The Information reported this week that five small companies, each with 20 to 70 employees, canceled their Salesforce and HubSpot contracts over the past six months. They replaced them with applications they built using Claude, Lovable, and Replit. Software costs dropped 40% to 80%.

Gartner followed up with a number that stopped me: $234 billion in enterprise SaaS spending is exposed to what they’re calling “agentic arbitrage” by 2030. That’s 20% of all enterprise SaaS. Retool’s own survey found 35% of enterprises have already replaced at least one SaaS tool with a custom-built alternative.

This is moving faster than most people realize.

The part everyone misses

The coverage is framing this as a cost story. Save $100,000 a year like Greenleaf Management did when they replaced Salesforce with a $300/month Claude Code app. Replace a $40,000 CRM with a Lovable build that costs $1,200 annually like Atonom did.

The savings are real. But that’s not what I find interesting.

What I find interesting is what the Atonom CRO said: “No one was using Salesforce to its fullest potential. I don’t know if we need every bell and whistle known to man on CRM.”

That’s not a cost complaint. That’s a fit complaint.

For 20 years, software vendors sold you their vision of how a business should work. You paid for it, trained your team on it, and bent your processes to match it. If your workflow didn’t map cleanly to the vendor’s data model, you hired a consultant to fix you, not the software.

AI flips that. For the first time, the cost of building software that fits your specific process is low enough that the build-vs-buy math actually changes for a 45-person company.

What this means in practice

At NexHam, we didn’t build a wrapper around Salesforce or HubSpot. We built what we needed: a content pipeline, a structured memory system, a client operations layer. Custom, because custom fits, and the off-the-shelf options were built for someone else’s version of this problem.

That doesn’t mean you should go build a CRM tomorrow. “AI can build it” is not the same as “you should build it.” The companies getting this right started with a specific, painful, well-understood process and replaced exactly that. Not their whole stack.

Start with the workflow that costs the most and fits the worst. If AI can model it accurately, you probably don’t need a vendor for it anymore.

That’s the real shift.