Inside the rebuild: how we're rethinking carbon accounting for the AI era

Last month, Impact Loop↗ wrote about a decision we made at Atmoz: we took a working carbon accounting platform, with customers flowing in, and rebuilt everything from ground zero.

The coverage told the what. This series is the reasoning, from inside the build. Over the coming weeks I will walk you through how a decade-old climate data company rethought its product from first principles, what that looks like day to day, and who is doing it. This first part is about the decision itself.

The uncomfortable question

Our first platform did its job well. It helped hundreds of companies turn their accounting data into climate data, and it was growing when we paused it. That matters, because this is not a story about fixing something broken. It is a story about noticing that the ground had moved.

Two things changed at once.

Our customers changed. They got larger, more complex, and they stopped asking for a report. They started asking for numbers they could defend: to an auditor, to a regulator, to a procurement team, to an investor whose job is to doubt them. Climate data was leaving the sustainability report and entering financial governance, and the standard of proof travels with it.

And AI changed what was possible. The hardest problem in climate data has always been the same one: reliable, detailed activity data is expensive and labour intensive to produce. For years, the whole industry worked around that constraint with templates and generic factors. Suddenly the constraint itself was negotiable.

So we asked the uncomfortable question. Not "how do we add AI to the product?" but "if we started from what must be true, would we build this product?"

Back to first principles

Strip carbon accounting down to what must be true and you get a short chain of reasoning.

A company's footprint comes from its activities. Its activities leave a trail of transactions. The invoice is the closest thing to proof of what actually happened.

And in the era our customers are entering, every number built on that proof must hold three properties:

• Immutability. Source data is never destroyed.
• Traceability. Every transformation is traceable.
• Recomputability. Every number can be recomputed from its origins.

Those are the principles we publish. The full list stays in the building.

Hold the old platform against those principles and the conclusion is honest, not dramatic: these properties cannot be retrofitted. Immutability, traceability and recomputability are not features you add in a release. They are the foundation, and a foundation is the one part of a building you cannot renovate.

That is why the decision, which looks brave from the outside, felt from the inside like the only honest conclusion. We knew exactly what the next generation of this product had to be, because we had spent a decade with the people who will have to defend its numbers. Keeping the old foundation would have meant knowingly building the future on principles we no longer believed in.

Where AI actually enters the story

AI was not the reason to rebuild. AI changed what was possible. First principles decided what to build. The order matters.

It matters twice, in fact. AI is in the product: agents that collect, classify, analyse and quality-assure emissions data, with every output traceable back to its source. And AI is in the building of the product: inside every loop of design, engineering and validation.

That second part is what changed the economics of the decision. A ground-zero rebuild used to be a multi-year gamble that killed companies. We paused development in autumn 2025, soft launched the new platform in summer 2026, and are onboarding customers. The cost of starting over collapsed, and that collapse is what turned a correct-but-impossible decision into a correct-and-feasible one.

First principles is the why. AI is the how.

What comes next

In this series I will show you how that actually works: the way our analysts and engineers build as one unit, the architecture that survived 165 variations of the product without flinching, how an invoice becomes a number an auditor can interrogate, and, at the end of it, the team doing it, between Colombo and Stockholm.

We have been in climate data for a decade. We have never been more sure of what we are building, or had more fun building it.

Next: we ship every Friday.