State of Sustainability Podcast

Solo Episode: AI, Agentic AI, and What It Means for Sustainability Teams

SAIF: Welcome back to another episode of the State of Sustainability. I'm your host Saif Hamid, founder and CEO of Altruistiq.

This episode is about AI — but a more evolved version of the conversation than we've had before. Longtime listeners may remember our episode on centaurs and cyborgs, which is already close to a year old. That episode covered two modes of human-AI collaboration: the cyborg model, where a human and AI iterate together on the same task; and the centaur model, where the work is split — AI does its piece, the human does theirs, and the outputs are combined. Those frameworks came from a Harvard Business School and BCG study of consulting teams that found all teams using AI outperformed those that didn't, and that the teams using AI adopted one of these two approaches.

I mention that partly to recap, and partly to say: that framing already feels out of date. The space has moved fast.

At Altruistiq, we've been on this journey for at least a year and a half. First phase: getting AI to run most of our product engineering — AI writing 80% or more of our code. Second phase: embedding AI into our product and platform for customers. Third phase: running agents across our business operations more broadly. Based on what we've been through, I want to spend this episode on three things I see as distinct but related: AI, agentic AI, and AI native services. And I want to talk through what each of these means specifically for sustainability teams.

AI vs Agentic AI vs AI Native Services

AI in the workplace today mostly means using a chat-based product — ChatGPT, Claude, Copilot — to help with everyday tasks. This is AI in its most accessible form, and most people using it are effectively using an LLM-powered chat interface. Quick note on LLMs: large language models are trained on large volumes of text and predict the next most likely word in a sequence. They're dominant right now, but they have real weaknesses — hallucinations, sycophancy, high compute costs — and it's worth remembering that in three years we may have moved to a different dominant approach entirely. For now, we run with LLMs.

Agentic AI is the next step. An agent is what you get when you take an AI and connect it to tools and context. Give an LLM access to your email, and it can read, summarise, draft, and send messages. Give it access to data feeds, a Slack workspace, a supplier database, and it can do meaningful operational work autonomously. The "context" is the fixed background information the AI needs — instructions, standing rules, organisational knowledge — that shapes how it behaves. Put tools and context together and you have an agent that can act on your behalf to deliver outcomes. Most large businesses have pockets of experimentation with agents — usually in IT or digital transformation teams — but it hasn't entered the mainstream of most business functions yet. Sustainability teams are at least as early as their organisations on this, and in many cases earlier.

AI native services is even newer, and most of our customers haven't encountered it yet — but they will within months. The concept is a hybrid delivery model: a solution provider uses a combination of human experts and agents to deliver defined business outcomes on your behalf. The key distinction from traditional consultancy is the commercial model. Consultancies sell capacity — a manager and two analysts for four weeks, billed at a day rate. AI native services providers sell outcomes. They commit to delivering three specific things and own the results. They don't want to lock in capacity because they're automating a large portion of the work internally, so a capacity commitment would make no economic sense for them.

A concrete example: imagine a supply chain manager responsible for three outcomes — ongoing environmental and human rights risk monitoring across suppliers, data acquisition for a corporate carbon footprint, and environmental diligence on newly onboarded suppliers. In an AI native services model, a provider takes responsibility for all three. The risk monitoring might be fully agentic: an AI connected to satellite data feeds, news feeds, and supplier databases flags exceptions and drops alerts into your inbox or Slack. A human somewhere monitors quality but doesn't touch individual cases unless something escalates. The data acquisition has a human making the initial supplier call, with all validation and verification done by agents depositing clean data into your carbon accounting system. The onboarding diligence involves agents scrutinising supplier data, other agents reviewing those agents' outputs for quality, and a human coming in only if the supplier wants to contest their assessment. That's AI native services: a human at the centre of an ecosystem of agents, delivering outcomes that previously required a full team.

What This Means for Sustainability Teams

I want to be honest about what I'm seeing right now. There is significant haemorrhaging in sustainability functions. Multiple CSOs within my network have been removed from their roles in recent months. The Chief Sustainability Officer as a C-suite role is probably not durable in its current form — I've been saying this on this podcast for a while, and it's now visibly happening. The role is being downgraded in organisational hierarchy, and sustainability team headcounts are taking meaningful hits.

AI and agentic AI will accelerate this. Here's why.

Most organisations are under pressure to demonstrate AI value, and that value almost always translates to labour savings. AI is artificial intelligence — by definition the only thing it substitutes for is human intelligence. The labour savings are going to show up first in work that is predictable, cyclical, repetitive, and requires a certain level of subject matter expertise but not creative or relational judgment. Sustainability work, right now, fits that description almost exactly. It happens on an annual cycle. It's structured. It requires knowledge but not invention. And it's out of favour culturally — which makes it an easier target for cost reduction conversations.

Most organisations will soon be choosing between two paths: building their own internal agents to handle sustainability workflows, or contracting with AI native service providers who deliver the same outcomes at lower cost through their own hybrid model. Either way, the implicit benchmark shifts. It's no longer about delivering 100% of the value at 100% of the cost. The new question is: can we get 80% of the value at 60% of the cost? For most organisations in the current environment, the answer to that question will be yes — and the decision will follow.

I say this with some personal difficulty. I know sustainability teams that are genuinely excellent — high bar, recognised within the industry as trailblazers. And I know that talented individuals within those teams are currently looking for jobs, not because their work is poor but because that level of quality is no longer what the organisation is prioritising.

What to Do About It: Three Levels

For a Chief Sustainability Officer. The answer is not to become a one-person show surrounded by agents. There are still capabilities that matter and that AI cannot replicate well — yet. Creative thinking, systems design, stakeholder relationships, strategic judgment. But the people you need for those things are probably different from the people you valued two or three years ago.

Two or three years ago, the most valuable person on a sustainability team was often the deep subject matter expert — the person who knew the arcane rules of FLAG accounting or could give you an instant and reliable read on where the GHG Protocol was heading. That expertise is being commoditised by AI, fast. The gap between your subject matter expert and what a well-prompted LLM can produce is narrowing. It's a bit like the famous finding that most drivers consider themselves above average — overconfidence in the uniqueness of human expertise relative to AI is probably not well-calibrated.

What you actually need now are people who can design and maintain systems — who understand what good looks like, how agents should plug into your processes, what context needs to exist and who maintains it, and how to get all the moving parts to work together. That's a different profile from the deep substantive expert of a few years ago.

For a Sustainability Manager. Your progression has historically been measured by the size of the team you manage. That metric is going in the wrong direction — organisations are moving toward fewer people, not more. The right question to ask yourself now is: how do I deliver the composite of outcomes I'm responsible for with fewer people and less resource? That means either leading your organisation's internal agentification of sustainability workflows, or helping it identify AI native service providers that can do it more cheaply from the outside. Either way, the role is shifting from managing a team of people to managing a set of outcomes — and the manager who understands that shift and gets ahead of it will be better positioned than the one who waits.

For a Sustainability Analyst. If the manager and analyst layers are going to collapse — and I think they are, for reasons I'll explain in a moment — then the question is: how do you make sure you're the one who comes out of that compression in the remaining seat? The answer is roughly the inverse of what I said to managers: show that you don't need the management layer above you. Show that the outputs you produce are already packaged in a way that makes decision-making easy for whoever is above you. Show that you can leverage agents to deliver what previously took a team. Don't wait for someone to give you the tools — find them, use them, and demonstrate leverage.

The reason the manager-analyst layer is under pressure is structural. Organisational hierarchies exist to do three things: manage the flow of information upward, channel decisions downward, and performance-manage the people doing the work. AI and agents make the first two significantly cheaper and easier — you no longer need multiple layers of gatekeepers to parse information into something a CEO can use, or to translate high-level objectives into operational tasks. If you don't need layers for information and decisions, the performance management function becomes derivative, and the argument for keeping those layers weakens.

A Cautionary Story

I want to close with a story from my consulting days at McKinsey. I was working with a large utility that had started a transformation at 12,000 employees and was down to 7,000 by the time I arrived. My job was to find where the next round of reductions would come from. I found a function of 1,200 people called the "change organisation." When I asked people what the change organisation did, almost no one could tell me.

After many conversations, I worked out what it was: a function of translators. The IT team — around 1,500 people — couldn't communicate with the rest of the business, and the rest of the business couldn't communicate with IT, so at some point someone had built a 1,200-person translation layer in between. At the time it was created, I'm sure it made complete sense. It probably felt like a smart and progressive move. But by the time I arrived, no one could even remember why it existed — and it was an obvious target.

I'm not saying sustainability is in exactly that position. But I would be wary of being a translator in a world that no longer knows what a translator does. The value of the sustainability function has historically been partly about bridging — between climate science and business language, between regulatory requirements and operational reality, between external pressure and internal decision-making. As AI absorbs more of that translation work, the question becomes: what remains that only you can do?

That's the question every sustainability professional should be working on right now.

I hope you enjoyed this episode of the State of Sustainability. If you did, please hit follow so you never miss another episode — and ratings, reviews, and shares are always very much appreciated.

‍