What Climate-Tech Looks Like After the Prototype

Going into my first conversation with someone building in the carbon market, I thought I broadly understood the problem. Nature-based projects measure what happens on the ground, use tools like satellite imagery to check it, get the credits certified, and sell them.

What I had underestimated was everything that happens around the technology.

One example stayed with me. I was told about a corporate buyer that continued conducting due diligence for nearly six months even after the project had gone through certification. The buyer examined the application, underlying datasets, and supporting evidence before becoming comfortable with the credits.

I had assumed certification was where uncertainty largely ended. Instead, I began to see climate technology as a much larger system, one where measurement is only the beginning.

Trust has to be designed into the process

Carbon markets have an unusual problem: the final product is essentially a claim about something that happened somewhere else.

A buyer needs to believe the intervention happened, that it produced the effect claimed, and that the evidence behind is credible. Certification provides an important independent layer of validation, but it may not answer every question a buyer has.

That changed how I think about what a climate-tech product has to produce.

It is not enough for a system to spit out an answer. It should also be possible for someone unfamiliar with the project to understand how that answer was reached: where the information came from, what evidence supports it, and what assumptions sit underneath.

In other words, the output is not only a number. The chain of evidence behind the number can be part of the product itself.

Better technology does not mean pretending to see everything

I liked the idea of satellite-based MRV because it looked scalable. In theory, one satellite system could watch thousands of farms without putting someone on every field.

Different agricultural interventions do not leave the same kind of signal, though.

Satellite imagery may help identify trees, water bodies, buildings, crop characteristics, or changes in vegetation over time. It cannot necessarily tell whether a farmer used direct-seeded rice or followed a particular irrigation practice such as alternate wetting and drying.

Those activities may require photographs, videos, documents, declarations, field visits, or other ground-level evidence.

That sounds obvious once you say it. It still undid one of my assumptions. The job is probably not to find one technology that verifies everything. It is to be precise about what each technology can actually establish, then build the rest of the evidence system around the gaps.

Admitting what a technology cannot observe is not a weakness. In a market built around credibility, it may actually make the system more trustworthy.

Economics can decide whether good technology gets used

Another point had almost nothing to do with sensors or software.

Registration or licensing for a carbon project can cost upwards of ₹15 lakh on its own, before validation and verification costs stack on top. For a small project, those fixed costs bite hard. For a project spread across a much larger area, the same costs can be distributed across more acres and more credits.

That changes the startup problem.

I normally think of scale as something that comes after a prototype works: build something small, prove it, and expand. But in this market, sufficient scale may sometimes be necessary for the economics to work in the first place.

That left me with a useful distinction. Alongside a minimum viable product, climate-tech builders may also need a minimum viable deployment.

Who has to participate? Over how much land? At what verification cost? How many credits do you need before the whole thing makes economic sense?

You cannot really separate the technology from the business model here.

AI may be most useful between the pieces

Perhaps the most interesting opportunity I came away thinking about was not using AI to magically “measure carbon.”

Nature-based projects can generate satellite images, photographs, videos, land records, declarations and other documents. Someone still has to turn that pile into a coherent account of what happened.

That points to a different job for AI.

Not as an unquestionable judge. More as a system that helps organise evidence, trace information back to its source, flag missing or contradictory records, check whether supporting material is relevant, and get the material ready for human reviewers, auditors, or buyers.

That distinction matters.

In a trust-sensitive industry, swapping one hard-to-inspect process for an opaque AI score just creates a new trust problem. Helping people inspect the evidence faster is probably the more useful move.

It changed the question I was asking — from “How can technology automate carbon verification?” to “Where does verification currently burn the most unnecessary human effort?”

What's next

This is just the beginning. I have a clearer frame for the conversations I want next. I want to know where evidence actually breaks. What do buyers push on most often? Which parts of verification eat the most time? Which agricultural interventions can you realistically observe from space? Where do project economics shut out smaller farmers or developers? And where can software cut friction without cutting accountability?

There is a larger question I want to keep coming back to as well. If farmers are the ones changing practices and carrying a lot of the burden, how should technology and carbon markets make sure value reaches them?

My biggest shift in thinking is fairly simple.

I started by looking for better ways to measure carbon.
Now, they are the systems that make climate claims credible enough to trust, economical enough to deploy, and transparent enough to question.
If you work in climate markets, carbon credits, MRV, or have thoughts on any of the questions above, I’d love to hear from you. I’m always open to conversations with people working on these problems.