The Spreadsheet

The guy next to me in the cafeteria was staring at his phone like it owed him money.

"What are you looking at?" I asked.

He tilted the screen toward me. A spreadsheet — hundreds of rows, tiny numbers, color-coded cells bleeding into each other. Utility bills. Fuel logs. Supplier invoices.

"Carbon audit," he said. "Been on this for three months."

"Three months on a spreadsheet?"

"Copy a number. Look up a factor. Multiply. Flag it. Repeat." He shrugged like a man who had made peace with his sentence. "Maybe two more months to go."

I didn't say anything. I just watched him scroll.

That afternoon I couldn't focus. I kept seeing those rows. So I started digging.

I read the GHG Protocol spec cover to cover. Then the GRI standards. Then the SEC climate disclosure filings. The scale of the problem got clearer with every page — but reading about it wasn't enough. I needed to talk to the people actually living it.

So I did. I reached out to ESG consultants, sustainability managers, and carbon auditing startups. I asked them to walk me through their workflows, step by step, without cleaning anything up for me.

What I heard was consistent, and consistently exhausting. Data arrived in every format imaginable — scanned receipts, email attachments, decade-old Excel exports. Someone had to touch every single one. Map it to a scope. Find the right emission factor for that fuel type in that region. Check it. Flag it. Get it reviewed.

"We use AI a little," one person told me. They were a founder. Smart, well-funded, deep in the problem. "Mostly to extract text from PDFs. Maybe summarize a document."

"But end-to-end?" I asked. "Data in, auditable report out?"

He shook his head. "Nothing like that exists."

I heard that same answer, in different words, from everyone I talked to. They weren't ignoring AI — they were using it. But they were using it the way you'd use a better highlighter. A faster copy-paste. The workflow was still fundamentally human. Someone still had to sit in the middle of it, moving information from one place to another, making sure nothing fell through the cracks.

No one had built the thing that removed that person from the loop entirely.

That was the gap. And once I saw it, I couldn't unsee it.

Six months later, I pointed SustainForge at a folder of invoices and watched five agents do in four minutes what used to take four months.

I thought about texting the cafeteria guy. I still might.

Built With

  • agentic
  • ai
  • langchain
  • langgraph
  • mcp
  • next
  • python
  • uvicorn
Share this project:

Updates