A 55-person property firm was paying roughly $100,000 a year for Salesforce. Today it runs the same work on a custom app that costs about $3,600. That single swap captures a shift now rippling through small businesses: a growing number are choosing to cut software costs with AI by building the tools they used to rent. Over the past six months, at least five companies with 20 to 70 staff have ended contracts with Salesforce and HubSpot and replaced them with applications assembled using AI coding tools, trimming their software bills by 40 to 80 percent.
The $100,000 swap that started a trend
The property firm, Greenleaf Management, is not an outlier. The Seattle Seawolves, a 70-person professional rugby organization, used Anthropic’s Claude Code to rebuild both its Salesforce CRM and its AXS ticketing system in four months. The club cut around $100,000 in annual software spending and lifted revenue 25 percent since the season began in March. Its AI tooling costs went up, but as owner Adrian Balfour put it, the cost-benefit math still landed firmly in the club’s favor.
What ties these stories together is not a single vendor but a method. Teams are describing the workflow they need in plain language and letting AI tools from Anthropic, Replit and Lovable generate the working software. The practice even has a nickname now: “vibe coding.” A year ago it produced toys. In 2026 it is producing systems that small businesses run on.
Why “build” is starting to beat “buy”
Two forces are pushing owners toward custom apps. The first is frustration. In a recent Salesforce admin survey, nearly 59 percent said the CRM has become increasingly complex to work with — bloated with features most small teams never touch but still pay for. The second is capability. The same AI that now lets a one-person business punch far above its weight can generate a lean CRM that does exactly what a 40-person company needs and nothing it doesn’t.
The savings are real because the pricing gap is enormous. A per-seat SaaS contract charges the same whether you use 10 percent or 90 percent of the product. A custom app, by contrast, mostly costs what it costs to run — and running costs keep falling as the price of AI inference collapses. When the tool is cheap to build and cheap to operate, the buy-versus-build calculation tips.
How big could this get?
Gartner estimates that $234 billion of enterprise application software spending is exposed to what it calls “agentic arbitrage” by 2030 — roughly 20 percent of all enterprise SaaS spend. The incumbents have noticed. Salesforce and ServiceNow are moving toward outcome-based pricing, while ServiceNow, SAP and Workday have introduced controls requiring outside AI agents to pass through metered integration layers before they can reach customer data. That is what a market defending its moat looks like.
Where this makes sense for your business — and where it doesn’t
Replacing your entire software stack overnight is the wrong lesson to take from these stories. The firms that succeeded picked one expensive, poorly-fitting tool and rebuilt that. Good early candidates are systems you overpay for relative to how you use them: a heavyweight CRM, a clunky ticketing tool, a reporting dashboard you export to a spreadsheet anyway. Poor candidates are anything carrying serious compliance, security or payment obligations, where a mature vendor’s guarantees are worth the price.
Treat it as a real project, not a weekend experiment. Map the one workflow you want to replace, build a version alongside the old tool, and run both in parallel until the new one has earned your trust. Keep a human owning data migration, backups and access. This is exactly the kind of scoped, measurable initiative that makes or breaks small-team AI adoption — and it is worth knowing why so many AI projects stall before you start yours.
The takeaway
For most owners, the headline is not that Salesforce is doomed — it isn’t. It is that the option to build has quietly become realistic, and that changes your negotiating position even if you never write a line of code. The next time a renewal quote lands, “we could just build this” is no longer a bluff. Used deliberately, the ability to cut software costs with AI is less about firing your vendors and more about paying only for what you actually use.
