How AI Is Changing the Economics of Everyday Work

It's Not Just About Replacing Jobs

There's a lot of discussion about which jobs AI will replace. I think that's a reasonable concern, but it isn't the most useful starting point when deciding how to use AI in a business.

I find it more useful to look at what people actually spend their time doing. Most jobs involve a mixture of activities. Some require experience, judgment, or dealing with customers. Others involve gathering information, checking documents, classifying data, and repeating the same steps every day.

AI is already changing how some of that work gets done. And when enough individual tasks change, the job itself may change considerably.

What interests me is how that affects the decisions businesses make about their software, processes, and where they spend their money.

I've Seen This Before, Without AI

At dPivot, I worked on a reporting system for a client whose product managers spent nearly an entire day preparing and reviewing reports. More than 50 people were involved in that process.

We built a reporting solution using AWS Athena, S3, and Parquet that reduced the process to approximately ten minutes.

There was no AI involved. It was a conventional data engineering problem, and the solution was conventional software.

This illustrates something I've always found important. The value wasn't simply that the reporting application was faster. It was that the people using it no longer had to spend so much of their day preparing information. They could spend more time actually using it for their analysis.

AI creates opportunities to do something similar with work that has traditionally been harder to automate, particularly where the inputs are documents, images, emails, or unstructured information.

Where AI Starts to Make Sense

I've been exploring this through several prototypes using Amazon Bedrock. The use cases included compliance research, image assessment, contract analysis, and customer-feedback classification.

These are different problems, but they have something in common. Each involves work that someone would otherwise have to review, interpret, or organize manually.

For example, let's take a look at customer feedback. A business might receive hundreds or thousands of comments from customers. Reading every comment and sorting them into useful categories takes time. AI can help with that initial classification and make patterns easier to identify.

But generating categories isn't necessarily the same as producing useful information. You still need to know whether the categories are accurate, whether important complaints are being missed, and whether anyone will act on the results.

That's why I don't start by asking where we can add AI. I start by asking which parts of an existing process are expensive, slow, or difficult to scale.

Saving Time Isn't the Whole Business Case

One thing I've learned from building software for clients is that an improvement can be technically impressive and still not justify its cost.

Suppose AI can summarize a document in seconds instead of the fifteen minutes someone would normally spend reading it. That sounds like an obvious saving.

But what happens next? Does someone still need to read the entire document to check the summary? How often does the model miss something important? What happens when it gets something wrong?

If the person reviewing the output spends almost as much time checking it as they would have spent doing the original work, the benefit may be quite small.

The answers can be very different depending on the application. Misclassifying a piece of customer feedback might be inconvenient. Missing an important clause in a contract could be considerably more serious.

The technology might be similar, but the acceptable level of automation isn't.

Sometimes the Workflow Needs to Change

I think this is where some businesses may miss the opportunity. They introduce AI into an existing process but leave everything else exactly as it was.

If AI can reliably handle the straightforward cases, perhaps people should spend more time reviewing exceptions. If documents can be categorized as they arrive, perhaps there's no longer a reason to wait until the end of the week to process them.

On the other hand, adding AI can sometimes make a process more complicated. Now you have model costs, monitoring, exception handling, and another system to maintain.

I've seen enough software projects to know that adding technology isn't always the answer. Sometimes a simpler workflow, a better database query, or a conventional automation script solves the problem more effectively.

That's one reason I think experience with existing business systems is particularly useful when evaluating AI. You need to understand the process before deciding which technology to use.

What Happens to the Work?

I don't think it's realistic to say AI will simply free everyone to do more interesting things. Some companies will use the efficiency gains to handle more work with the same number of people. Others will need fewer people for particular tasks.

At the same time, work that was previously too expensive may become practical. A small company might be able to analyze its customer feedback, review documents, or develop internal tools that it couldn't previously justify paying for.

From my own experience, projects which were shelved due to cost or complexity became feasible with the help of AI.

I'm not convinced anyone can predict exactly how that balance will play out across different industries.

From a technology perspective, though, my approach hasn't changed very much. Understand the problem, work out what the current process costs, test alternatives, and measure whether the result is actually better.

AI gives us another set of options, and in some cases they're remarkably powerful. The challenge is knowing when to use them and when something simpler will do.