Claude Design for CRE Decks (+ an AI Governance Framework)

1. Anthropic Drops Claude Opus 4.7 — And It's a Beast Anthropic released Claude Opus 4.7 last Wednesday, reclaiming the top spot among generally available AI models.

2. Anthropic Launches Claude Design — From Idea to Visual in Minutes The day after Opus 4.7, Anthropic dropped Claude Design — a new tool that lets you go from a text description to polished slides, prototypes, one-pagers, and presentations without touching a design tool. During onboarding, it reads your existing brand assets and builds a design system using your colors, typography, and components. Outputs export as PDF, PPTX, or directly to Canva.

3. HBR: The Hidden Demand for AI Inside Your Company Harvard Business Review's latest piece explores a growing problem: employee demand for AI tools is outstripping what most companies are willing to provide. The result? Shadow AI — people signing up for their own ChatGPT accounts, feeding proprietary deal data into consumer tools, and building unsanctioned workflows with zero security oversight.

4. PwC: 74% of AI's Economic Gains Go to Just 20% of Companies The most important data point of the week comes from PwC's 2026 AI Performance Study: a small group of companies is pulling dramatically ahead in the AI race, generating 7.2x more AI-driven revenue and efficiency gains than the average competitor. The differentiator isn't spending more on AI — it's redesigning workflows around AI rather than layering tools on top of existing processes. Sound familiar?

Claude dropped Claude Design last week on April 17, a dedicated design interface that can design decks, website prototypes and more from a single prompt. I've been down a rabbit hole putting it to the test on CRE decks. While AI-powered deck generation isn’t a new phenomenon, Claude Design doesn’t disappoint. The editing features combined with the accessibility of this kind of design horse-power in an AI platform that many of us use every day could be a massive unlock for CRE professionals. Polished decks, ready for export to Powerpoint or a handoff to Canva for finishing touches in 10-15 minutes (plus the time you need to review.)

All that said, it’s still early and the platform isn’t perfect. Here’s some things you should know before you get started:

This is a powerful design tool, and the more you understand the ins and outs, the better marketing materials it will help you produce.

I created a full guide on Claude Design for CRE decks, with everything from my testing & research along with the recommended prompt framework. Feel free to check it out in CRE AI Studio.

Claude Design is one door into the Claude lineup. If you're deciding where to start, read Claude Chat vs. Cowork vs. Claude Code, then How to Set Up Claude Cowork.

- Topher

By: Nadine Ezzie

In late 2025, a researcher named Richard Weiss was poking around Claude 4.5 Opus on release day, trying to extract its system prompt. The model kept referencing a document called "soul_overview" that wasn't in its system message. He figured the model was making it up. Then he regenerated the response ten times. The text came back nearly identical every time.¹ 

Turns out the soul was real. What Weiss had surfaced was an internal document that Anthropic's personality alignment lead, philosopher Amanda Askell, had been quietly writing for years. On January 22, 2026, Anthropic published the full 80-page successor, Claude's constitution, under a Creative Commons public domain license.² 

Then, earlier this month, Anthropic released Claude Mythos Preview, a model the company itself described as its best-aligned model to date and the one with the greatest alignment-related risk.³ Mythos can autonomously find zero-day vulnerabilities, (which is security industry shorthand for software flaws nobody knows about yet, not even the people who built the software. Mythos found thousands, including a 27-year-old flaw in OpenBSD, an operating system whose entire reputation is built on being secure). Anthropic paired the release with Project Glasswing, a $100 million coordinated defense effort, because releasing the model without one was considered irresponsible.⁴ 

In four months, Anthropic went from publishing the governance document to releasing the model the document was written to govern. In the same four months, what did the CRE industry publish about AI governance?

The question matters because of what CRE firms actually touch. In a single organization, we hold tenant data, employee records, investor financials, asset data, and counterparty deal terms. Five different kinds of sensitive information, one firm. And unlike financial services or healthcare, CRE has no sector-specific AI regulator breathing down its neck. 

At least not yet.

In a recent conversation with a long-standing voice in the AI community, someone who has been at this since 1980 and has advised presidents, the World Economic Forum, and two popes, framed true AI governance as three tiers: (1) the laws, (2) the firm's own rules, and (3) the individual's own judgment. You need all three. A firm without its own rules is relying on either regulation (slow) or individual conscience (uneven). A firm with rules but no individual buy-in has produced a PDF, not governance.

At tier one, the law is catching up. The EU AI Act goes into full enforcement in August 2026, with penalties up to EUR 35 million or 7% of global revenue.² The enterprise AI governance market is projected to grow from $2.55 billion this year to over $11 billion by 2036.⁵ Entire companies are being built to sell what CRE has, until now, treated as an afterthought. 

Meanwhile, MIT's April 2026 mapping of the global AI governance landscape found that Real Estate and Rental and Leasing is one of the least covered sectors in the conversation.⁶ The world is talking about AI governance. It is not talking about us.

At tier two, the firm-level code, Anthropic's constitution is instructive. Not because CRE firms should publish 80 pages (please don't!). But because of what a serious code actually does. It explains why before what. It establishes a priority order, so that when values conflict, the model knows what wins: safety first, ethics second, compliance third, helpfulness fourth.⁷ 

In case I wasn't clear, it gives the model standing to refuse its author. Claude is explicitly told it can act as a conscientious objector and refuse to help Anthropic if a request seems inconsistent with its values.⁸ Now ask yourself, does your firm's AI policy give a junior analyst standing to refuse a senior director’s request? If the answer is no, you don't have a code. You have vibes.

Here’s where CRE needs to look in the mirror. Everywhere I look nowadays, everyone is talking "human-centered" this and “human-centered” that. Every tenant experience platform, every workplace strategy deck, every ESG panel description invokes human centricity as a core principle. Fine. The vocabulary is polished.

But human-centered AI governance is the same principle applied one floor up. If you take human centricity seriously in the lobby, you can’t abandon it at the technology layer. Governance is how you operationalize human centricity when the counterparty isn't a tenant, it is a model. An industry that put "human-centered" on every tenant experience deck now has to prove it wasn't just a good panel title.

At tier three, the individual code of ethics, is where most of CRE quietly lives today. A quick tour of what passes for governance in most shops: the Slack channel where someone posted a ChatGPT link eight months ago. The IT memo from March that nobody opened. The vendor questionnaire that asks whether the tool is "SOC 2 compliant," as if that settles anything about what your people type into it. The training deck that ends with "use good judgment," which is what you write when you don't have a framework. The quarterly AI Committee meeting, which runs four times slower than the technology changes. 

I am describing, with a straight face, well-run institutional firms. Not laggards. Firms that would describe themselves as early adopters.

The numbers tell us how thin this is stretched. Grant Thornton's 2026 survey found that among organizations still in the pilot phase, only 7% are confident they could pass an independent AI governance audit within 90 days.⁹

So what do you actually do? Tier one is out of your hands. But tier two is sitting on your desk waiting to be written. Draft a one-page firm code of ethics on AI. State the reasoning, establish the priority order, name the person with standing to refuse. Revisit it quarterly, not annually, because the technology cycle doesn't wait for your annual policy review. At tier three, model it yourself. Every time a leader in your firm pauses before pasting something into a consumer AI account, or cites the firm's code to decline a shortcut, the individual layer gets built. That is the layer no PDF can produce.

Claude has a soul document because the people who built it decided character couldn't be an afterthought. They hired a philosopher, wrote 80 pages, and published it. Then they shipped Mythos. They built the governance before they built the capability that required it. The CRE industry is about to deploy these same tools into every lease, every underwriting model, every investor letter. The tools are here. The governance is not.

The question isn't whether we'll catch up. It's whether we'll write our governance on purpose, or inherit it by accident. One of those is a choice. The other is what happens while you're deciding.