Real Estate Jobs Are Bigger Than the Task List

Every Inc. is close to the kind of company people imagine when they talk about AI replacing work. Almost 30 people, agents across the business, AI handling most of the CEO's work email, and a team using the newest models across coding, writing, design, customer service and internal operations. If there was ever going to be a clean example of the replacement story, this would be a decent place to look.

Dan Shipper's recent essay, After Automation, caught my attention for that reason. Every has pushed hard into agents, and yet the lesson seems to be more complicated than the early AI employment story suggested. They have automated a lot of work, the company looks very different because of it, and they still have people doing customer service, writing, editing, engineering and the harder work of steering, judging and improving what the agents produce.

A few months ago, I wrote about the fear that AI was going to take everyone's work. My view then was that the most exposed work would be highly repeatable, structured and easy to define, while jobs built around trust, judgement, curation, history and wisdom would be much harder to replace. I still think that is broadly right, but the more I look at what is happening inside businesses now, the less useful the simple "job replacement" frame feels.

A job is rarely just the sum of the tasks attached to it. That might be how it looks in a position description, but it is not how work actually happens during the day. Most roles are a live mix of repeated work, exceptions, interruptions, handoffs, incomplete information, conversations, internal context, small decisions and judgement calls that only become obvious once something has gone slightly off track.

That is especially true in real estate, where an agency admin role can look highly automatable from the outside. Updating the CRM, preparing documents, booking photography, chasing missing information, drafting emails, compiling vendor reports, moving details between systems, setting reminders, checking contracts, arranging marketing and keeping campaign information current. On paper, a lot of that work looks repetitive. In practice, it arrives out of order, gets interrupted constantly, depends on other people, and often requires someone to decide what matters now rather than what was next on the list.

The same is true for agents. A buyer asks for the contract just as a vendor calls with a concern about price feedback. Someone from the office asks where the keys are. A conveyancer needs a document. A past client messages about a property they saw online. A junior agent needs help with a difficult inspection comment. A campaign report needs one more piece of interpretation before it goes out. None of that sits neatly inside a list of tasks, but all of it is work.

Microsoft's WorkLab data gives this a useful shape. Its 2025 "infinite workday" report says Microsoft 365 users are interrupted every two minutes during core work hours by meetings, emails or notifications, with the 275-pings-a-day figure applying to the top 20% of users by ping volume. It also found that most meetings are either ad hoc or unscheduled, which says something about how much modern work now happens in the spaces between the formal plan.

That feels very familiar if you have spent time inside an agency. The day is calls, walk-ins, texts, inspections, vendor nerves, buyer questions, internal questions, CRM notes, contract requests, price feedback, diary changes and follow-up that has to happen while the signal is still fresh. The interruptions are not always distractions from the job. Quite often, they are the job.

The Every example is useful because Shipper describes a pattern they call the "human sandwich": a human frames the work, AI carries much of the middle, then a human comes back to judge, adjust and extend the result. It sounds slightly clunky as a phrase, but it is a pretty good description of where a lot of professional work seems to be heading.

For real estate, that pattern is easy to see. The agent has the conversation with the owner, buyer or past client. AI then pulls together the history, drafts the follow-up, summarises the call, checks the CRM, prepares comparable sales, sets the next action, updates the record and gives shape to the information that would otherwise sit across email, phone notes, memory and a half-finished task. The agent or assistant then comes back at the end to apply judgement, tone, local feel and accountability, which is a much more realistic model than imagining the person simply disappears from the loop.

Take an appraisal... the human part is reading the owner, understanding motivation, framing the market honestly, judging timing, building trust and deciding how hard or soft the next conversation should be. Wrapped around that is a lot of repeated preparation work: property history, comparable sales, suburb activity, likely buyer segments, past database interactions, a pre-appraisal note, a follow-up sequence and a clean record of what was said. AI can carry more of that middle layer, but the moment still needs a person who understands what is really happening in the room.

A buyer follow-up loop works the same way. The buyer calls or inspects, the system links them to the property, checks previous enquiry history, identifies whether they have asked for contracts before, records the key points, drafts the next message and brings them back into view at the right time. The useful part is not just speed. It is that the intent does not evaporate because the agent had six other calls before getting back to the desk.

Vendor reporting is another obvious one. The numbers can be assembled, inspection comments can be grouped, buyer sentiment can be summarised, campaign activity can be pulled together and a draft report can be prepared. But the value of that report still sits in interpretation. What does the feedback mean? Is the price issue real or tactical? Is the vendor ready to hear the harder message? Should the next step be a report, a call, or a more direct conversation?

That is why I think the admin layer will change in a more nuanced way than a lot of people first expected. It may not be one large replacement event. More likely, it becomes a set of smaller agentic loops sitting underneath the business: appraisal preparation, buyer follow-up, open-home feedback, vendor reporting, contract requests, CRM hygiene, past-client prompts, campaign summaries, marketing drafts, compliance records and diary handling. Each loop has a person at the start, AI carrying more of the middle, and a person at the end where judgement still matters.

McKinsey has been describing a similar idea in real estate through its distinction between steps and thoughts. The steps are repeatable tasks that benefit from speed, consistency and clean handoffs. The thoughts are judgement calls involving discretion, exceptions, trust, taste, risk and trade-offs. Read through a residential sales lens, that is a useful way to think about the business, because a lot of the steps around the relationship are becoming easier to systemise while the relationship itself still depends heavily on human judgement.

There is a Jevons paradox quality to this. When something becomes cheaper and easier to do, people tend to do more of it, and once that happens automation does not always make work smaller. If appraisal preparation becomes easier, more appraisals can be prepared properly. If buyer follow-up becomes easier, more buyers can be followed up with context. If vendor reporting becomes easier, reports can become more useful rather than just faster. If database work becomes easier, more old relationships can come back into view. If basic content, editing and campaign material become easier to produce, more of that work may come back inside the agency rather than being pushed to external suppliers.

The office may still feel busy, possibly even busier in some places, because more of the "we should really get to that" work starts becoming realistic. The more useful labour question is not only what cost AI removes, but what capacity it creates once repeated admin, basic production, simple research, first drafts, some external contract work and parts of the SaaS stack become easier to carry. The larger effect may be that the people who remain can hold more live work, with better context and fewer loose ends.

For small agencies and individual operators, that could be a very big shift. A solo agent or small team has always been limited by the amount of coordination they can carry around the relationship. The calls, records, follow-up, reporting, reminders, prep work, content and admin all sit around the parts of the job that actually build trust and win business. If more of that support layer becomes system-driven, a smaller operator starts to feel more like an always-on agency, provided the human remains firmly in control of the loop. Every's essay also points to the second problem sitting inside that opportunity, which is sameness.

When everyone has access to capable agents, the baseline quality of work lifts, but it also starts to flatten. Emails get cleaner, summaries get neater, drafts get more structured and reports become easier to produce. A lot of it can start to feel competent and generic at the same time.

Real estate is already vulnerable to that. Listing copy starts sounding the same, market updates start sounding the same, follow-up emails start sounding the same, social posts start sounding the same. The more AI lifts the baseline, the more valuable the specific human layer becomes: local knowledge, restraint, timing, memory, tone, taste, commercial instinct and the ability to say something that actually fits the client, the property and the moment.

The recent pendulum swing around AI and labour is worth watching through that lens. Klarna became one of the clearest examples of aggressive AI use in customer service, then later talked about keeping a human connection for higher-value support. Starbucks recently retired an AI inventory-counting tool in North America after problems with execution and accuracy. These are not signs that AI is going away. They are reminders that operational reality is usually messier than the first productivity story makes it sound.

Real estate has always been messy in that way. The work is personal, commercial, fragmented and highly time-sensitive. A lot of the operating layer is repetitive, but the sequence is rarely stable. A lot of the communication can be drafted, but the judgement behind it cannot be assumed. A lot of the follow-up can be prompted, but someone still has to know when the prompt is useful and when the situation calls for something else.

So when I think about AI inside an agency now, I am less interested in whether one role disappears cleanly and more interested in where the work moves once parts of it become cheaper, faster and easier to carry. Does the saved time become better vendor communication? Does faster appraisal prep become more consistent prospecting? Does easier reporting become sharper campaign advice? Does cleaner CRM data become more useful past-client work? Does AI-generated content become more of the same, or does it free people to produce something more specific and more useful? Those are operating questions as much as technology questions, and they point to whether the agency actually understands how its own work moves.

The agencies that benefit most will probably be the ones that can break the work into the right loops. Where does this piece of work start? What information does it need? Which steps can be carried by a system? Where does a human need to approve, adjust or decide? What should happen next without relying on memory? Where does the workflow currently leak?

That feels like the practical version of the Every lesson for real estate. The human layer does not become less important because AI gets better. In many ways, it becomes more exposed. When the system can carry more of the middle, the quality of the human framing at the start and the human judgement at the end matters more, because that is where the work becomes specific, trusted and commercially useful.

A few months ago, the fear was that AI would take the work. The better question now is what happens when AI makes more of the work possible. My guess is that the best real estate operators will use it to carry more of the work that already should have been done: more buyers followed up properly, more appraisals prepared well, more vendors updated with context, more database signals brought back into view, and more human attention spent where it actually changes the outcome.

After automation, the job does not vanish as cleanly as the early story suggested. It gets redrawn around better systems, more capacity, and a sharper need to know where people still matter most.