Build or Buy?

Your manager reads about a shiny new agent product, and asks, should we buy that, or should you build one? Engineers love to build. Vendors love to sell. Neither of those is a good reason. Here's a better way to think about it.

• A shiny new product
• Build or buy?
• Neither love is a reason

A simple guide

Ask what makes your company special. If it's a common task, such as coding help, or searching documents, buy it. Thousands of people are improving those tools. If it's core to your business, using your own data and rules, build it. And there's a middle road, of buying the platform and building your own tools on top. That's where most companies end up.

Questions for any vendor

Before you buy, ask these. Can we export our data? How do we see what the agent did? Can we limit what it can touch? Where does our data go, and is it used for training? What happens when the model changes? And run it on your own test set, rather than on their demo. A demo always works.

The total cost

Building has hidden costs. Evaluation, monitoring, security reviews, and keeping up with models that change each month. Buying has hidden costs as well. Per seat prices, lock in, and adapting your process to the tool. Estimate both over two years. And include your people's time, as that's usually the largest number.

Summary