AI and the Last Mile

Everyone is excited about AI generating documents. Ask ChatGPT to write a contract, a proposal, a report — and it does a decent job. The first draft appears in seconds. It feels like magic.

But then what?

You need to format it correctly. You need to get it into PDF. You need to route it to the right people for review. You need to collect signatures. You need to track who's signed and who hasn't. You need to store the final version somewhere. You need to make it searchable later.

This is the last mile of document work. And AI is terrible at it.

The 10/90 problem

Here's a pattern that shows up everywhere in technology: the first 90% of a task is solved by some impressive new capability, but the last 10% — the boring, detail-oriented, operational part — takes 90% of the effort.

AI-generated text is the 90% that feels like 10%. The actual delivery pipeline — formatting, compliance, routing, signing, archiving — is the 10% that takes 90% of the effort in any real business process.

Think about what happens when a sales team needs to send a contract. The AI can draft the contract in 30 seconds. Then a human spends the next three hours getting it into the right template, adding the correct legal clauses for that jurisdiction, routing it through legal review, setting up the signature fields, sending it to the client, following up when they don't sign, and filing the executed copy.

The drafting was the easy part. It was always the easy part, even before AI.

Why the last mile is hard

The last mile of document processing is hard for a specific reason: it involves interacting with the real world. Real people with real email addresses who forget to check their inbox. Real legal requirements that vary by country and industry. Real file formats that need to work on every device. Real audit trails that regulators might examine.

AI operates in the world of text. It's very good at manipulating strings of characters. But the last mile isn't about text — it's about process. It's about ensuring that a specific document reaches a specific person, they verify their identity, they review the correct version, they sign in a legally binding way, and the signed copy is stored in compliance with the relevant regulations.

None of this is glamorous. But all of it is essential.

The infrastructure layer

The companies that will capture the most value from AI in documents aren't the ones building better language models. They're the ones building better infrastructure for the last mile.

This is counterintuitive. When a new technology appears, everyone assumes the value will accrue to the technology itself. But with electricity, the value didn't go to power plants — it went to appliances and factories. With the internet, the value didn't go to TCP/IP — it went to applications built on top of it.

With AI, the value in document processing will go to the platforms that take AI-generated content and handle everything that happens next. The formatting. The workflows. The signatures. The compliance. The storage. The search.

What good infrastructure looks like

Good document infrastructure is invisible. When it works, nobody notices. You draft a document, click send, and it flows through the entire process without you thinking about the mechanics.

Bad document infrastructure is painfully visible. You've experienced it. The PDF that won't open on your phone. The signature tool that requires you to create an account. The contract that gets lost in someone's email. The file you can't find three months later.

The gap between good and bad infrastructure is worth billions of dollars in lost productivity. McKinsey estimates that knowledge workers spend 19% of their time searching for and gathering information. Most of that time is wasted in the last mile — not in creating documents, but in processing them.

AI makes the last mile more important, not less

Here's the thing that most AI enthusiasts miss: as AI makes document creation faster, the bottleneck shifts entirely to the last mile. If you can generate a first draft in 30 seconds instead of 3 hours, the formatting-signing-delivery pipeline becomes 99% of the total time instead of 50%.

This means the companies that solve the last mile will become more valuable as AI improves, not less. Every improvement in AI document generation increases the pressure on the delivery infrastructure.

The future of documents isn't AI that writes them. It's infrastructure that moves them. The writing was never the hard part.