We’re about to see work reinvented. We’re now at a moment in time when change goes from slowly to suddenly, and huge opportunities open up to multiply our cognitive capabilities and transform our productivity. Let the revolution commence.

A Sunday Epiphany: Watching Agents at Work

Last Sunday I was playing around with OpenAI’s new ChatGPT Agent service which, according to them, ‘can now do work for you using its own computer, handling complex tasks from start to finish.’

You give it a task, it fires up its own ‘virtual computer’ within your browser window and then gives you a running commentary about what it is deciding, autonomously, to do in order to fulfil the task you gave it. It feels like you have a ‘ghost in the machine’, an alien intelligence moving the digital world on your behalf while remaining utterly intangible.

And to me, it felt like a revelatory moment. This was something truly different.

A Break in the Timeline

In evolutionary biology there is a term, ‘punctuated equilibrium’, that suggests species typically experience long periods of little or no change—called stasis—interrupted by short, rapid bursts of significant evolutionary change. This model contrasts with the traditional idea that evolution happens slowly and gradually over time in a steady, continuous way.

And this is what the rapid development of generative AI is replicating. Since the launch of ChatGPT on November 30th, 2022 there has been a whirlwind of progress. And the whirlwind has even been speeding up. The arrival of ‘reasoning’ models in December 2024 has seemingly cranked up the gears across all the frontier labs, OpenAI, Google and Anthropic, and the cadence of new releases and features is now weekly.

The previous equilibrium has been rent asunder and we’re in a new era. The AI Age has truly arrived.

This Time is Different

These four words are meant to presage a financial crash. During boom times whenever you get to the stage where people say ‘This Time is Different’ is usually the first ‘sell’ signal. And it has repeatedly proved true throughout history.

But I’m going to steal them anyway and apply the spirit to the world of work.

Using ChatGPT Agent strongly pushes me to say ‘This Time is Different’. This way of getting work done is something else entirely. It is not something we’ve ever had before, and it is going to redefine how we ‘get stuff done’ and how we conceive of a ‘workforce’.

Now, to be clear, ‘Agent’ is not the first service in this area. Notably, the Chinese-backed company Manus has had a service out there for some months doing similar things, if in a slightly different way. But you’ve probably never heard of them. Whereas you’re 99.9% certain to have heard of OpenAI and ChatGPT. With over 500 million weekly active users, this new way of working suddenly has massive distribution and public awareness. Meaning it will be taken up by significant numbers very quickly.

What This Means for Real Work

So what will that mean?

Well let me give you some examples of how I’ve used it to date, and some indications of where this is heading.

First, I asked it to visit the index page of this newsletter, read every edition, summarise each one, add these to a table and then separately write a meta summary of the topics I cover in the newsletter.

It took about 15 minutes, and did the job perfectly.

Secondly, I gave it a lengthy ‘Deep Research’ report into which companies within real estate were providing software services involving generative AI. I asked it to extract every company mentioned, find their website address, visit said website, and summarise the asset classes they operated in and the services they provide. And, again, put it all in a table.

Job done in 20 minutes.

Thirdly, I uploaded an Offering Memorandum and told it to find comparable sales within 5 miles. Then assess for sensitivity on rent growth 0% vs. 3% vs. 5%. Then provide a one-paragraph executive summary for investors. Then build a PowerPoint slide with deal highlights.

Job done in 15 minutes.

You might have read Arthur C Clarke’s Third Law where he writes ‘Any sufficiently advanced technology is indistinguishable from magic’. Well, that is exactly what I felt about all of this. Magic was afoot.

It’s really quite a bizarre thing to ask a multi step question, provide no other guidance, and then see a ‘machine’ itemise in text how it is going about arriving at an answer.

This Time is Different.

From toy to Power Tool

Obviously all of this does not work perfectly yet and one would not point it at mission critical work. But …. give it six months, or a year, and one can be pretty confident it will be ready for important work.

And when that happens, how we work might change fundamentally. I can see no reason why we won’t be running dozens of these ‘Agents’ 24/7 on repeatable questions, processes or workflows. Constant implementation of X, constant research about Y.

Let’s extrapolate a couple of potential use cases:

1. AI Underwriting Assistant

For input we might upload Offering Memoranda (PDF), Rent Rolls (Excel), or deal data to a shared drive or folder. And set a ‘trigger’ of a new file detected in the "Deals to Underwrite" folder (e.g., via Zapier or cloud API).

Then the ‘Agent’ would autonomously do the following -

• Extract key data: lease terms, income, expenses, cap rate, market comparables
• Perform rent growth/cap rate sensitivity
• Benchmark against portfolio and market data
• Flag inconsistencies or missing assumptions
• Write a one-page underwriting summary
• Set Risk flags (e.g., short WAULT, high OpEx ratio)
• Provide a sensitivity matrix (e.g., IRR vs. rent growth)
• Optionally create an IC-ready memo or PowerPoint slide deck
• Notification : Email, Slack, or CRM update sent to Acquisitions team with summary and link to outputs.

2. Market Intelligence Synthesiser

For input we might have a predefined watchlist of locations, asset classes, tenants, and competitors.. And set a ‘trigger’ that runs on a daily schedule (e.g. 6:30am London time).

Then the ‘Agent’ would autonomously do the following -

• Scrape news from key outlets (e.g., FT, Property Week, Bisnow, CoStar)
• Pull comps and pricing signals from public data feeds
• Aggregate macro indicators (e.g., Gilt yields, inflation prints)
• Cluster insights by theme: pricing, regulation, supply, occupier trends
• Produce a smart digest with summaries, headlines, and source links
• Create charts for rent/yield movements and liquidity trends
• Optional: brief commentary or strategy nudge
• Notification :Delivered to email or Notion dashboard; summary ping in Slack.

And so on. Think of a process that follows this pattern. I suspect you have many. Maybe soon ‘Agent’ will be doing them for you.

How to Build Your Own Agent Workflows

So far I’ve only played around with this new tool. But the obvious direction of travel will be to define your own uses cases, then:

• Define the use case
• Craft and refine your prompt
• Specify the required outputs
• Provide data inputs (internal + third-party)
• Upload presentation templates
• Set automation triggers
• Establish review and notification criteria

It will take some setting up, and some iteration, but these extensive workflows will be possible to achieve. If not today, then ‘soon’.