What Actually Happened
Eric Nowoslawski cancelled his Apollo.io subscription and replaced its list-building function with a custom Claude skill he published open-source on GitHub. Not a prototype. Not a proof-of-concept. A working replacement he's running in production, with 30+ additional outbound skills in the same repo available to fork today.
This matters beyond the headline. Apollo.io isn't a niche tool — it's a standard fixture in mid-market RevOps stacks, running $50–100+ per seat per month. Cancelling it in favor of a configurable Claude skill means someone with operator-level judgment decided the SaaS value proposition no longer held.
Why This Is a RevOps Architecture Signal, Not a Hack
The standard objection to DIY AI tooling in revenue teams is maintenance burden: who owns it when the prompt breaks, the model updates, or the data source changes? That objection is real, but it's been overstated as a reason to keep paying SaaS vendors for logic you can now own directly.
What Nowoslawski demonstrated is that a Claude skill can encode prospecting logic — ICP filters, enrichment sequencing, list criteria — in a way that's auditable, version-controlled, and modifiable without a support ticket. That's a meaningful capability shift. You can read the logic. You can change it this afternoon. You can adapt it when your ICP shifts next quarter.
Contrast that with a SaaS tool where the list-building methodology is a black box, the filters are constrained by the vendor's data model, and any customization requires a feature request or an enterprise tier.
For non-tech revenue teams specifically, the unlock is control without engineering dependency. If you can write a clear brief, you can direct a Claude skill.
The Actionable Takeaway
Before your next SaaS renewal — especially any prospecting, enrichment, or outbound sequencing tool — run a one-question audit: Is the core value here data I can't get elsewhere, or logic I'm paying someone else to maintain for me?
If it's logic, it's now a candidate for consolidation into an AI-native workflow. Pull the GitHub repo, map the skill against your current tool's primary use case, and pressure-test whether a configured Claude skill covers 80% of your actual usage.
This isn't about eliminating all SaaS. It's about identifying which line items are paying for configurability you can now own. Start with the tools where you've customized the least — those are the ones where you're getting generic logic at a premium price.