A company can renew its central human-resources system while declining to buy a new assistant attached to it. That possibility puts the next feature sale, rather than necessarily the entire platform, at the centre of the debate over customer-built software.
Business Insider’s 30 September reporting, available in its authorized Taiwan edition, describes Spotify’s internally built HR Bot and Twilio’s sales-coaching tool Jarvis. These are named examples of companies supplying particular functions themselves. They are not proof that enterprise customers have abandoned their core systems, or an independently audited estimate of savings across the software industry.
The financial question is how an internal alternative changes the price a customer will pay for the next purchased capability. AI-assisted development can make that alternative easier to consider. Whether it is cheaper over its useful life still depends on who operates it, which services it consumes and what happens when it fails.
The threatened transaction is an expansion sale
Software vendors often sell a base service and seek more revenue from additional capabilities. A customer that builds a useful narrow tool may still retain the main platform, its records and its existing subscriptions. The commercial pressure can therefore appear in a foregone expansion sale before it appears in a cancelled renewal.
That distinction matters for an investor interpreting a customer story. Keeping the account does not establish that the vendor will capture the same future spending from it. Equally, a customer’s decision to build one feature does not establish that all its spending with the vendor will disappear. The relevant comparison is the purchased feature’s value against the complete cost and reliability of the internal alternative.
A buyer with that alternative has another negotiating position. It may buy at a lower price, defer the purchase or build for a subset of employees. Those are possible outcomes, not measured effects established by the reported examples. The evidence does not supply a market-wide revenue loss or a timetable for one.
Spotify’s own August account of building Xirp supplies primary evidence for the broader mechanism. Xirp is a vendor-neutral environment for coordinating AI development work, built around the company’s engineering needs. This is a separate project from the HR Bot described in the news report. Spotify’s account shows custom development alongside external models and tools, rather than a simple choice to buy nothing.
Ownership continues after the first working prototype
The purchase-versus-build comparison becomes incomplete when it prices only the first version. An internal tool needs an owner after launch: someone must address failures, update connections and maintain the business rules that make its answers useful. Faster initial coding does not establish that those responsibilities have disappeared.
An employee-facing assistant illustrates the difference without requiring a forecast. A leave-policy answer depends on the employee’s situation, the current policy and permission to access the relevant information. An assistant can produce fluent language while using the wrong rule or disclosing information to the wrong person. The cost comparison must account for review, access control and correction, not just the number of developer hours saved.
In an October 1 account of its engineering history, Spotify describes how fragmented services and unclear ownership led it toward common standards and the Backstage developer platform. It also says operating such infrastructure remains an ongoing commitment and presents its managed Portal service as an alternative. This is a company account with a commercial interest, not an independent productivity benchmark.
The practical lesson is a cost boundary: custom features rely on surrounding systems that somebody must maintain. A company with established engineering ownership may absorb that work differently from one without it. Neither the subscription quote nor an impressive prototype alone answers which organization has the lower lifetime cost.
Platforms can sell the ground beneath custom tools
Incumbent vendors have a response besides trying to prevent customers from building. They can make their data, controls and execution environment useful to those builders. A customer-owned interface can then sit above a vendor-owned platform.
Workday’s June 2 announcement described Developer Agent for custom applications, Agent-Ready Tools for governed access to HR and finance functions, and Agent Passport for verification. The announcement specified early access and future availability plans at that time; it does not by itself establish their October availability or prove that any particular customer agent is safe.
The direction is economically relevant even without assuming the products achieve their promises. If development remains on the incumbent’s platform, the customer’s decision to build may redistribute revenue between features and platform services. It need not remove the underlying relationship. The unanswered question is what customers will pay for those layers and whether that offsets any lost feature sale.
Twilio’s May platform announcement describes Agent Connect as allowing businesses to choose models and agents while retaining the communications integration. That offers another example of the same distinction: writing a custom agent and purchasing the channels beneath it can coexist. The announcement establishes product strategy, not a guaranteed revenue outcome.
A smaller add-on bill need not mean lower total spending
For the customer, spending can move from an application subscription to model usage, infrastructure and internal operations. A build can still be attractive through customization or faster adaptation even if it does not minimize every cost. Conversely, avoiding an add-on fee is not sufficient evidence of net savings once the supporting services and ownership burden are included.
For the vendor, the information that would make the threat measurable includes changes in expansion revenue, renewal terms and demand for the development or infrastructure services beneath customer applications. Customer disclosures about sustained usage, maintenance effort, failures and total costs would test the other side of the bargain. The reported cases and product announcements do not yet settle those measures.
The analysis would become more negative for a vendor if customers repeatedly declined paid features without increasing spending on its underlying services. It would become less negative if custom development strengthened platform use and supported pricing. Both remain scenarios that need financial evidence.
The new competition is a credible internal option for a specific task. Its value comes from what it does reliably over time and which purchased layers it still requires. That makes the next SaaS feature sale contestable while leaving the broader spending outcome open.