This is the second article in a series on AI and the future of real estate brokerage. The first piece - Portals, CRMs, and Traditional Agents: Which Part of Real Estate Survives the Autopilot? - examined the structural shift across the whole industry. This article goes deeper on the agent - and what the structural shift means. These are my own conclusions from building properti. The research confirms what I observed.

There is a question I hear constantly at real estate conferences, from agents, from investors, from competitors: 'Will AI replace the agent?'

In my opinion it is the wrong question. And I know it is the wrong question because I have watched 11'000+ transactions unfold across one of Europe's most demanding real estate markets - which gives me a particular vantage point that no researcher, however rigorous, fully has.

The right question is: which parts of what agents do today will AI do better - and which parts become more valuable precisely because AI takes everything else off the table?

My answer is not optimistic or pessimistic. It is structural.

McKinsey estimates that AI applied to real estate knowledge work could unlock $430-550 billion in annual value globally. Sequoia Capital named real estate brokerage one of the largest unsolved autopilot opportunities on the planet. I also spoke with over 20 founders rebuilding their organisations around AI and found that some of the founders have planned to cut 120-person engineering teams to 25 - not because the work disappears, but because humans move up the stack.

And here is the uncomfortable truth that most brokerage executives will not say out loud: the majority of agents working today are not advisors. They are paper processors. AI does not replace those agents. It simply makes visible what was always true - that the value they claimed to deliver was never really there.

The Two Ingredients of Every Transaction

McKinsey's agentic AI framework introduces a distinction that every real estate professional should memorise: the difference between steps and thoughts.

"Steps are repeatable tasks that benefit from speed, consistency, and clean handoffs. Thoughts are judgment calls that require discretion: exceptions, trade-offs, expressions of taste or creativity, gestures that preserve trust, and decisions that carry financial, reputational, or regulatory risk." - McKinsey, March 2026

Sequoia Capital calls the same thing intelligence work and judgement work. The terminology differs. The insight is identical.

Every real estate transaction is a combination of steps and thoughts. The steps are automatable. The thoughts are not - and in fact become more valuable as AI handles everything else.

The problem is that most agents today spend 70-80% of their time on steps. Generating valuations. Writing listing descriptions. Processing follow-up sequences. Scheduling viewings. Chasing documents. Updating CRMs. Coordinating paperwork.

These are not judgment calls. They are repeatable, rule-based, learnable tasks that AI already does faster, more consistently, and at a fraction of the cost.

It's simple: automate steps aggressively. Protect thoughts deliberately.

The agent who does not make this distinction is not just inefficient. They are structurally obsolete.

What I Have Learned from 11'000+ Transactions

Every transaction I have been part of building - as a founder, as a platform operator, as someone who has sat with the data and the people - has reinforced one distinction above all others.

McKinsey calls it steps versus thoughts. Sequoia calls it intelligence versus judgement. I call it the difference between what a computer can do and what a human must do.

"Steps are repeatable tasks that benefit from speed, consistency, and clean handoffs. Thoughts are judgment calls that require discretion — exceptions, trade-offs, expressions of taste or creativity, gestures that preserve trust, and decisions that carry financial, reputational, or regulatory risk." — McKinsey, March 2026

The problem is not that AI is coming for agents. The problem is that most agents today spend 70-80% of their time on steps - generating valuations, writing listing descriptions, processing follow-up sequences, scheduling viewings, chasing documents, coordinating paperwork.

None of that is judgment. All of it is automatable. And the agent who continues spending 80% of their time on work that AI does better is not just inefficient. They are structurally obsolete.

The agents who recognise this shift early - and spend their time on what only a human can do - will not compete in a shrinking market. They will compete in an expanding one. This is the Jevons Paradox applied to real estate: more automation creates more demand for the judgment that automation cannot replace.

The Architecture That Makes the Shift Possible

This is where I want to be precise, because most discussions about AI and real estate get this wrong.

The shift from steps to thoughts does not happen spontaneously. An independent agent with six fragmented tools cannot eliminate the steps - they are too busy doing them to develop the judgment work that makes them irreplaceable.

What is required is what Benedict Evans calls the 'cars on dirt roads' problem. The car existed. But its value could not materialise until roads, highways, and cities were redesigned around it. Adding a new engine to old infrastructure produces incremental gains, not transformation.

This is why I believe the future of real estate brokerage is not the independent agent with AI tools. It is a two-party architecture - and only two parties - each with a function the other cannot perform.

The two parties are the platform and the agent. Not four roles. Not a management layer. Two.

Neither party can do what the other does. The platform cannot build trust in a first meeting. The agent cannot accumulate the decision intelligence of 11'000+ transactions. Together, they produce something that neither an independent agent nor a pure software platform can replicate.

This is what Foundation Capital describes as the Context Graph: a searchable record of how decisions were made and why - not just what happened, but under which conditions, with which strategy, and to what result. Every transaction properti closes adds to this graph. Every agent on the platform benefits from every transaction that came before them.

What Thoughts Look Like in Practice

I want to be concrete about what the agent's side of this architecture actually involves. These are not hypothetical scenarios. These are patterns I have observed directly.

The seller who is not ready

A 67-year-old woman has lived in her house for 34 years. Her children are encouraging her to sell. She has agreed to list. But in the first meeting, she is not asking about the price. She is asking about the garden, the neighbour's tree, the light in the kitchen in the morning.

The platform scores her as high-conversion. The property is perfectly positioned. The mandate paperwork is ready.

A good agent recognises that she is not ready to sell. That pushing for the mandate today means a failed listing in four months. That the right move is to slow down, come back in six weeks, and wait until she has made peace with the decision.

No model produces this insight. It requires experience, presence, and the willingness to walk away from a short-term conversion in service of a long-term outcome.

The offer that is not what it appears

Three offers arrive. The highest is 5% above asking. Standard logic says take it.

An experienced agent checks the financing. The buyer is pre-approved but at the limit of their loan-to-value. The property has a structural detail a bank valuer might flag. If the valuation comes back below the offer price, the deal collapses three weeks before closing - after the seller has already committed elsewhere.

The second offer is 2% below asking. Enough cash on the bank, can choose the mortgage. Notary appointment already proposed.

Which offer to recommend? The platform provides the data. The agent makes the call - and stands behind it.

The negotiation that requires presence

The buyer's final offer is CHF 30'000 below the seller's minimum. The seller wants to relist.

The agent has fifteen minutes on the phone. Not to split the difference - but to understand what is actually driving the impasse. Is the buyer at their limit or testing? Is the seller's minimum anchored to a number they told their friends?

Knowing which dynamic is in play requires pattern recognition built from hundreds of similar conversations. Acting on it requires the kind of trust that a seller extends to a human being - not to a software interface.

"Residents remember how you handled one bad moment, not nine routine ones. When agents handle routine steps consistently, people can focus on the work that requires judgment, taste, creativity, and presence." — McKinsey, March 2026

The Market That Is Already Bifurcating

This is not a prediction about 2030. It is a description of what is already visible.

On one side: agents who have recognised the shift. Who are part of platforms that handle the intelligence layer. Who spend their time on the negotiation, advisory work, and high-trust moments that define good brokerage. These agents close 35, 50+ transactions per year. They earn more. They serve clients better. They are structurally irreplaceable because their value proposition is judgment and trust - not information.

On the other side: agents who are still information brokers. Who spend 70% of their time on tasks that AI already does better. Who compete on commission rates because they cannot demonstrate superior outcomes. In Switzerland, over 5'000 agents close fewer than 5 transactions per year. Most of that activity will disappear over the next five years - not because AI replaces these agents, but because agents who make the shift make agents who do not entirely uncompetitive. 

The Question Every Agent Should Ask Today

Not: 'Will AI replace me?'

But: 'What percentage of my working week am I spending on steps that AI already does better - and what would I do with that time if I got it back?'

If the answer is above 50%, the platform question is not optional. It is urgent.

The shift from information broker to transaction strategist does not happen by itself. It requires infrastructure that handles the steps so thoroughly that the agent has no choice but to focus on the thoughts.

I believe that infrastructure - when it is truly built around AI (or even AI-First) and not just equipped with AI tools - changes not just how many deals an agent closes, but what kind of professional they become. More present. More decisive. More valuable. This is the model we are building in Switzerland and scaling across Europe.

That is the bet properti is making.

The agents who find this platform - and commit to the shift it enables - will not just survive the next five years.

They will own them.