THE BLOG
Agents, Agents, Agents
The low hanging fruit of generative AI. Waiting to be picked.
AI 'Agents' come in many forms, and OpenAI’s ChatGPT contains three that you can learn and leverage in no time at all - Custom GPTs, Projects and ‘Agent' mode.
So What are AI ‘Agents’
Let’s start with what AI ‘Agents’ are. They are simply intelligent systems that can be designed at different levels of complexity — from lightweight assistants to fully autonomous problem-solvers. They come in many forms, each suited to different purposes, whether guiding a single workflow, managing a project, or operating as a flexible digital teammate.
To be clear this is a sliding scale, all the way up to what purists mean when they talk about ‘Agentic AI’:
‘A true AI agent is an autonomous system that persistently pursues goals through iterative environmental sensing, decision-making, and action-taking, whilst adapting its strategies based on feedback and changing conditions.’
These exist today, in limited numbers, but mostly the AI Agents we work with today are not fully autonomous and are very much designed with a ‘human in the loop’.
Think of these three as rungs on a ladder: Custom GPTs for repeatable text, Projects for structured workflows, and Agents for proactive automation.
ChatGPT Custom GPTs
What are they?
They’re bespoke versions of ChatGPT tailored for a specific role, task, or style of work. They are configured with your own instructions, tone of voice, and reference material so outputs are consistent and repeatable. And you can embed templates, checklists, or frameworks relevant to your domain (e.g., investment memos, ESG plans, board packs) into them.
When to use them?
You use them for repeatable, text-driven tasks such as memos, reports, checklists or templates. They work best when you want a GPT that consistently “thinks like your team” without re-explaining context each time, and when you want to share a standardised tool with colleagues, so everyone produces outputs in the same style and structure.
In a nutshell, you use a Custom GPT when you want repeatable outputs, in a consistent style, that you can share with others.
Use Cases
Here are four use cases:
Investment Committee Support: draft polished IC memos in your firm’s preferred format.
Fund Reporting: produce NAV updates and ESG reports in a consistent structure.
Recruitment: creates job specs, interview packs, and scoring templates with the right tone.
Heads of Terms Negotiation: draft clauses, flags risks, and ensure standardised outputs.
In the #GenerativeAIforRealEstatePeople course we have 20+ ‘TDH GPTs’ that do everything from provide career advice, act as sustainability consultants, act as IC Committee Advisors, and help you negotiate Leases.
Outside real estate progressive companies use Custom GPTs throughout their business. Vaccine developer Moderna has over 3,000 among a workforce of 5,600. Nearly every workflow could benefit from a Custom GPT.
One of the TDH Custom GPTs will even help you work out where best to use them in your own business.
ChatGPT Projects
What are they?
They are workspaces inside ChatGPT designed for multi-step, data-driven, or ongoing workflows. They let you store files, instructions, and conversation history so you can return and build on work over time. And they support advanced data analysis (spreadsheets, models, scenario runs) alongside natural language prompting.
When to use them?
They are best used for complex workflows that need structured inputs and iterative runs (e.g., portfolio stress testing, budgeting, capex prioritisation), and when you need to upload and reuse data or documents (e.g., lease schedules, ESG data, financial models). In addition when you want to track progress across sessions, and not just deal with one-off answers.
A major difference though is that Projects are not shareable - they are personal workspaces, unlike Custom GPTs which can be distributed across a team.
Use Cases
Here are four use cases:
Portfolio Stress Testing: run vacancy and interest rate scenarios with uploaded data, saving results for comparison over time.
Capex Prioritisation: rank ESG retrofits and fit-outs using criteria and data files, updating iteratively as assumptions change.
Budgeting & Forecasting: manage Opex/Capex scenarios, store models, and track forecasts across sessions.
Planned Preventive Maintenance (PPM): schedule, prioritise, and update maintenance tasks across sessions.
Within the course we have a Project that lets you see a building through the eyes of its occupiers. It surfaces the pain points that drive churn, disputes, or reputational risk, and generates practical AI/tech interventions to fix those problems.
You can push Projects pretty hard; for workflows that you repeat, are quite complex but follow certain patterns, and require updated datasets they can be incredibly useful.
ChatGPT Agent
What are they?
They are AI assistants that can work continuously in the background, not just when prompted. They can monitor systems, fetch data, update trackers, and send alerts across different tools, and are designed for proactive workflows that go beyond “ask and answer.”
When to use them?
They are best used when a workflow needs ongoing monitoring (e.g., lease events, arrears, compliance deadlines). Also when tasks require cross-tool coordination (e.g., pulling from CRM, data room, spreadsheets, and messaging platforms). Overall, when you want the AI to act without being asked each time, and are best for time-sensitive or repetitive processes where a missed step carries risk.
Note: Agents are still evolving - they often need careful setup and integration with your existing systems. They’re powerful, but not always “plug and play.”
Use Cases
Here are four use cases:
Lease Event Management: track renewals, break clauses, and re-gears, fetching market data and prompting timely action.
Arrears Management: monitor payments, flag arrears, and draft notices automatically.
Compliance (Safety, etc.): track fire, asbestos, and statutory deadlines, sending reminders and updating logs across systems.
Deal Pipeline Tracking: continuously monitor NDAs, bids, and due diligence statuses, reducing manual oversight.
Most of the above would take some setting up, and are perhaps more aspirational than practical today. Nevertheless, one should keep track of the capabilities of ‘Agent’ mode because it is developing fast. In a year it’ll be unrecognisable. Within a year, expect agents that handle entire leasing workflows — scheduling viewings, updating deal trackers, and flagging risks automatically — with minimal human input.
That said, in Agent mode today you can send your ‘Agent’ off to visit websites, read news, do calculations and report back. You just have to try ideas out and see how they get on. It’ll be good practice for when you have dozens of these virtual helpers working for you 24/7.
Conclusion
These three ‘Agents’ are not widely used, but they should be. And no doubt will be over time. But for now, as I repeatedly stress, if you lean into these advanced uses of Generative AI you’ll be doing yourself an enormous favour. These are low hanging superpowers - I’d be amazed if you weren’t amazed at what they are capable of if you give them a serious go.
OVER TO YOU
What workflows resonate with you? Where do you think you could use a Custom GPT, a Project or an Agent? There are probably dozens of use cases - just pick a few, and dive in!
This week, try building one Custom GPT that mirrors your team’s tone. Use it three times. If it doesn’t save you time, email me and tell me why.
Changing Assumptions
"When the facts change, I change my mind” - John Maynard Keynes
Real estate is heading towards an operating model where job functions run on prompts—and agents do the rest. The shift is coming faster than expected.
The Hypothesis
It is my belief that a (very) large percentage of workflows in real estate can be broken down into a series of tasks, and that these tasks can be completely, or nearly, automated by the application of ‘Prompt Packs’.
Prompts are Enough
Each ‘Prompt’ in the pack - they’d work sequentially through 3-7 steps - would contain the essence of that task. By which we mean they’d contain:
Inputs – the required data (rent rolls, EPCs, abstracts, comps).
Processing steps – filtering, benchmarking, compliance checks.
Outputs – tables, reports, approvals.
Examples – to make the flow transparent.
We looked at workflows across nine categories, such as Leasing & Occupier Management, Valuation & Investment and Asset & Portfolio Strategy. And realised that almost all the workflows could be fitted into the I/P/O/E framework above.
It became clear why Morgan Stanley (after analysing tasks performed by 162 real estate investment trust and commercial real estate firms, with a combined $92 billion of labor costs and 525,000 employees) recently wrote that ‘AI can automate 37% of tasks in real estate, representing $34 billion in operating efficiencies.’ **
Two Breakthroughs This Year
This still holds true from when we devised this at the beginning of the year, but since then, two breakthroughs have emerged to push the whole idea forward.
Automation of automation (Master Prompts)
First, it may be possible to ‘automate the automation’. Since GPT-5 came out in early August we’ve seen how much better it is at multi-step problem solving, instruction-following for complex and evolving tasks, and invoking additional ‘tools’ as and when required. And with these new capabilities we’ve found it is possible to develop ‘Master Prompts’ that allow you to enter a workflow and the models work out the entire input/process/output/example framework and build the series of prompts required.
In effect a prompt can create ‘Prompt Packs’.
Now this is at an early stage but the technology has developed to such an extent that a lot of shortcuts to the future are now available.
Practical Agents (Agentic AI)
The second breakthrough to emerge is that the long-forecasted world of ‘Agentic AI’ is arriving fast. Whilst still somewhat brittle, the idea of creating discrete software services that can be given instructions and that can then be left to autonomously work out how to complete them, is coming to pass. Anyone who has used ChatGPT Agent will have experienced the rather odd sensation of watching a virtual entity thinking and acting its way through a problem whilst boxed in a computer within a computer.
This is opening up a whole new world of opportunity and whether more robust forms of Agents take 3, 6, 9 or 18 months to arrive, they will definitely be arriving. And so one can plan for them.
From Prompt Packs to Agents
Which means that in the near future we are very likely to see large swathes of the real estate industry become industrialised. Prompt Packs will naturally morph into collections of Agents.
Just as lean manufacturing, via ‘The Toyota Production System’, codified shop-floor know-how into standard work instructions, Agentic AI will be doing the same for CRE.
Now, ’The Toyota Production System’ is one of the most consequential management innovations of the 20th century — arguably as influential to industrial organisation as double-entry bookkeeping was to finance.
The fundamental point is that the concept of codifying tacit knowledge is hardly new in management thinking. So real estate should not be surprised that it appears to be finally reaching our industry.
There are already domain specific AI companies offering a range of services dealing with high value, document heavy and repeatable real estate workflows. Where this ‘Prompt Pack/Agentic’ framework differs is that it applies to all the other workflows one has to deal with day-to-day that do not merit VC backed startups addressing. The implementation of ‘Prompt Packs/Agents’ will be led by all of us. Each creating our own swarm of virtual helpers to suit our particular needs.
As we’ve discussed before , the days of us being ‘Agent Bosses’ are near. We’ll create, monitor and curate these tools. We’ll industrialise ourselves!
So, between the high-end outsourced agent creators and our ‘build your own’ efforts, a huge amount of what we’ve been paid for as an industry in the past is about to be automated.
Critical Questions
Which leads to some critical questions:
1. Do we buy or build?
It has long been unfashionable within real estate to entertain industry players building their own technology. Always better to buy in technology from specialists has been the mantra. On the basis that ‘you don’t know what you’re doing, don’t have the talent, and can’t afford to do it properly’.
Today the tables are turned. Given the ease with which a lot of new tools can be developed, or utilised, every real estate company needs to build a level of technical competency, or at least literacy, in-house. You need to know what you’re doing, you need technical talent, and now YOU CAN afford it.
2. If we buy, what does that mean?
As we’ve seen ‘Prompt Packs’ and Agents are going to become easy enough to create and curate in-house, but for the heavy lifts you are going to need help from the AI services companies working on things such as Lease Extraction, Asset Performance Reporting and ESG Analysis.
These are high value, repeatable tasks that are complicated and nuanced - but they can be largely automated. Just not, most likely, by you.
Which means that, unless you are a large player who SHOULD be building this capability in-house, over time the value in this work will accrue to the AI service providers. If you are using someone else’s tools you are vulnerable to being commoditised. And most likely will be.
AI unbundles knowledge from jobs, and reduces the cost of intelligence. Value will move to those that enable this.
The strategic trade-off is clear:
Build = defensibility, talent, control point.
Buy = speed, commoditisation risk, margin erosion.
3. If the ‘machines’ are doing all this work, what are we humans supposed to do?
This is not nearly as hard, or as worrying, as often stated. You just have to be clear about relative competencies. Think of it like this:
AI provides:
Rapid processing of complex, multi-dimensional datasets
Identification of patterns humans might miss
Consistent analytical frameworks across large portfolios
Probabilistic insights to inform human decision-making
Humans provide:
Market context and nuanced interpretation
e.g Local knowledge, regulatory nuanceStrategic judgment and risk assessment
e.g Portfolio capital allocation, risk appetite.Stakeholder relationship management
e.g Tenant trust, investor alignment.Creative problem-solving for novel situations
e.g Complex mixed-use repositioning, adaptive reuse.
Changing Assumptions
So you need to be paying huge attention to what we discussed in ‘AI Fluency is Not Enough’ - as AI removes existing constraints (eg intelligence) it creates others (eg data, coordination and trust). You need to be over-indexing on how the above will reshape the industry.
Where can we add value?
Where can we act as a ‘control point’?
How can we absorb customer risk and guarantee an outcome?
How can we change the story - what was expensive is now cheap, but you now need XY or Z.
And above all else, changing your assumptions about the future:
What constraints within our industry will go away, but what will endure?
What do we need to know that we did not know before, and how long will it take for us to acquire this new knowledge?
What coordination problems will be commoditised (for example which reports)?
What will be defensible - of those four ‘Human’ skills listed above which ones am I/We strongest at?
Should I be updating my five-year plans to two. Or one?
OVER TO YOU
Ask yourself:
Which of my workflows are most at risk of commoditisation?
Where can I act as a control point?
Which human skill is my differentiator?
What’s my planning horizon — five years, or one?
** Obstacles remain, of course, such as fragmented data, tacit knowledge, and organisational inertia. But the trajectory is clear: agents will chip away at each, and this rebundling will force CRE leaders to redefine where they add value.
Two Years In: The Workforce Changes Reshaping Office Demand
Extraordinary AI capability gains are reshaping labour markets faster than most realise—with profound implications for which tenants will need more space, and which will need far less.
Executive Summary
Two years in, AI has leapfrogged in capability. Labour markets are shifting, with junior roles evaporating fastest. For CRE, this means some tenants will shrink, but others—those leveraging AI to expand—will grow, reshaping demand patterns.
Capability Development Phases
Two years ago I ran the first cohort of my #GenerativeAIforRealEstatePeople course. This week we’re starting the 12th.
I thought this would be a good time to review how the AI scene has changed over this period.
We’ve seen three phases of development:
Phase 1: Multimodality (late 2023 – mid 2024)
Focus: AI systems expand from text-only to unified models with images, voice, data, and search.
Nov 2023 – ChatGPT unified model (text, images, data, web search)
Nov 2023 – Arrival of Custom GPTs
Feb 2024 – Gemini expands to 1m token context window
May 2024 – ChatGPT 4o (advanced voice + vision)
June 2024 – Anthropic launches Claude Artifacts & Projects
Phase 2: Reasoning (mid 2024 – late 2024)
Focus: Models get better at structured thinking, coding, and sustained tasks.
Jun 2024 – Claude Code launched (dramatic leap in machine coding)
Sep 2024 – ChatGPT Projects launched
Oct 2024 – ChatGPT o1 (first “reasoning” model; advanced voice & search)
Nov 2024 – Google NotebookLM with audio overview
Nov 2024 – Lovable launched (“vibe coding” movement)
Dec 2024 – ChatGPT o3 (inference training + multi-pass consensus)
Dec 2024 – Gemini 2 & 2.5 Pro (“Thinking mode”, fully multimodal)
Dec 2024 – Deep Research arrives across models
Phase 3: Autonomy (2025 onwards)
Focus: Agents, orchestration, and early autonomous computing.
Jul 2025 – ChatGPT Agent launched (autonomous computing)
Aug 2025 – GPT-5 launched
Exponential Improvement Trajectory
Which means, in short, an enormous increase in the breadth and depth of capabilities.
When ChatGPT-5 launched recently the response was somewhat underwhelming. It was better, but not dramatically so, to the previous models.
However, if one compares the capabilities of GPT-5 with those of GPT-4, launched in March 2023, the difference is enormous.
It’s like the frog that’s been slowly brought to the boil without noticing.
Some have said AI progress is slowing down but, as David Shapiro has said ”AI is slowing down, insofar as most people are not smart enough to benefit from the gains from here on out!”
We have come a long way and have incredible power at our fingertips. All the time whilst, as Google announced last week, their energy cost of AI queries had dropped by 33X in just one year, and for any given unit of intelligence the ongoing trend of becoming 10X cheaper each year is holding steady.
Labour Market Impact Evidence
All of which is starting to have big consequences, even whilst it is still only circa 10% of ‘knowledge’ workers who use these tools every day as part of their workflows.
This week, three Stanford professors released a report, ‘Canaries in the Coalmine’, which examines early signs of how AI is reshaping the US labor market, especially for young and entry-level workers.
Key findings were:
Young workers (22-25) in AI-exposed jobs have seen a 13% drop in employment since late 2022, even as older colleagues in the same roles maintain or increase employment.
The decline is not due to industry-wide layoffs, but to firms quietly not backfilling entry-level positions as they become vacant, erasing the bottom rungs of traditional career ladders.
AI-driven job reductions are concentrated in roles where AI automates codified or rule-based tasks ("book learning"); jobs where AI augments human skills (especially those reliant on experience or tacit knowledge) are not impacted similarly.
Older and more experienced workers in the same occupations are not affected and may even see job growth, underscoring that this is age-specific displacement, not a general workforce reduction.
Wages across age groups remain flat, indicating that firms are reducing headcount rather than cutting salaries—for now.
Researchers controlled for industry and firm factors, showing that the phenomenon is linked specifically to AI-exposure rather than broader economic trends or unique company shocks.
Office Demand Implications
Meaning what for office demand?
Probably less people needing space for doing entry level work, and more space for more senior types capitalising on their ‘tacit’ knowledge.
Anecdotally one is hearing similar dynamics occurring across the professional services arena. The great intake of yearly graduates, populating the base of the pyramid, may be coming to an end?
Similarly, stories such as that recounted by Salesforce boss Mark Beniof last week, that he has sacked 4,000 out of 9,000 customer support staff, are moving from rarity to commonplace.
Now I am sceptical that this trend will last, and indeed my projection is that junior people will come to do well out of this new tech - see Issue #33 - but for now at least things aren’t looking good.
Either way, if younger people DO end up doing better than the research is suggesting right now, that might well be at the cost of more senior people. Same capability, but cheaper, has consequences.
For the office market, we’re going to need a lot of growth to support the market. The dynamic is 100% less people visiting, maybe more space per person for ‘collaboration’, and no mass return to 5 days a week. Without growth, this is a solid ceiling. If you don’t have the ‘right’ space, you’re in trouble.
In Pursuit of Tacit Knowledge
A key question for younger people is how to gain tacit knowledge? I think this might happen in two ways:
‘Someone’ will aggregate corporate tacit knowledge and publish it. This will happen across industries, and will be incentivised by the high value of this knowledge. Maybe ‘old timers’ codifying all they’ve learnt and passing it on. Or maybe through some sort of AI driven automated research tool, that gamifies in some way the passing on of knowledge.
Within corporates, Slack Channels, email, diaries, notetakers and the like will be able to codify all the tacit knowledge in this unstructured data. Hybrid working just makes this easier - as we increasingly work asynchronously, our ‘office’ becomes virtual and all our knowledge and information is contained inside this virtual ‘beast’. Getting to understand the actual workings of a company has never been easier. Tacit knowledge will become explicit just through the process of working.
And so either of these levels the playing field for younger people. And solves the problem of it being all very well getting rid of the cost of juniors but they’ll be no-one to enter the ranks of the seniors. Companies cutting down their intake lose the optionality that has historically given them.
‘Forecasting is hard - especially about the future’, said Nobel prize-winning Quantum physicist Niels Bohr. But…. I’d wager we’re going to see a different dynamic emerge anyway.
And this goes back to something we covered in #Issue4 and #Issue11 - https://www.flexos.work/trillion-dollar-hashtag
Strategic Imperatives
The future belongs to fast, agile, ultra-productive superteams and those designing their business ‘for a bigger pie’.
Incumbents planning their futures on doing what they’ve always done, just faster and cheaper, will be the death of the workforce. A mentality going nowhere and destroying all notions of a ‘social contract’.
Without new ways of doing, and aiming for growth, we’ll all be in terrible trouble.
So let’s not go there.
The lesson of the last two years of Generative AI is that it IS going to be increasingly easy to substitute machines for labour, and that many companies will be incentivised to do just that, ahead of any restructuring or redesigning of their businesses.
So, within CRE, we need to become very good at identifying these types of company, as they are definitely going to be needing less space. And frankly, are customers to avoid.
Focus on Two Things
Where we need to focus is on two things:
First, identifying those companies leaning in to the new technologies, coming up with new business models, and with aggressive growth targets.
And secondly, applying the same thinking to ourselves; as individuals, teams, divisions and companies. We need to be the ones ‘smart enough to benefit from the gains from here on out!’
In Preparation For Next Week
Next week I am going to build on this hypothesis: that we are moving towards an operating model where we’ll execute complete job functions from detailed prompts—define inputs, methodology, outputs, and let the model deliver results. And once optimised, these prompts will be evolved into autonomous agents.
The human role will transform into three functions: workflow design, data curation, and strategic intervention at critical decision points.
On the course we dig deep into where Humans + Machine works best. Where can we achieve AI Human Synergy - doing together what neither could do on their own?
This is going to require new types of working. And a very different mindset.
But, from the work I’ve been doing over the last few weeks, and based on the speed of progress since our first cohort, I’m increasingly convinced that this is not the pipe dream I thought it was a year ago, but actually the framework for what is definitely going to happen.
We shall see. Soon!
OVER TO YOU
Are you feeling the change? Are you thinking of moving job, redesigning your business, looking for different markets?
Do you think we can go straight for ‘a bigger pie’, or will we have to go through the painful transition first?
All my Blog Posts Are Now in My Weekly Newsletter
I now write a weekly newsletter - https://www.flexos.work/trillion-dollar-hashtag - and all my ‘blog posts are now printed there.
I list them all in the Index of the Archive on this site‘
I think you’ll like the newsletter - please do register and let me know.
AI And Bad Workmen
A new report has been caricatured as saying Generative AI is a failure - whereas it actually says that most ‘Enterprise’ users are simply using it badly
Don’t Believe Everything You Read
You might well have heard about a report that came out last week from MIT NANDA (Networked Agents and Decentralized AI) allegedly proclaiming that Generative AI was a busted flush and that 95% of enterprise pilots were failing.
The press, and a certain type of Linkedin ‘Expert’ were all over it. To the extent that, on August 19th, the UK’s Financial Times proclaimed that:
“US tech stocks sold off on Tuesday as warnings that the hype surrounding artificial intelligence could be overdone hit some of the year’s best-performing shares.”
Now, as Mark Twain quipped, one should "Never let the truth get in the way of a good story” but in this case I thought I better check on the veracity behind the cacophony.
And knock me down with a feather, it turns out that the report actually argues that Generative AI is a really useful tool that almost every large company is using incorrectly. Right there in the opening paragraph it states
“Just 5% of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact. This divide does not seem to be driven by model quality or regulation, but seems to be determined by approach.”
Not My Fault Guv’nor
Bad workmen are blaming their tools!
They are calling this the ‘Gen AI Divide’ - the stark gap between those generating considerable value and none at all.
And then they are very clear what the issue is:
“The core barrier to scaling is not infrastructure, regulation, or talent. It is learning. Most GenAI systems do not retain feedback, adapt to context, or improve over time.”
And the solution:
“A small group of vendors and buyers are achieving faster progress by addressing these limitations directly. Buyers who succeed demand process-specific customization and evaluate tools based on business outcomes rather than software benchmarks. They expect systems that integrate with existing processes and improve over time. Vendors meeting these expectations are securing multi-million-dollar deployments within months.”
In order to ‘address these limitations’ they go on to explain how Agentic AI systems are what is required because these can be configured with memory, adaptability, and autonomous workflow integration.
And all this is in the first three pages of the report, which you would not have thought would be too taxing for any ‘expert’ to read.
But then, writing up a report saying enterprises have an incredible new tool at their disposal if used correctly maybe isn’t as click worthy as ‘the bubble has burst - yet again the emperor has no clothes’.
The Enterprise Prerogative
Essentially, at enterprise scale, one has to ensure:
Integration with internal systems (CRM, ERP, lease databases)
Governance/traceability (audit trails, compliance logs)
Continuous improvement loops tied to business outcomes.
They need embedded, auditable, adaptive systems. And off the shelf, or even lightly customised LLMs don’t provide these.
So the report emphasises that enterprises should ‘Stop building static, non-learning tools that require endless prompting’.
What’s Wrong with Custom GPTs?
So why not build Custom GPTs?
The answer is that a Custom GPT can hold static context (instructions, tone, reference docs), but they don’t adapt dynamically from user corrections or workflow usage over time. If you correct an output today, it won’t automatically do it better tomorrow unless you manually retrain or rewrite the system prompt.
Which makes them configured assistants, rather than learning systems.
Big Is Boring - But Where The Money Is
WHICH IS FINE when using them for individual productivity, and within SMEs where the flexibility and adaptability is very much a feature rather than a bug, but in large enterprises you really need ‘process machines’.
It’s noticeable in the report that they say whilst most budget is currently pointed towards front-office applications (sales & marketing, customer operations etc) most of the ROI actually comes from building agents systems to manage back-office operations. Enterprises are process factories, not havens of entrepreneurial spirit. Frankly they need boring systems to do boring jobs that currently bored humans do.
2 out of 9 IS Bad
In their research covering nine industry categories, they found only two, Technology & Media and Telecoms, were on the right side of the ‘Gen AI Divide’. In these two industries, the impact of Gen AI was such that it was acting as a forcing function for structural change, and entire business models and operating models were being reimagined. For all the rest it was a world of bolt-on pilots, static tools and little ‘change’. In addition the levels of training were poor (only 40% paid for ChatGPT or the like Licenses), leading to a large amount of shadow IT, and a lot of ‘build-it-ourselves’ thinking was leading to botched implementations.
Where AI Is Working: Individual Adoption
Where these companies were having success was in areas they weren’t focussing on. Individual use by employees. In interviews they found that drafting, analysis, summarisation and outreach were very popular use cases at the edges, and adoption of consumer tools like ChatGPT was way ahead of anything handed down from above.
It Probably Does Not Apply To You Anyway
This report is very much focussed on larger enterprises, with upwards of thousands of employees. Mostly CRE companies don’t fit in this bucket. For instance the ‘large’ UK REIT, British Land, only directly employed 634 people as of mid 2024, and most developers, asset managers, and agencies operate with dozens to a few hundred staff. CBRE, JLL, Cushmans et al. are of course a lot bigger, but en masse CRE is an industry of SMEs. That often thrive by letting the bottom up approach reign.
As such, whilst (as we’ve written about before) we are going to see a lot of use of ‘agentic systems’ within our industry, we are not going to be constrained by the same imperatives that much larger companies have to operate under. And therefore should, as we already are, see a lot of Gen AI use at an individual or team level. People just making stuff happen, and pushing AI models to help them with whatever they have to deal with. No massive, centralised, bureaucratic plan needed.
And agility is a super-power in CRE. There are so many workflow/operating model innovations possible - leasing teams piloting deal-structuring agents; asset managers automating ESG reporting; development teams using AI for scenario modelling, and on and on.
The Bottom Line
Ultimately, the media's misreporting of this study perfectly illustrates the report's actual findings. The report describes a failure of implementation at the enterprise level, where structural change is needed but not forthcoming. We're seeing a lot of 'sound and fury... signifying nothing'. A complete bum steer.
Frankly this is not surprising - this technology is a ‘disruptive’ innovation, and that hurts, harms and annoys many people. So it is little wonder we’re seeing a degree of backlash right now. Who wouldn’t rather be safe and secure than throwing themselves into the rough and tumble of the ‘new’?
I suspect most readers of this newsletter are happy to be somewhat unnerved by all the change going on. We’re not the complacent type are we?
OVER TO YOU
What’s your greatest success with Gen AI? Where are you seeing how you work, or think, or act, change? Does this bother you? Let me know.