Superpower or Superwaste: What AI lacks is a mission

I already knew it before AI, but in privileged societies Maslow’s hierarchy has been hollowed out and filled with five-second clips.

What is it all for? I’ve trying to work out if it’s just the creeping cynicism of someone heading towards mid-life motorbike purchases or something more empirical. Yes, I’m in my forties. Yes, I have a child — sleep-loss could be cynicism-contributing. But they will inherit whatever we decide to build or burn, just as we have from “the boomers”. Yes, I’ve spent twenty years making software products. Indeed, I’ve fought through rooms of stakeholders, legal, accessibility, procurement, and the particular hell of a roadmap that has half a dozen “owners” and no clear north star — all in service of shipping something that actually worked, that people actually needed, that made some small corner of the world function better. And, YES, also, because it paid to do it. Maybe the difference was naivety? Or desperation…

And now I open a chat window, type a few sentences, and something that would have taken my team three months, or more with those damn stakeholders, assembles itself in thirty seconds.

It should feel like a superpower. Mostly it feels like a question I can’t stop asking.

Is this all it’s for?

Sending Superman to do Laundry

The candid answer, if I look at my own chat history, is semi-embarrassing.

I’ve used AI to write docs I was too tired to write (and everyone was too tired to read). To summarise meetings I was barely present for. To prototype a “fun” app I had no real intention of finishing so I could post it somewhere and feel briefly like I was still a person who ships things. Vibe coding, they call it. Click yes. Nudge the prompt. Click yes again. Watch something that increasingly works emerge from nothing. A small dopamine hit dressed up as productivity. The like button of B2B. The Big-Mac of productivity. For what?

I don’t think I’m alone in this, this… question. This uneasiness. This thirst for something bigger here.

The most sophisticated reasoning tool in human history, and I used it last Tuesday to help me decide how to cook barbecue. (Dry brining, for the record. Worth it.)

To be absolutely clear: this is not an argument against efficiency and technology (it is my career). I am not a luddite (the opposite). This is not a defence of “craft” or a reaction to feeling replaceable (I do). Those feelings are fine to have. This is something simpler and yet harder to pin down.

It’s an argument about having a superpower in my pocket and feeling, most days, like I am wasting it. I have sent Superman to do my f***ing laundry.

AI on a {first_name} Basis

Meanwhile, inboxes fill up with outreach that thinks using {first_name} is intimacy. Social media encourages me to “generate a post” — sure, let’s go with My agent just raised my child, here are 3 things it taught me about B2B sales progression #grateful. The six-fingered children of badly proofed display ads smile blankly from every sidebar (what a wonderful 6 months that was). I ask a customer support bot a technical question, it deflects and I ask it how to do barbecue better (still dry brine). The most common deployment of the most powerful technology since the internet is: helping companies sell things to people who don’t want them, slightly more efficiently.

Is this where the bottle landed? Truth or dare, but the only option is #HustleCulture.

Nobody intended it, just like nobody really intends to eat too much barbecue. But it was there and umami. “This is the predictable result of transformational technology deployed inside existing economic incentives”, says the banal tech strategist in me. I hate that guy. “Optimise for growth and then eat the harder problems”. SHIT, is he right? I mean, of course the first wave of AI applications optimises for revenue. Of course it gets aimed at marketing and sales pipelines. That’s where the money is fastest. That’s how venture math works. That is how capitalism works mate. It is the profit motive.

That doesn’t make it less of a waste of Superman to check the washer dryer.

We can do better with AI

Until six weeks ago, the furthest any human being had physically travelled from Earth was 400,171 kilometres. Apollo 13. A near-disaster in 1970, where the record was set accidentally, as a side effect of bringing three men home alive. We didn’t beat it for fifty-six years. Then, on the 6th of April 2026, the Artemis II crew pushed past it — reaching 406,771 kilometres from Earth, the far side of the Moon, views no human eye had ever taken in directly.¹

I find it genuinely extraordinary and a sobering fact to hold in my head: the entire frontier of human physical presence in the universe currently fits inside a number that a frequent flyer counter could theoretically reach. We are, cosmically speaking, stepping out of our front door, looking terrified and running back in for 5 decades.

Now hold that thought, and add this spiky bastards into the melting pot.

We face a climate crisis with a narrowing window. The International Energy Agency’s Net Zero by 2050 roadmap is explicit: we needed to stop approving new fossil fuel development after 2021.² We didn’t. Antimicrobial resistance kills more than a million people a year directly — 1.27 million in 2019 alone, according to the most comprehensive study ever conducted on it, published in The Lancet in 2022 — and the pipeline for new antibiotics is nearly empty, with projections suggesting 39 million deaths attributable to AMR between 2025 and 2050 if current trends hold.³ In the global south, people still die of diseases we know how to cure. Suicide is the third leading cause of death among people aged 15 to 29 worldwide — and in parts of western Europe and high-income Asia Pacific, it’s the leading cause, full stop.⁴⁵

These are not small problems made to sound large. They are exactly as serious as they appear. Hiding in our house won’t help. Speaking personally, they TERRIFY me. But at least I can simplify my Linkedin posts in one click. Why am I allowed to do that?

And we have, for the first time, a general-purpose reasoning tool that can accelerate scientific discovery, model complex systems, and find patterns in data at a scale no human team could match.

The question isn’t whether the tool is powerful enough (it is getting there). The question is what we’re pointing it at. How much of it we are pointing. PLEASE CAN WE POINT AS MUCH AS POSSIBLE AT THE WORLD ENDING PROBLEMS.

Last year, Google DeepMind’s GNoME model identified 2.2 million theoretical crystal structures — roughly 45 times the total number science had found in its entire prior history, according to a paper published in Nature in November 2023.⁶ The headline is contested: critics, including researchers at UC Santa Barbara, have argued that many of those structures are variants of already-known compounds rather than genuinely novel materials, and as of late 2025 the paper faced calls for retraction over concerns about duplicates. DeepMind disputes this. The actual number of meaningfully new, synthesisable candidates is smaller than the headlines suggested.

Even so. Even with the caveats, the tool expanded the frontier. Of the 2.2 million predictions, around 380,000 are assessed as the most stable — promising candidates for next-generation batteries, solar cells and superconductors. External researchers around the world had already independently synthesised 736 of these new structures in concurrent work. That’s not optimisation. That’s new territory, however large or small the map turns out to be.

And GNoME is one example, from one lab, pointed at one corner of the problem. We need more. All of them please.

Some Better News

It is not all existential terror and wastefulness. There are angles we can push, that I would love to help push too:

Steel produces somewhere between 7 and 8 percent of global greenhouse gas emissions annually. The vast majority comes from burning metallurgical coal to reach the temperatures needed to smelt iron ore. The known alternative — hydrogen-based direct reduction — works. It’s used at scale in Sweden by a consortium called HYBRIT, producing what they call “fossil-free steel,” and the technology is proven. The problem is speed and capital. AI doesn’t replace the hydrogen. But it optimises the existing plants right now — extracting more from the same energy input, identifying inefficiencies invisible to human operators, buying time while the infrastructure scales.

Cement is, if anything, worse. Every tonne of clinker — the binding ingredient in concrete — releases roughly a tonne of CO₂, and about half of that is a direct chemical consequence of the production process itself, not the fuel. You can’t eliminate it just by switching energy source. Cement and concrete together account for around 8 percent of global emissions. A UK company called Carbon Re deployed its AI operating system at a Heidelberg Materials plant in the Czech Republic in 2024. In an initial evaluation, it delivered a 4.1 percent cut in fuel costs and a 2.2 percent reduction in specific heat consumption. That doesn’t sound dramatic. But Carbon Re’s co-founder estimates that each deployment reduces emissions by around 10,000 tonnes of CO₂ per plant per year. Scale that across the thousands of cement plants operating globally, and it becomes a meaningful number — not a solution, but a meaningful, deployable intervention available now.⁷

Shipping moves around 90 percent of global trade and accounts for roughly 3 percent of global greenhouse gas emissions — a figure the International Maritime Organization projects could rise to 10 percent by 2050 if unchecked. The fuel these ships burn is among the dirtiest in use anywhere. Route optimisation is an obvious target. A management company called Anglo-Eastern ran AI-driven performance systems across 800 vessels and nearly 46,000 voyages between 2023 and 2025. Result: over 700,000 metric tonnes of CO₂ avoided, 225,000 tonnes of fuel saved, and approximately $135 million in cost reductions.⁸ Those numbers came from a single fleet operator, not from a global programme. The technology exists. The deployment is the constraint.

The grid is the connective tissue that everything else depends on. Renewable energy is now growing faster than at any point in history — 582 gigawatts of new capacity added in 2024, the strongest annual gain on record — but it doesn’t dispatch on demand. Wind and solar are intermittent. The harder the grid leans on them, the more it needs intelligence to balance supply against consumption in real time. DeepMind applied machine learning to predict wind power output 36 hours in advance across 700 megawatts of capacity, increasing the value of that energy by around 20 percent through smarter grid scheduling. Open Climate Fix, a non-profit, is working with a state grid operator in India — a country with an ambition of 450 gigawatts of renewable capacity by 2030 — to bring the same kind of forecasting to solar. Early results show a 10 percent reduction in large forecasting errors across a 24-48 hour horizon.⁹ In grid terms, that matters.

Agriculture produces somewhere between 20 and 25 percent of global greenhouse gas emissions, mostly through methane from livestock, nitrous oxide from fertilised soils — a gas with 265 times the warming power of CO₂ — and land-use change. Precision agriculture, guided by AI analysis of satellite and sensor data, reduces the amount of fertiliser applied per unit of yield, which directly cuts nitrous oxide emissions. A peer-reviewed study published in Smart Agricultural Technology in 2024 found that AI-driven tools can increase crop yields by 15-20 percent while reducing required investment by 25-30 percent.¹⁰ A Nature Sustainability study from 2025 found that AI-driven precision systems outperformed conventional farming on both fertiliser use and carbon emissions by adapting to field-specific conditions in real time.¹¹

Chemical industry. Less visible, more pervasive. The production of plastics, fertilisers, pharmaceuticals and industrial chemicals accounts for around 6 percent of global emissions, and the feedstocks are almost entirely fossil-derived. AI is beginning to accelerate the discovery of bio-based alternatives — materials derived from plant or microbial processes rather than petroleum — and to identify substitution pathways through the enormous chemical space that human researchers, working conventionally, couldn’t survey in multiple lifetimes. This is slower, less legible work than optimising a cement kiln or a shipping route. But it’s where the deeper transformation happens.

The collage I am trying to create for you is nuanced. AI, heavily backed, partially wasted. The challenges, large and diverse. The specific solutions, progressing, under-resourced. The math seems obvious to me in these terms.

I Am the Problem

I should acknowledge something here (and in everything I write). I drink oat flat whites without shame. I have an air fryer. I live in a comfortable apartment in Berlin and spend my evenings reading about climate science while my child (sometimes) sleeps. The distance between my daily material reality and the problems I’m describing is not something I’m unaware of. There’s a particular flavour of comfortable, liberal guilt that produces essays exactly like this one, and I want to be honest about that proximity.

But I don’t think the right response to this privilege is silence. I think it is action. I can, I should, I will, do more. The banal strategist in me sees opportunity in the threat. The entrepreneur sees a mission. Twenty years inside the industry that built these tools. I see a profound mismatch between capability and ambition. Not malice. Not conspiracy. Just the familiar human tendency to aim the new thing at the nearest revenue opportunity and call it progress.

What’s missing is not the tools. Not the people. Not even, really, the technology.

It’s the mission. The declaration. The moment where someone with credibility and reach says: this is what “ALL THIS” is for.

This is What AI is For

We got to the Moon the first time because someone declared it. Not because the market demanded it. Not because the unit economics made sense. Artemis II proves we can still do that, when we decide to. Four people, a spacecraft named Integrity, the far side of the Moon for the first time in human history. It happened because people chose to make it happen. We can reduce CFC emissions, when decide to do it, together. So, let us choose.

I’ve spent two decades making products that made other people’s businesses more efficient. I’m proud of some of it. Some of it, honestly, I could take or leave. What I know is that I’m not interested in spending the next twenty years doing the same — not when the tools available now make the gap between “product that optimises a sales pipeline” and “product that bends a carbon curve” smaller than it has ever been.

Maybe it always comes down to that. AI is not going to dissolve capitalism or cure cultural short-termism. Maybe we are, in some fundamental sense, doomed to spend our best compute on financial derivatives and engagement optimisation. But maybe it also gives us a fulcrum. A moment where the gap between what’s possible and what we’re doing is visible enough that some people — the right people — decide to close it. So… I guess this is what I want it to be for.

References

NASA, “Artemis II Crew Eclipses Record for Farthest Human Spaceflight” (6 Apr 2026) — the crew reached 406,771 km from Earth on the Moon’s far side, breaking Apollo 13’s 1970 record. Source: NASA. ↩
IEA, Net Zero by 2050: A Roadmap for the Global Energy Sector (2021) — its central finding: no new oil and gas fields or coal mines approved beyond 2021 on a 1.5 °C pathway. Source: IEA. ↩
GBD 2021 Antimicrobial Resistance Collaborators, “Global burden of bacterial antimicrobial resistance 1990–2021: a systematic analysis with forecasts to 2050,” The Lancet (2024) — ~1.14M deaths directly attributable to bacterial AMR in 2021, with a projected 39M cumulative deaths 2025–2050. Source: The Lancet. ↩
WHO, Global Health Estimates 2021 — Leading causes of death — the WHO dataset behind the essay’s cause-of-death rankings by age group. Source: WHO. ↩
GBD 2021 Suicide Collaborators, “Global, regional, and national burden of suicide, 1990–2021,” The Lancet Public Health (Feb 2025) — finds suicide the third-leading cause of death among people aged 15–29 (~727,000 suicide deaths in 2021). Source: The Lancet Public Health. ↩
Merchant, A. et al. (Google DeepMind), “Scaling deep learning for materials discovery,” Nature 624 (Nov 2023) — GNoME identified 2.2 million candidate crystal structures, ~381,000 of them predicted stable. Source: Nature. ↩
UNIDO, “How AI is shaping decarbonization pathways in heavy industry” (Dec 2025) — profiles Carbon Re’s cement-kiln AI, whose co-founder estimates ~10,000 tonnes of CO₂ saved per plant per year. Source: UNIDO. ↩
Anglo-Eastern, SAPS performance platform (2023–2025) — across 800+ vessels and ~46,000 voyages, reported 700,000+ tonnes of CO₂ avoided, 225,000 tonnes of fuel saved and ~$135M in cost reductions. Source: Anglo-Eastern (via Yahoo Finance). ↩
Open Climate Fix with Google DeepMind — AI solar/wind forecasting for grid operators, including a project with an Indian operator (India targets 450 GW of renewables by 2030). Source: Open Climate Fix. ↩
“Adaption of smart applications in agriculture to enhance production,” Smart Agricultural Technology (2024) — reports AI-guided tools improving crop yield by 15–20% while reducing required investment by 25–30%. Source: ScienceDirect. ↩
Study in Nature Sustainability (2025) — AI-driven precision-agriculture systems outperform conventional farming on both fertiliser use and carbon emissions by adapting to field-specific conditions in real time. Source: Nature Sustainability. ↩