Eight years ago, radiologists were declared obsolete in the face of AI. They're still here, but the job has changed. That's exactly the scenario waiting for real estate agents, and this week sketches out its shape. Cold calling is closing down and forcing a return to local marketing, Google AI Overviews confirms the traffic drop on French sites, water is becoming a valuation criterion while the official indicator picks the wrong cities, and Uber faces nearly a billion dollars in fines for letting an algorithm decide alone. Five signals about what AI actually shifts: not the jobs, but where the value sits inside them.
A week that answers the real question: what does AI actually shift?
The question is no longer "will AI replace real estate agents?" but "which part of the job still holds value?". This week, the answer emerges from five unrelated stories: an article about radiologists, two pieces of French regulation, a web traffic study, and a record fine. They all point to the same place.
Radiologists didn't disappear - but their job is no longer the same
Back in 2016, declaring the end of radiologists was fashionable: AI already read images better than they did, so it was just a matter of time. Eight years later, they're still here, and in short supply. What changed isn't their existence, it's what fills their days: less time spent detecting, far more spent arbitrating, contextualizing, explaining to the patient, and owning the final call.
The parallel with real estate is almost too obvious. Whatever is automatable in an agent's job - writing a description, retouching a photo, generating a floor plan, pre-qualifying an inquiry - will be automated, and fast. What remains is everything that requires putting your name on a decision: pricing accurately in a market you know, telling a seller what they don't want to hear, holding a negotiation. The real risk isn't being replaced, it's continuing to bill time on the part that's about to become free.
Source: Ars Technica, "AI won't replace radiologists, but it will dramatically change their jobs", 2026.
Cold calling closes down: local marketing becomes mandatory again
New consent rules on phone prospecting sharply narrow what a professional can do: no more cold outreach to numbers collected along the way, and prior consent that must be explicit and traceable. For many agencies, an entire acquisition channel shuts down overnight.
At the same time, the conversation about reclaiming local marketing is back in force - and the timing is no coincidence. When you can't go get sellers by phone anymore, they have to come to you, which means being visible and credible in your area: local content, presence on the ground, client reviews, and serious work on what the agency actually publishes. The regulatory constraint mechanically pushes toward slower marketing, but also more durable marketing - an asset the agency owns, unlike a list of phone numbers.
Sources: Journal de l'Agence, "Phone prospecting: the new consent rules you can't afford to miss" and "How to reclaim local marketing?", 2026.
Google AI Overviews: the traffic drop is confirmed on French sites
The data keeps accumulating and points the same way: since AI Overviews rolled out, French sites are seeing organic traffic decline. The mechanism is simple: Google answers directly on the results page, and the user has no reason left to click the three links that produced that answer.
For an agency, the consequence is strategic. Ranking well on "apartment valuation + city" no longer guarantees a visit; what matters now is being the source the engine cites in its answer. That changes how you write: clear, self-contained answers, dated and sourced figures, pages structured to be quoted as-is rather than to keep the visitor as long as possible. It also strengthens the case for channels you don't rent from Google: the newsletter, the direct relationship, local reputation.
Source: Immobilier 2.0, "Google AI Overviews: the traffic drop for French sites is confirmed", 2026.
Water becomes a value criterion - and the official indicator picks the wrong cities
Two stories that answer each other. The first: drought risk, consumption, and water-related equipment are settling in as property valuation criteria, especially in southern France - a very concrete topic when you work, as we do, between Aix and Marseille. The second: the summer comfort indicator in the French energy performance certificate targets the wrong cities, flagging municipalities that aren't particularly at risk, and missing others.
Side by side, they say something important about data. A criterion becomes decisive for a property's value at the exact moment the official indicator meant to measure it shows its limits, because it rests on theoretical averages rather than the reality of a given building. That's a useful reminder as more and more valuations get automated: a model fed a biased indicator will produce biased valuations, with great confidence. On the ground, actual exposure, wall inertia, and available shade beat a theoretical rating.
Sources: Journal de l'Agence, "Drought risk, consumption, equipment: water, a new value criterion for properties" and Mon Immeuble, "Summer comfort in the energy certificate: the official indicator picks the wrong cities", 2026.
Uber: nearly a billion in fines for letting the algorithm decide
Uber faces a fine of nearly $1 billion for suspending drivers automatically, without meaningful human involvement in the decision. The amount isn't there to punish the use of an algorithm, but the absence of a responsible person at the end of the chain when that algorithm gets it wrong.
The lesson goes well beyond ride-hailing, and echoes what we're seeing elsewhere: at Meta, AI agents deployed to replace teams caused large-scale disruptive actions before the plan was walked back. In an agency, the principle is identical: AI can prepare a rental file, prioritize inquiries, draft a first valuation. It must never be the last link on a decision that affects someone. This isn't a moral stance, it's risk management - and now it has a price tag.
Sources: TechCrunch, "Uber faces fine of nearly $1B over automated driver suspensions" and Ars Technica, "AI agents meant to replace Meta workers made large-scale, disruptive actions", 2026.
The thread this week: the job isn't disappearing, it's refocusing
Radiologists still here but transformed, cold calling shut down, Google traffic eroding, official indicators proving imperfect, a record fine for automated decisions: all these signals push the same way. What's losing value is production - of text, of contacts, of clicks. What's gaining value is judgment: knowing your area, making the call, and owning it.
So, in a typical week: how many hours do you still spend on tasks a machine will handle just as well next year?
