way more reliable because I designed it from the bottom up with clear architecture. AI cannot magically read your mind—you have to go through the creative process to learn what you actually need.  

Conrad: The amount of learning you get from taking on one of these projects is insane. It is like getting a master's degree in systems over a few months. And as far as job security goes, an AI-native kid fresh out of school cannot just step in and automate your function because they do not know your specific procurement process. You know your process. Learning to build alongside agents is what enables you to manage them in the future.  

Zach: 100%! I am super hopeful for the future of work. Humans still need to bring imagination, process knowledge, taste, and final decision-making to the table. Once you see the art of the possible and gain confidence with these tools, you will never go back.  

Conrad: I agree completely. Life and business are bigger than drudgery, and automating non-value-added admin gives us time back for what really matters. If anyone listening feels stuck or wants to see the art of the possible, reach out to us on LinkedIn. We run hands-on AI workshops and love helping people transform their workflows. Thanks for joining us today, Zach!

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