The Impact Edit: Automated Decisions, Human Consequences
Hello, it has been a minute.
In my defense, I spent the spring and summer finishing up a book, recording a podcast season, and researching a question that became a bit of a rabbit hole. All while delivering client work, meeting deadlines, and keeping Liveable moving.
These are three things I would have strongly advised a friend not to attempt at the same time. I never said I was good at following my own advice.
Anyhoo-the research is what I want to tell you about. It's 2026, so of course, it's about AI.
Our buildings are making decisions now
Somewhere along the way, without a memo, podcast or a press release, the software in our buildings started making decisions about people. No, not assisting with decisions- making them.
For example, dynamic pricing engines are adjusting renewal rates. Screening platforms are ranking rental applicants before a human ever opens the file. Security analytics are deciding whose behavior looks "anomalous." Maintenance algorithms are deciding whose repair request can wait another week. If you work in or around real estate, some version of this is already running in your portfolio, whether or not anyone showed it to you.
Here is the number that sent me down the rabbit hole. According to JLL, 92% of corporate occupiers and 88% of investors are piloting AI right now. Only 5% say they have achieved most of their goals with it. Everyone is experimenting, almost no one is succeeding, and meanwhile, the systems already deployed are quietly shaping who gets approved, what they pay, and how they are watched.
These tools are not neutral. Far from it- they learn from historical data, and housing data carries history's patterns. A screening model trained on decades of biased decisions does not correct those decisions. It repeats them, faster and at scale, behind an interface that looks objective. That is the part that should keep us up at night, and it is what our new spring brief is about.
Diving into the findings
We mapped where these systems sit across the building lifecycle, from planning to leasing, and where accountability breaks down.
Three patterns kept showing up:
Decisions get delegated by default. High-stakes calls about who is approved and how they are billed get handed to software without anyone formally accepting the risk.
Systems operate as black boxes. Organizations rely on outcomes they cannot explain, test, or challenge. Ask a leasing team why the algorithm priced two renewals differently and watch the room go quiet.
Responsibility fragments. The owner assumes the vendor validated the logic. The vendor assumes the operator is monitoring outcomes. The operator assumes someone upstairs signed off. Nobody did.
I do want to be careful here, because this is usually where it is tempting to tip into alarmism, and that is not where I am. What pulled me in was how much of this is already documented, in court filings and settlements rather than opinion pieces.
You can read our full brief- it is free, and it includes the five questions we now ask of any automated system: where does it affect people, who owns that influence, what evidence supports it, how are outcomes reviewed, and what happens when something goes wrong. Read it here: https://makeitliveable.com/spring-brief
Three conversations that made me think differently
The same question runs through the new season of my podcast, Changing Tomorrow, where I asked three people who sit close to these systems what People Positive AI actually looks like. Each of them surprised me.
Janet Pogue McLaurin, FAIA, FIIDA who leads Gensler's global workplace research, shared a finding I did not believe at first. Across 16,000 workers in 16 countries, the heaviest AI users spend less time working alone and report stronger relationships with their teams. We assumed AI would empty the workplace. Instead it is changing what the workplace is for.
Zainab Garba-Sani, a Stanford research scholar, said something that I think all of us need to hear: opting out of AI is not neutral. If your community is missing from the data, the systems get built anyway, and they fail you specifically. She has watched it happen in healthcare, where an algorithm used past spending as a proxy for medical need and quietly deprioritized the patients who had received the least care all along.
And Archita Mandal award winning filmmaker and tech founder of UpendNow Inc., fixed a phrase I did not realize was broken. We keep saying "human in the loop," as a definition of best practice in AI systems- almost as if the machine runs the process and we just check its work. She argues it should be AI in the loop. Humans direct- and the tool executes. The difference sounds small until you notice which one your organization is actually practicing.
All episodes are live now, wherever you listen: https://changingtomorrow.buzzsprout.com
With light,
Gayathri
The Conversation Continues...
This post is part of our ongoing exploration into how automated decision-making in buildings quietly shapes human outcomes, and why closing the AI accountability gap requires moving from black-box delegation to human-directed governance. As problem-solvers, we believe the best insights emerge when diverse perspectives meet. Have you encountered similar challenges or discovered different approaches? Share your story.
Connect with us as we continue to prototype, test, and learn:
Subscribe to our newsletter
Join us on Linkedin
Explore our resources
We acknowledge that social sustainability is always a work in progress. These insights represent our current understanding, shaped by our partners, communities, and continuous learning.