Introducing: NotebookLM, Operator, Deep Research and GPT-4.5

NotebookLM

Google’s NotebookLM is a research and note-taking assistant that uses their latest Gemini LLM, with its enormous 2 million token - 1,750,000 word - context window (think ‘memory’) to synthesise documents, generate summaries, and create podcast-style “Audio Overviews”. You can add hundreds of documents, videos, audio files, web site links or Google Docs/Slides and then interrogate them intensively and specifically. The system is ‘grounded’ in the sources you upload so has minimal hallucinations and provides citations.

Effectively, this gives everyone access to expert-level analysis—at

the scale of an entire project.

Which has consequences: Imagine reviewing lengthy lease agreements or environmental reports as audio summaries while commuting or multitasking. Or using it as a collaborative research assistant for geographically dispersed teams, synthesising insights in real time. It could even redefine how companies share knowledge—leasing teams, for instance, might receive AI-generated podcast updates on market trends, technical due diligence, and portfolio changes, making internal communication more dynamic and efficient.

NotebookLM will streamline the analysis of market reports, technical due diligence documents, even property portfolios.

Operator

OpenAI’s ‘Operator’ is an AI ‘agent’ designed to interact with websites and perform goal-based tasks, such as booking appointments or scraping data. Essentially it can operate your PC just as you can. It is programmable, meaning you can assign it tasks, and it will determine (either through explicit instructions or independently) the best way to complete them. So it could autonomously handle workflows that can proactively manage operations, anticipate issues, and optimise performance. Without constant human intervention.

So it might operate your building’s systems, go on deal sourcing hunts (24/7), or dynamically respond to Helpdesk requests.

Or it might be programmed (via natural language not code) to automate routine tasks such as data extraction from public property records or scheduling property inspections. Eventually it might integrate with CRE platforms and automate repetitive workflows (e.g., updating property listings or extracting market data from multiple online sources).

Think of it as a ‘digital employee’. But unlike traditional RPA, which follows rigid scripts, Operator is a digital employee that learns, adapts, and executes autonomously.

In this world, the human no long does daily tasks, but rather spends their time designing and overseeing these autonomous systems.

System design, high-value strategic thinking and client engagement become the killer skills.

Deep Research

OpenAI’s Deep Research (Google and Perplexity have similar products) is an AI research agent that autonomously browses, synthesises, and produces cited reports. Unlike what has previously been possible, these ‘reasoning’ systems take their time, anything from 5 to 30 minutes, to work their way up and down, back and forward,through a problem. They do pretty much what a human researcher would do, just much faster.

What they also do is open up the world of hyper-local market intelligence and niche market expertise. These systems can be instructed to go exceptionally deep—making them ideal for due diligence, competitor research, and trend analysis.

Are they perfect? No. As of today they can make mistakes and theirwork needs to be checked. As one would with an intern or junior researcher. But increasingly you will be able to give them secure access to all your internal knowledge and intelligence. And by doing so provide them with known facts and data, greatly reducing the likelihood of inaccurate or misleading information.

GPT-4.5

GPT-4.5 is the latest, and last, non reasoning model produced by OpenAI. It is a model 10 times the size of GPT-4, with improved writing capabilities, greater world knowledge, and a more refined conversational personality. Whilst not seeming to be a major  breakthrough on the AI Evals Leader Board, it is said to be the first model that really ‘feels’ like one is interacting with another human.

In practice it is going to be used as the ‘Teacher Model’ from which many, domain specific ‘Student Models’ will be ‘distilled’. Distillation creates smaller, faster models that retain most of the accuracy and capabilities of the original but are fine-tuned for specific tasks.

Within CRE we are likely to see it used to enhance communication and contracts. Its refined language abilities could revolutionise the drafting of contracts, marketing materials, and client communications, providing clear, persuasive, and data-backed narratives.

We know GPT-5 is coming and that will add ‘reasoning’ capabilities to what we are seeing in GPT-4.5. And from recent model development history, we know that the power of reasoning models is partly a function of the strength of the ‘pre-trained’ models they are built upon.

So expect to see more of us treating these LLMs as creative partners in CRE, augmenting human intuition and blurring the lines between human and AI creativity. We are going to redesign creative work in CRE, shifting it from individual creativity to collaborative AI-human creativity.

For CRE, GPT4.5 offers practical gains in everyday tasks—from generating property descriptions to automating client correspondence—thereby enhancing productivity while setting the stage for more transformative future models.

What These Tools Tell Us About the Future of AI

Domain Specialisation

First off there is domain specialisation. AI is moving toward tools that are not one-size-fits-all but are increasingly tailored to specific industries, blending general language capabilities with domain-specific data. Each of these tools can be targeted and fine-tuned for our very specific data, information and knowledge needs.

Hybrid Human–Machine Workflows

Secondly hybrid human–machine workflows will become the norm because, despite their impressive automation capabilities, these tools still require human oversight for quality control and ethical decision-making. The future lies in creating seamless collaborations where AI handles data-heavy tasks and humans provide judgment and strategic insight. With the caveat that this human input has to be of the highest quality. And that this does not apply to all workflows (though they will be the ones humans are actually interested in).

Incremental Evolution vs. Radical Disruption

Thirdly, we should expect more incremental evolution than radical disruption. Taking GPT4.5 as an example, its benefits will take time to evolve and emerge. But they will come bit by bit, so whilst a big bang should not be expected these small incremental improvements will have a cumulative impact and whilst a tad slower, one must still expect and anticipate significant transformation.

This is akin to the way Dave Brailsford adopted the strategy of the "Aggregation of marginal gains" to improve the Sky cycling team's performance. By improving hundreds of tiny factors by just 1%, his team achieved remarkable cumulative gains—winning the Tour de France two years ahead of schedule.

Operational Efficiency and Risk Management

Operator in particular will be a boon for operational efficiency and risk management but it is going to take some time for it to be ready for mainstream use. And then the design of the systems we apply it to, and where we place the ‘human in the loop’ is going to be critical. That said, once these are finalised we will have an AI that won’t stop doing whatever we want it to, and the productivity gains could be extraordinary.

Redesigning Workflows

How we reimagine work processes to leverage AI's efficiency is going to represent a super-skill amongst humans. It has been shown that adding an AI into a human workflow can actually make it work less well, can make it perform as well as ‘the best human’, or can make 2+2=5. Designing for the latter will take special, and very valuable, talents.

Curating Intelligent Outcomes

When humans don’t need to actually undertake tasks their input is going to be mostly around creating, and curating. Designing prompts, setting the parameters, and fine-tuning the outcomes to align with strategic business goals. Redesigning workflows and curating intelligent outcomes will become a new and large job category.

Reinventing the Office

And finally, while AI’s impact spans all asset types, the office will feel it most acutely. Routine tasks will fade, replaced by ideation, strategy, and creative collaboration. Offices must transform from places of execution into catalysts of human potential.

Conclusion: Embracing the Unseen Revolution

The four tools we’ve discussed are harbingers of the democratisation of expertise, autonomous workflows, niche market mastery, and AI as a creative partner. And this has broad implication for CRE: we’re moving beyond automation, and are shifting towards a more data-driven, proactive, specialised, and creatively augmented future.

Automation will be everywhere but it will no longer be the end point, the goal. Instead we’ll be building products and services of a higher order. We’ll be looking to build a better built environment, not just optimise and manage what we currently have. These and other tools are going to allow us to raise the bar of what is possible, and what we aspire to.

As models improve we’ll increasingly be able to focus on human-centric goals such as well-being, community, sustainability, and economic opportunity. To be sure, we will have to as much of what we humans do in CRE today will be co-opted ‘by the machines’, but that is a feature not a bug. These tools are incredible enablers, and they will open the door to building a built environment we can barely imagine today.

What to do?

One can guess, probably pretty accurately, that certain areas are going to be impacted the most, and soonest: property management, investment, brokerage, development, and corporate real estate. So here are seven things to be doing to prepare:

TOP PRIORITY

1. Experiment with Tools: Use NotebookLM, Deep Research, Operator, and GPT-4.5 to gain direct experience.
   ‍Why: Hands-on experience is paramount to understanding the capabilities (and limitations) of these tools. Before you can effectively have internal conversations, identify opportunities, or plan strategically, people need to see what these tools can do.This builds conviction and sparks ideas.
2. Start Internal Conversations: Discuss the implications of AI within teams and across your company.
   ‍Why: Before anything else, creating internal awareness and alignment is crucial. Without a shared understanding of the opportunities and challenges, other efforts will be less effective. This helps build a culture of experimentation and innovation. It's also the least resource-intensive way to start.
3. Invest in Upskilling: Prioritise training programs for teams to leverage AI tools effectively, and boost AI literacy.
   ‍Why: AI tools are only as good as the people who use them. Upskilling ensures your team can effectively leverage these technologies, maximising their potential. This might involve internal workshops, online courses, or bringing in external experts.
4. Identify Automation Opportunities: Pinpoint workflows ripe for AI-driven automation.
   ‍Why: By identifying specific processes that could benefit from automation, you can focus your resources on implementing AI solutions that deliver tangible results. This can also help build a business case for further AI investments.

HIGH PRIORITY

1. Forge Strategic Partnerships: Collaborate with tech experts and AI solution providers - get to know and understand the ecosystem that is developing.
   ‍Why: Building relationships with tech experts and AI solution providers provides access to specialised knowledge, resources, and support. These partnerships can help you navigate the complex AI landscape and find the right solutions for your business.

MEDIUM PRIORITY

1. Assess Tech Infrastructure: Evaluate current systems for AI readiness.
   ‍Why: Understanding your current tech capabilities is essential for determining which AI solutions can be implemented and what upgrades may be necessary. This is more of a technical step that can be addressed once your automation opportunities are clearer.
2. Plan for Change: Develop a roadmap for integrating AI that maintains human oversight and creative input.
   ‍Why: Creating a roadmap ensures that AI integration is well-planned and aligned with your overall business goals. This helps avoid the pitfalls of implementing AI in a piecemeal or uncoordinated fashion.

But philosophically, really engage with the notion of "Building a Better Built Environment": Reflect on what this means in an AI-powered context and how AI can help achieve our goals.

Be as aspirational as you can be. And then some more.

Everything IS possible.