This is not my main game
Vibe coding made building our own tools feel free. The razor I use for build vs. buy.
Last week I caught myself staring at our CRM, wondering if I could just rebuild it in Markdown files, Obsidian, and Claude Code.
A few webhooks, an API or two, and some MCPs to get data flowing between all the different tools we run. Maybe Obsidian, something like Karpathy’s Wiki, or a chief-of-staff framework like Dex. Honestly, any of them felt like they’d make more sense as a CRM than an actual CRM.
And then I actually said it out loud to myself, “This is not my main game.”
I’m glad I didn’t get too far. It was just a thought experiment I outlined on paper, but it was far enough to feel the pull. I really wanted to do this, and I took it far enough to see the cliff.
The moment you scratch the surface of a CRM, you realize how much actually goes into building one that does the things you want it to do in the first place.
You might have felt the same pull to build things rather than buy things. I think there are a few signs right now that the market is about to shift how we feel about this question.
We took the free sample
Vibe coding and all the hype around it made “we can just build this ourselves” feel true. I certainly felt it. When you can stand up something that works in 39 minutes, of course you think you can and should own this thing.
But of course, the price of building this right now, given AI tooling, isn’t a real price.
It’s easy to stay comfortably inside our CloudMax subscriptions right now. It’s $200 a month for a lot more than $200 in value, so it’s a great deal. It’s also subsidizing this build vs. buy conversation. All the big labs are selling compute below cost, which is great for getting us all using it and experimenting and playing.
But for a lot of businesses, there’s a huge risk involved. Ours included.
When an AI workflow or an AI-enabled workflow becomes load-bearing and runs whether or not someone is watching, it can become a liability later on.
The moment, for example, Anthropic decides to bump the rate cards from subscription to metered API, the math doesn’t just shift a little bit. It explodes.
This has implications for our build vs. buy decision-making. We need to be more mindful of our spend. What fits within a Claude Max subscription today could become a gigantic bill tomorrow. That includes AI-assisted workflows, but also all of these tools that we’re building instead of buying. We own those costs.
I haven’t seen it bite our customers yet, but it’s starting to come up in conversations, and the warning signs are adding up.
The State of Brand has written several articles about this price hike and its implications. If we’re not ready for it, it can be a scary, scary thing.
Building isn’t the cost. Maintenance is.
Even if the tokens stay cheap forever, there’s the thing that most of us don’t price in very well, and that’s people.
Every hour I spend building, tweaking, fixing, and maintaining an internal CRM is an hour I’m not selling our product. Unless this home-baked tool makes me 10x more effective in the hours I do sell, it doesn’t pay for itself.
I think about it like I think about our little family farm.
We have enough room to grow vegetables, run a few cows and some sheep. Every hour I spend in the garden is an hour taken away from the business and from my family. One of those things has to suffer. It’s not a discipline problem; it’s just how time works.
Some companies have the margin and the people to build their own tools. Ramp seems to be doing this a lot, but most companies can’t. The ones that do pull it off often take dedicated teams and months of effort to get there. To me, it’s like having a business within a business whose only customer is the main business. If that tracks.
The razor
Despite my shiny object syndrome and my strong urge to tinker with everything, I’ve had to keep asking myself one stupid simple question:
Is this my main game?
Be honest about the resourcing, not just to build it but to run it, to maintain it. Keep it alive. Keep it valuable.
What business are you actually in? If building this thing lets you do that business dramatically better, build it. If it would cost you dedicated staff time and money to get there when a tool already gets you 80 to 90% today, buy it and get back to work.
Most tools are about 80% of what you want them to be, especially given that every person and team across an org wants slightly different things. I have to remind myself that the same is true for the tools I build — unless I spend all of my time making it perfect, it’ll always be 80 to 90% of the way there.
Over the last year, dozens of prospects told us they’d build it themselves. A handful of them came back six to nine months later and said it was harder than they expected.
The rest? Some gave up, and some just did nothing about the problem.
And that seems to be the real shape of build vs. buy: not some triumphant in-house platform. Mostly silence.
Keep building and playing, though
Given how fast all of this is moving, we need to tinker and hack things together. We should be poking at these tools to see what they can do, not just for ourselves, but for our products and for our customers.
Remember, though, that every hour spent spinning up a tool that you probably won’t keep is an hour that you could have been working on the main game. For most of us, the main game is:
growth
keeping customers happy
winning new ones
selling
All that said, if you’ve been building anything with AI that you couldn’t have built before, I’d love to see what you’re doing. After all, it’s still fun to mess around and see what we can do.
Happy building, people.
Peter