Once a quarter, TXI stops to take a break from delivery work. The whole company, all at once, sets down client work for a few hours to learn together. This quarter, our DEIB education arc landed on environmental justice, one of our four justice pillars alongside racial, disability, and gender justice. And this year, we've been weaving that education directly into our core strategy: evolving TXI from a digital product company into an intelligent product company. So I went looking for someone who lives at the intersection of both. I found her in Dr. Sasha Luccioni.
Dr. Luccioni is one of the most credentialed voices working on the environmental and social costs of AI today. She holds a PhD in AI, is Co-Founder and Chief Scientific Officer of the Sustainable AI Group, helped build Hugging Face's climate work before that, and was named one of TIME's 100 Most Influential People in AI and one of the BBC's 100 Women. She's given two TED talks on the subject, and her research has become a reference point for how the industry measures AI's energy use. This was, without question, the biggest name we've brought to an all-hands.
We didn't keep her to ourselves, either. Climate, sustainability, and environmental justice are bigger than any one company's quarterly meeting, so we opened the room. We invited our competitors and collaborators from the Kermit Collective, the group I founded to foster cooperation among consulting firms who'd rather solve industry-wide problems together than pretend we're not all wrestling with the same ones. We also invited our community partners at the Chicagoland DEI Alliance to dial in. If the conversation about building AI responsibly is going to matter, it can't just happen inside our own walls.
What Dr. Luccioni gave us over the next hour reframed how a lot of us think about the systems we're building.
The cost we don't see
Dr. Luccioni opened with scale. Computers used to train AI models have grown roughly fourfold every year, and the energy that goes with it is growing right alongside it. Nobody, she pointed out, actually has a clean number for how much energy data centers use today, and the range of predictions for 2030 varies by a factor of four depending on who you ask. That uncertainty has real, local consequences. She walked us through a Reuters story from just days before her talk: pollution from a Tennessee data center concentrated in predominantly Black neighborhoods already carrying disproportionate environmental burdens. The gas turbines many data centers use to generate power on-site are technically classified as mobile equipment, which means a lot of environmental protection law simply doesn't apply to them in the same way it would to a permanent facility. In Virginia, the buildout of data centers is already driving electricity prices up for residents who aren't using any more power than before, they're just absorbing the cost of infrastructure built for someone else's workload.
Environmental justice, she made clear, isn't an abstraction layered onto AI. It's already showing up tangibly in specific zip codes.
Whose labor, whose livelihood
One reframe that stuck with a lot of us: when we talk about AI, automation, saving money, or time, we tend to only count what's countable. Dr. Luccioni used the example of agricultural robots, widely marketed as saving farmers time and money. What that framing leaves out is the labor of the workers those robots replace, people who often don't show up in official labor statistics at all. The investment shifts from people to the companies building the machines, and the "savings" we measure are the ones easiest to put a dollar figure on, not the ones that show up in a person's livelihood.
It's a pattern that shows up again in how she talked about the foundations AI is built on. As she put it plainly: it's our data, it's our dollars, it's our tokens. The people whose work trained these systems, and the people bearing the downstream costs of running them, are rarely the ones the benefits accrue to.
Time is not a savings account
We've all heard some version of "AI will give you your time back." Dr. Luccioni pushed on that. Time is finite, so if AI saves you an hour, that hour goes somewhere, and research shows it often goes toward more consumption, not less, more travel, more purchases, more resource use. She also pointed to an MIT study showing that people who rely heavily on tools like ChatGPT for cognitive tasks show measurably different brain activity patterns over time than people who work through problems themselves. And she was candid about where AI has actually generated the most value so far: not optimizing traffic flow or transforming logistics at scale, but targeted advertising, which remains the dominant revenue engine funding the biggest AI buildouts.
She also took apart a widely circulated study that compared the carbon footprint of AI-generated writing and images to human authors and illustrators, concluding AI produced hundreds of times less CO2. Her critique was sharp and, we thought, important: human beings are more than just what they do for a living, and the comparison ignored the cost of training the models in the first place, plus the simple fact that AI could not exist without the human-created work it was trained on. You can't build the tool from human output and then use the tool to justify replacing the humans.
What good actually looks like
The talk wasn't all critique. I asked Dr Luccioni to give us both insight and some inspiration. Is there any hope for sustainability given AI’s impact on energy resources? Dr. Luccioni spent real time on what companies like ours can actually do.
Measurement comes first. She's helped build open tools like CodeCarbon and the AI Energy Score project specifically so organizations aren't flying blind on which models are efficient and which aren't, sometimes the gap between two models doing the identical task is 100x in energy use. Model choice matters too: a simple task doesn't need a frontier reasoning model, and increasingly, smaller models that run locally on a laptop or phone are catching up in capability for everyday tasks, without ever touching a data center.
The most useful reframe for us, as a consultancy that regularly advises clients on exactly this kind of decision, was transparency as a procurement question. We already ask vendors detailed questions about data security and compliance. Dr. Luccioni's point: we should be asking the same rigor of questions about the AI we deploy, where it runs, how much energy it uses, how that energy is generated, the same way we'd ask about recycled content in office paper. And when a client's motivation isn't primarily environmental, she offered a reframe that lands regardless of where someone sits politically: think about it as de-risking AI. Not knowing which data center your vendor's model runs in, or what regulatory or geopolitical exposure comes with it, is itself a business risk worth asking about.
She closed with a definition of sustainability that reframed the whole talk: true sustainability requires three pillars together, environmental, economic, and social, not one traded off for the others. An AI system that's efficient but built on inequitable data isn't actually sustainable. Neither is one that's socially fair but wildly wasteful. The goal is the intersection of all three.
Why this mattered enough to open the doors
This is the third quarter we've built our all-company education arc around the intersection of a justice pillar and the technology shaping our industry. And each time, the goal has been the same: learn something uncomfortable, together, in public, rather than alone and quietly. This might have been the sharpest example yet of why that ritual matters. As TXI moves further into building intelligent products for clients in healthcare, manufacturing, and industrial sectors, the environmental and social costs of the systems we recommend are not somebody else's problem to solve later. They're a design constraint now, the same as security or performance.
That's also why we brought other companies into the room. The tradeoffs Dr. Luccioni described, energy versus water in data center cooling, efficiency versus capability in model choice, whose labor gets displaced and whose doesn't, aren't unique to TXI. Every company building or buying AI is making these calls, often without asking the questions that would surface them. If more of us learn this together, out loud, we get better at asking the right questions before we build, not after.
What I'm still thinking about
The line that's stayed with me most is her answer to the "it's already too late" fatalism she says she hears constantly: it's our data, our dollars, our tokens. That's not a guilt trip, it's actually the opposite. It means we have more agency in this than the doom narrative suggests, in which model we pick, which questions we ask a vendor, whether we default to the biggest model or the right-sized one.
We're a mid-sized, employee-owned consultancy building intelligent products for clients who run enterprises, factories, and supply chains. We don't control how the frontier labs build their data centers. But we do control the questions we ask before we recommend a model, whether we default to measuring impact or just assuming it's negligible, and whether "intelligent" for us means fast and clever, or fast, clever, and honest about its cost.
That's the standard we're trying to hold ourselves to as we keep building.
Want to go deeper?
