The AI Infrastructure Trap Associations Need to Avoid

A few weeks ago, Sherwin Wu, the lead engineer behind OpenAI's Codex, said something on Lenny Rachitsky's podcast that should make every association technology leader pause.

"The models will eat your scaffolding for breakfast."

He was talking about all the infrastructure that organizations have built around AI over the past few years. Vector stores. Agent frameworks. File-based context management. The tooling that made early AI actually work.

His point: much of that scaffolding is already obsolete. And whatever scaffolding we're building today will likely be obsolete soon too. The models keep getting better, and as they do, they absorb capabilities that used to require external architecture.

This raises an uncomfortable question for associations thinking about AI investment. If the infrastructure keeps changing, what exactly should you be building?

The Two Traps

Associations thinking about AI infrastructure typically fall into one of two traps.

Trap 1: The off-the-shelf AI wrapper. These are the products that layer a nice interface on top of existing AI models. They're easy to buy, quick to implement, and feel like progress. The problem is they're static. The vendor built their architecture in 2024 or 2025, and that's what you get. When the models evolve and the scaffolding becomes obsolete, you're stuck with a product that was built for yesterday's capabilities.

Trap 2: Building everything internally. Your IT team takes on the job of tracking model developments, evaluating what's changed, swapping out components, and adjusting architecture. Continuously. Forever. For organizations where technology is a support function rather than the core mission, that's an enormous commitment. Most associations don't have the resources to keep up.

The way out is a third option: long-term infrastructure partners who manage the evolution for you.

What the Right Partner Looks Like

Not all AI partnerships are created equal. Here's what associations should be looking for.

You own your data and your data destiny. This matters more than most organizations realize. If your AI infrastructure locks you into a proprietary system, you've traded one dependency for another. Open-source foundations mean you retain control. You can switch providers. You can build on top of what exists. Your data doesn't become someone else's asset.

They actually understand associations. AI infrastructure built for generic enterprise use cases will miss the nuances of member data, credentialing workflows, event management, and the dozen other things that make associations different. Domain expertise isn't a nice-to-have. It determines whether the solution actually fits your operations.

Your data stays protected. This is especially critical for associations handling member information. The right architecture shares metadata with AI models so they understand the shape of your data and can run analysis. But the actual data never leaves your environment. Members' personal information doesn't get sent to external models for processing.

They're continuously updating the plumbing. This is the Sherwin Wu point. Vector stores may matter less next year. Agent frameworks are already shifting. The models themselves keep changing. A good partner is watching all of this and adjusting the architecture so you don't have to. They're switching in new models, updating how artifacts get generated, improving transparency and explainability. That's their job, not yours.

What This Looks Like in Practice

The foundation of effective AI transformation is a unified data layer that connects your AMS, LMS, event systems, and other member touchpoints. This creates a single source of truth for member information, enabling AI agents and analytics tools to work from complete, accurate data. Without this foundation, AI implementations remain siloed and limited in impact.

But the deeper value is what happens over time. When model capabilities change, the infrastructure gets updated. When new approaches emerge for things like search or context management, they get evaluated and integrated. When the scaffolding needs to evolve, it evolves.

Associations using this approach aren't betting on a static product. They're accessing a team that's continuously rebuilding for where the models are going.

This is the approach we took when building Member Junction at Blue Cypress. But the principles apply regardless of which partner you choose.

The Bitter Lesson, Applied

There's a concept in AI research called "the bitter lesson." The short version: simpler approaches that scale with computing power consistently beat complicated hand-engineered solutions. Researchers keep learning this the hard way.

Sherwin Wu pointed out there's a version of this lesson for building with AI too. Organizations keep architecting elaborate systems around the models, and the models keep absorbing that complexity.

For associations, the implication is this: don't get locked into static SaaS wrappers that won't evolve. And don't try to build and maintain everything yourself. Look for long-term infrastructure solutions that manage the continuous evolution for you.

Someone who builds for where the models are going, not where they are today. Someone who treats infrastructure as a living system, not a one-time purchase.

That's how you avoid the trap.

This is part of our "Strategic Window" series on AI investment for associations. Previously: "The 10% Framework: A New Way to Think About Association Reserves," "The Secret Cyborg Problem: Why Your Association's AI Gains Are Stuck at 10-20%," and "What I Heard at CESSE: Associations Are Rethinking What They Actually Are."

For the complete framework and supporting research, download our white paper: "The Window Is Open: A Framework for Deploying Association Reserves into AI Transformation."

About the Author

Johanna Kasper Snider is the CEO of Blue Cypress. Blue Cypress is building the AI ecosystem that helps associations transform how they serve members—with a bold goal of making associations as powerful as Fortune 500 companies by 2030. The Blue Cypress family includes AI products like Betty, Izzy, rasa.io, Skip and SoundPost; services companies Cimatri, Elastik Teams, and Tasio; Sidecar for AI education; and Blue Cypress Consulting for strategic transformation. BC Labs incubates and launches new AI solutions for associations. Johanna has over a decade of experience in SaaS and professional services, holds a BA and MBA from Tulane University, and is a contributor to "Ascend: Unlocking the Power of AI for Associations."