I felt something interesting for the first time last week. Something I haven’t felt yet in my AI Agent journey.
I felt like an AI tool was… expensive.
Not because it actually was expensive. But because my expectations have fundamentally shifted.
Let me explain.
The $50k AI Agent That Feels Cheap
On one end of the spectrum, we’re seeing AI agents that cost $50k a year and replace 10 SDRs. And somehow, that feels cheap. Or at least, OK value if it really, truly works. You’re paying $5k per SDR-equivalent, maybe less when you factor in benefits, management overhead, and the fact that they work 24/7 without complaining about quota.
The math just works. No one blinks at $50k when the alternative is $800k+ in fully-loaded headcount.
But on the other end? Our expectations of what should be included — what should just be part of the package — are evolving just as fast.
The $25 Add On That Felt Expensive
Here’s my example from the other day.
I’m building a new vibe coded game called Founderscape. I wanted to add native SSO — proper authentication, not just the basic stuff. So I asked my Replit agent which solution to use: native Replit auth, Clerk, or WorkOS.
The AI agent recommended Clerk. So I went with it.
Setup wasn’t quite as simple as I’d hoped, but net-net it was pretty easy. Works well. So far, so good.
Then I looked at pricing: $25-30/month to start.
Now, let’s be rational here. Is $30/month a lot for an SSO solution that has decades of combined engineering time behind it? That handles all the edge cases, security requirements, and enterprise features that would take months to build yourself?
Of course not. $30 is cheap. Objectively, laughably cheap for what you get.
But…
Compared to Replit, it seemed expensive.
That’s what all of Replit costs for a base subscription. The entire AI-powered development environment. Unlimited agents. Code generation. Deployment. Everything.
So Clerk at $30 felt kind of expensive. Even though it isn’t. And when I hit some issues trying to go from dev to production, I thought about churning at that price point.
We’ve Been Here Before
This is giving me flashbacks to the early days of SaaS.
Back when everyone was paying $99/month for Salesforce (most of us pay much more now), adding another $20/month for each SaaS add-on like EchoSign felt genuinely expensive. “Wait, I’m already paying for Salesforce. Now I need to pay 20% more for this integration? For this feature that should probably just be included?”
The psychology was real. The resistance was real. Even if the ACVs were actually pretty modest back then.
Then something shifted. Budgets exploded. Companies started buying 100, 200, 500 SaaS apps. And suddenly $20/month didn’t feel expensive anymore. It was a rounding error. Nobody cared, and then in 2020 and 2021, it really seemed likely nobody cared.
Until… it did feel expensive again. Until the CFO started asking why we have 847 software subscriptions and whether we actually need all of them.
We went through the whole cycle.
Where’s the Wall in AI Spend?
So here’s the question I’m sitting with: Where do we hit the wall in AI spend?
When does it start to feel like too many subscriptions? Too many AI agents we’re paying for? Too many $30/month tools that each feel reasonable in isolation but collectively start to add up? Too many $50k-$100k a year enterprise AI Agents running around, autonomously?
I don’t have the answer yet.
But I know I felt it for the first time the other day, at the start of 2026.
And if I’m feeling it — someone who has been all-in on AI, who runs an AI-first company with 20+ agents, who spends hours every day vibe coding new applications — then plenty of other buyers are going to feel it too.
What This Means for AI Founders
A few things to think about if you’re building AI products:
1. Your pricing anchor isn’t other point solutions anymore. It’s ChatGPT at $20/month. It’s Replit at $30/month. It’s Claude Code at whatever it costs. These horizontal AI platforms are becoming the new baseline that everything gets compared against.
2. The “should this just be included?” pressure is coming. Just like features got absorbed into platforms in SaaS, certain AI capabilities are going to feel like they should be table stakes. Authentication. Basic integrations. Simple automations. If you’re charging separately for something that feels like it should be bundled, you’ll face resistance.
3. Value demonstration has to be immediate and obvious. In the early SaaS days, you could sell on future value, on potential, on roadmap. In AI? People expect magic on day one. If your $30/month tool doesn’t feel like a no-brainer in the first five minutes, you’re going to churn.
4. Bundling and consolidation are inevitable. Just like we went from 500 point solutions to platform plays in SaaS, we’ll see the same in AI. The companies that win will be the ones that give you more for less, that reduce the number of subscriptions, that become the “one tool to rule them all” for their category.
Is The Pricing Honeymoon in AI Ending?
We’re entering a new phase in AI pricing psychology.
The honeymoon period — where everything AI-related seemed worth whatever they charged — is ending. Buyers are starting to develop intuitions about what things “should” cost. They’re starting to compare. They’re starting to feel that twinge of “wait, is this expensive?”
It’s not rational. A $30 SSO tool is not expensive. An $50k AI agent that replaces $800k in headcount is not expensive.
But rationality isn’t what drives purchasing decisions. Perception is. And ultimately, the size of IT budgets. They are expanding at recording rates, but they can’t expand infinitely and indefinitely. Not really.
And perceptions are shifting fast.
If you’re building in AI, you need to be thinking about this now. Not when the wall hits. Now.
Because I just felt the first brick get laid.
