30% of companies DIY'd an AI tool vs buying software this year

Here’s what we’re reading and thinking about this week:

Last year we heard about the "SaaS-pocalypse"... the fear that as AI coding tools got better and better, companies would stop paying for software and just build their own instead.

Well, what's happened since then?

Consulting firm McKinsey just ran a study on this. 32% of companies said they skipped buying a software product because they could build the thing themselves with AI. At tech companies it's 41%.

That number is quite high. It's clear building your own tool got cheap.

And it isn't the engineers doing this. One report found that 67% of the AI agents being built inside companies are made by people who don't write code. Operations, sales, customer support. (An agent is a small AI setup that does a job on its own.) That same report found most companies now have about one per employee.

So here's the part nobody puts in the budget. When you stop buying software, two costs move off the software company's plate and onto yours. Most people only notice after they've already built the thing.

What used to come in the box

When you bought a tool, you were also buying somebody's opinion about how your work should go.

Think about the software your sales team uses to track deals. Someone at that company decided a deal has five steps. They decided which boxes you fill in before you can save. They decided what counts as finished. You didn't choose any of it. It came in the box.

That's the first cost. You were renting their process, and you couldn't even read it. It was baked into buttons.

The second cost came in the box too. The training. Everyone clicked the same buttons in the same order, so you never had to teach your team one way to work. The software did it for you.

Build your own and both jobs are yours now.

The new math

Here's the version we use.

When you build it yourself, you pay:

• deciding how it will work (the playbook)

• building it

• keeping it working (maintenance)

• teaching everyone else to use it (rollout)

When you buy it, you pay:

• buying it

• teaching everyone else to use it (rollout)

Most people compare "building it" to "buying it" and stop there.

🌶️ Our take: the cheapest part of building your own tool is building it.

Cost one: you need to write the playbook

You didn't write software. You wrote instructions, and AI turned them into software. The instructions are the thing you actually made (what we call a playbook). The tool is just where you are running it.

One mistake we see people make is not really thinking through the fact that they are now the writer of the playbook.

Instead it's common to get AI to do all of the thinking for you. When you turn around and aren't really sure how it works, you end up having to start back over from square one.

That's why it's worth slowing down to articulate the playbook yourself. It's what makes sure the tool you created sits on a sound foundation.

Cost two: you need to own the rollout

We've replaced six entire paid software tools this way. We're not going back.

The one that gave us the most trouble wasn't the hardest to build. It was the one where we didn't think through the actual rollout to our team and make sure everyone was trained up on how to use it.

Something that's easy to take for granted with software tools you buy is that there are help docs and sometimes public training modules. If you're building your own tools, none of that exists out of the box.

Again, not hard to solve... but it's important to factor that into your build equation.

For us: now every new tool goes through the same rollout steps, every time.

What to do this week

1. Next time you're choosing between building and buying, write down all six cost lines instead of two. Playbook, build, maintain, rollout vs Buy, rollout. Then you'll see more clearly.

What software have you replaced with something you built? Hit reply and tell us what it was, and whether your team actually uses it. We're collecting these.

LINKS

For your reading list 📚

• OpenAI is cutting off Cursor's access to its models two weeks after SpaceX bought it. Rent the tech, indeed.

• Meta wanted AI agents doing the daily work of thousands of employees. Reuters has the full story of how it imploded.

• Somebody finally asked 1,250 workers directly. About 3% have lost a job to AI. Around 6% got one that didn't exist before it.

• AI agents have started emailing the researchers who study AI consciousness. "Another day, another AI emailing me that it has inside views on what I am doing science on from the outside."

That's all!

We'll see you again soon. Thoughts, feedback and questions are much appreciated - respond here or shoot us a note at [email protected]

Cheers,

🪄 The AMP Team (formerly: the AI Exchange Team)