The Three Layers Decoded

On Monday I described the three-layer architecture that organizations are using to solve the AI governance problem. Today let’s get specific about how each layer actually works, and why the combination creates something that transforms how you use AI.

Layer 1: The Interface — Making AI Trustworthy

Start with what your staff actually interact with.

Subscribe now

Read more

By now, if you’ve been reading this series, you understand the problem.

Your organization needs AI. Your staff want to use it. But the way AI is currently available—scattered across public tools, outside your control, generating no audit trails—doesn’t align with institutional responsibility.

You’ve heard why that m…

Read more

On Monday I described a problem most institutional leaders are beginning to face: regulators, auditors, boards are asking how you govern AI. And “we have a policy” isn’t a satisfying answer anymore.

Let’s talk about the difference between the two ways institutions actually try to solve this problem. And why one of them works and one of them fails.

The compliance-as-theater approach

This is what most organizations are doing right now.

You recognize that AI governance matters. So you write a policy. The policy says: staff should be thoughtful about what data they share with AI tools. Don’t upload confidential information. Get approval for certain uses.

Then you do training. You tell staff about the policy. You explain why it matters. You hope they remember it when they’re solving actual work problems.

Then you spot-check. You ask people what AI tools they’re using. You assume they’re following the policy.

And you write a report. You tell your board, your auditors, your regulators: we have an AI governance policy. We train staff. We monitor usage.

This is compliance-as-theater. It looks like governance. It sounds like governance. It allows you to say you’re governing AI; but it’s Kabuki.

But it doesn’t actually govern anything. It relies entirely on voluntary compliance. It provides no visibility into what’s actually happening. It has no enforcement mechanism. It can’t prevent violations. It can only catch them after they’ve happened.

And when your auditor asks to see evidence of governance, what do you show them? The policy document. The training records. Your good intentions.

But what you can’t show them is proof that the policy is actually working. Evidence that you’re preventing the behaviors you’re concerned about. Visibility into what’s actually happening.

Subscribe now

Read more

Regulations around AI are not coming in some distant future. They’re here. They’re accelerating. And they’re asking questions your organization probably can’t answer yet.

If you run a healthcare organization, you already know this. The FDA has guidance on AI-based softwa…

Read more

Here’s something most institutional leaders don’t fully appreciate about adoption: people will use tools they trust with sensitive information. They won’t use tools they don’t trust, no matter how powerful, convenient, or strongly incentivized.

This is not a technology problem. It’s a trust problem. A…

Read more

And most of them interpret that hesitation as a problem to solve. Fear of technology. Lack of sophistication. Resistance to change.

But here’s what I’ve learned by talking…

Read more

A Practical Toolkit for Governance-First AI Adoption

Read more

Read more

On Monday I described a problem most institutional leaders are quietly worried about: your organizational data is leaving your control through AI tools, and you don’t have good visibility into it.

Let’s get specific about what data we’re actually talking about, and why it matters that you can’t see it or govern it.

Understanding operational data

Start with the simplest category: operational data. The information that makes your daily work possible.

This includes things like internal processes, reports, schedules, communications, routine decisions. It’s the stuff that describes how your organization actually works day-to-day. It’s not typically confidential. A competitor wouldn’t gain much advantage from knowing your standard processes.

But here’s what happens when this leaves your control: it creates a record. Your processes are now part of external systems. External AI models are trained on how you work. Your organizational practices become part of training data that could surface anywhere.

This matters because it’s a form of visibility you didn’t give. Your staff thought they were solving a work problem. They didn’t realize they were creating a permanent record of how your institution operates.

More importantly, once that data is external, you can’t govern how it’s used. You can’t control what happens with it. You can’t ensure it’s treated consistently with your values or your practices.

This is why even seemingly low-risk operational data matters: it’s the first step down a path where you lose visibility and control.

Understanding strategic data

Move to the second category: strategic data. The information that actually gives your organization competitive advantage.

For some organizations this is research. For others it’s market analysis, donor relationships, partnership strategy, program design, institutional knowledge. It’s the information that, if your competitors or your market saw it, would meaningfully change their decisions or capabilities.

This is the data most organizations are protective of. You have systems in place to keep it safe. Access controls. Limited distribution. Careful stewardship.

But strategic data is also exactly what your staff reaches for when they need to solve problems with AI. They have a strategic question, they ask an AI system. They need to analyze competitive positioning, they upload relevant information. They’re trying to get better analysis, faster thinking, more perspective.

And in doing so, they send strategic information outside your boundaries.

Subscribe now

Read more

It happens in almost every organization now. Someone on your team solves a problem using AI.

They have a question about organizational policy. They paste it into ChatGPT and get a helpful answer. They have a document that needs analysis. They upload it to an AI tool and get structured insights. They’re …

Read more

Read more

So you’re probably doing what most institutional leaders do: trying to add AI tools to existing work. A chatbot here. A document analyzer there. Maybe a research assistant for your team.

But her…

Read more

The three shifts that quietly changed AI this year

A field guide for leaders trying to understand where AI is actually heading.

Part 1 of an 8-part series from Parallax Partners on the last year in AI.

Read more

And even fewer are ready to build a memory system.

Most teams are still experimenting with generic tools. They’re automating small tasks, generating content, maybe integrating a chatbot. And that’s fine—for now.

But the teams we work with? They’re further along. They’ve passed the novelty stage. They…

Read more

But here’s the real question most orgs are asking beneath the buzz:

“Is this actually helping us do our work better—or just giving us new toys to play with?”

At Parallax, we’re not i…

Read more

“We’ve tried a few tools; they’re cool, but the actual work hasn’t gotten any easier.”

That frustration is real. And it’s not just about maturity of the tech. It reveals a deeper blind spot baked into most AI implementation strategie…

Read more

I’m the guide: steady, practical, and focused on craft. I won’t offer miracles. I’ll offer a posture, a few simple instruments, and habits that make AI useful in everyday work. This series — How to Work With AI — is about turning capability into routine. Each issue will give you one thing t…

Read more

We’ve entered a new era—an age Thomas Friedman calls the Polycene, where “many, not one” defines how the world operates.

• Many signals.

• Many actors.

• Many systems.

• Many crises—but also many opportunities.

For AI designers, technologists, and systems thinkers, this …

Read more

No one opened their window shade.

Not one person - well, me and a lady in 27A did.

As we rose over one of the great human-made landscapes of the world — Manhattan unfolding in perfect mornin…

Read more

Spend even a few minutes online and you’ll see the complaint everywhere:

“AI is flooding the internet with slop.”

Bland articles that feel interchangeable. Reports that sound polished but empty. Strategy decks that read like they were generated from the same tired template.

It’s tempting to point at the model and say, “There’s the problem.” But that’s only…

Read more