| Key Takeaways
● AI agents are making customised enterprise software faster and less expensive to build. ● General-purpose SaaS tools face greater pressure than specialised, industry-specific applications. ● The strongest case for building internally is where software strengthens a strategic capability or removes a meaningful bottleneck. ● SaaS vendors may need to move from seat-based pricing and feature competition towards business outcomes. |
For decades, enterprise software vendors benefited from a difficult choice facing business leaders: either accept an expensive, painful implementation or invest heavily in building a system internally. The effort required to design, integrate and maintain custom software created a durable advantage for packaged enterprise applications, first through traditional enterprise licences and later through software as a service.
Artificial intelligence is beginning to weaken that advantage. New tools can build applications around a company’s own workflows and data with far less capital and development time than were previously required. The result is not necessarily the immediate disappearance of SaaS, but a renewed willingness among companies to ask whether they should buy software at all, particularly when an off-the-shelf product is costly, inflexible or poorly aligned with how the business operates.
The build option is becoming economically credible
The most persuasive evidence is appearing in operating businesses rather than only in software markets. One building supplies company replaced its enterprise resource planning system with a system built using Claude. The core of the system was developed over a weekend, followed by two or three weeks of further work before it entered production.
The system was not a narrow experiment. An application helped installers take measurements at a building site, which were then converted into bills of materials and engineering drawings. The information flowed through invoicing, finance and marketing before jobs were queued into production machinery. Photographs were translated into drawings, drawings into computer-aided design files and those files into instructions for manufacturing equipment.
The company initially operated the new system in parallel with its existing processes so that it could test security, reliability and performance. Over time, however, the internally developed platform moved into a secure cloud environment and became the operating system for the business. The example suggests that AI is changing the economics of customisation, allowing a mid-market company to create software around its own process rather than redesigning its process around a vendor’s product.
Other organisations are using agents in accounting and administrative work, with different objectives. In one case, agents were introduced to help employees focus on higher-value activity. In another, they were used to reduce hiring requirements. The distinction matters because the economic effect of AI is not limited to replacing software subscriptions or reducing headcount. Agents can also increase the amount of valuable work existing employees are able to perform.
The first pressure point is SaaS bloat
Most businesses have accumulated more software than they fully use. Large applications often contain extensive functionality that remains dormant, while peripheral tools multiply across departments and create overlapping capabilities, recurring costs and integration demands. When a business can describe its need in a prompt and generate a tailored application or workflow, the value of unused features becomes harder to defend.
This creates a particularly direct challenge for general-purpose SaaS products that support broad administrative activities. A company may no longer need to purchase a full package for every employee when a smaller number of AI agents can perform much of the underlying work. The economic calculation changes from paying for access across a large user base to paying for the capacity required to complete a specific task or produce a defined result.
Customer relationship management illustrates the problem. Adoption and compliance have long been difficult because salespeople often regard data entry as administrative work that takes time away from customers. A system designed with the sales team, rather than imposed on it, could be prototyped around its actual needs. Agents could also absorb part of the administrative burden, giving salespeople more time for client-facing work without necessarily reducing the size of the sales team.
The commercial opportunity for buyers is therefore larger than simply lowering their software bill. A lower-cost system that frees human capacity can create more value if employees use that capacity to sell, serve customers or solve operational problems. The relevant question is not only how much a company spends on SaaS, but what work its current software prevents people from doing.
Not every SaaS category faces the same threat
The term “SaaSpocalypse” implies a uniform crisis across the software sector, but the pressure is likely to vary significantly by category. General-purpose tools that handle routine, horizontal activities may be easier to reproduce with AI agents than products built around specialised knowledge, regulatory requirements or deeply embedded industry processes.
Specialised software can remain valuable when it reflects expertise that is difficult to recreate, connects to complex operating requirements or supports activities where errors carry significant consequences. That does not make these products immune to change, but it may make them more resistant than applications whose main advantage has been packaging common tasks into a subscription.
Software companies are already responding by incorporating agents into their products and reconsidering how they charge customers. A move away from pricing solely by seat and towards pricing based partly on execution or outcomes reflects a different view of value. If a smaller number of agents can perform work previously assigned to many licensed users, charging for access alone becomes increasingly disconnected from the benefit the customer receives.
For incumbents, the challenge is difficult because the existing model may still generate substantial recurring revenue. A vendor must protect contracts built around the old model while developing a new one that may produce revenue at a different level and with different valuation characteristics. The transition is not simply a product upgrade. It may require changes to pricing, sales incentives, customer education and the internal operating model.
Incumbents face an operating-model problem, not only a technology problem
The shift from on-premise software to SaaS required enterprise software companies to rethink how they sold, delivered and monetised their products. The next transition presents a similar challenge, although the destination is less settled. SaaS providers may need to move towards models built around agents, outcomes or more adaptable combinations of software and services.
The difficulty is that an incumbent must manage both models at once. Existing contracts, customer expectations and sales structures are tied to recurring subscriptions, while the new model may involve fewer seats, more automation and a closer connection between price and business performance. Vendors must decide how quickly to change without undermining the revenue base that funds the transition.
They also need to consider how an AI-native challenger would approach the market. Such a challenger is likely to focus on customers that believe their current software is too expensive, too slow to change or too poorly fitted to their operations. A credible alternative that promises comparable capability at a fraction of the cost is likely to attract executive attention, particularly where customers are already dissatisfied with their incumbent systems.
Adding AI features to an existing product may not be enough if the underlying business model remains unchanged. The more consequential question is whether the incumbent can redesign its offer around the value customers are trying to create, rather than continuing to compete mainly through an expanding catalogue of features.
Build where software strengthens a strategic capability
The prospect of cheaper custom software does not mean every company should become a software developer. The more useful question is which capabilities will determine the company’s future competitiveness and what role software should play in bringing those capabilities to life.
Internal development becomes more compelling when it can create a new source of value more quickly, remove a bottleneck that limits performance or enable the organisation to compete in a way that standard software cannot support. In those cases, the software may be an important element of a strategic capability, particularly when it contains intellectual property that is closely connected to the company’s distinctive operations.
For other activities, buying a standard solution may remain the better choice. If an application does not differentiate the business, does not remove a meaningful constraint and can be replaced or adapted without affecting competitive advantage, outsourcing the capability may be more efficient than maintaining it internally.
This approach also gives executives a more disciplined way to evaluate build-versus-buy decisions. Rather than treating software as a strategic asset by default, leaders can start with the capability they need, the value they want to create and the constraints they need to remove. They can then determine whether a vendor product, a customised internal system or a combination of agents and existing tools is the most effective means of achieving that outcome.
The most important decision may be when to commit
Companies considering a major enterprise software purchase now face a more complicated time horizon. A system that appears suitable for a five-year commitment may be exposed to rapid changes in AI capabilities, pricing models and vendor viability. Waiting may preserve flexibility, but it can also delay improvements that would create value immediately.
One response is to experiment before making an irreversible commitment. Businesses can test whether agents can handle specific workflows, whether internally developed applications meet security and reliability requirements and whether employees will use systems they help design. Early experiments can also show which capabilities are genuinely strategic and which are simply administrative functions that do not justify long-term ownership.
For vendors, the same uncertainty makes speed more important. They must assess whether their existing customer relationships and recurring revenue give them enough time to adapt, or whether challengers will use dissatisfaction with slow releases, high costs and rigid workflows to enter the market. The threat is not limited to a competing software company. It can come from customers themselves, now able to assemble systems that previously required a large implementation programme.
The SaaS model is therefore likely to evolve unevenly. General-purpose horizontal applications may face the earliest and strongest pressure, while specialised products may prove more durable. The profit pool for software need not disappear, but a larger share may move towards AI-enabled solutions and the companies that can connect technology directly to customer outcomes.
For senior leaders, the central task is to separate the software they happen to use from the capabilities they must own or strengthen. AI is making the build option more credible, but its strategic value depends on where it improves the business rather than merely adding another technical project. Organisations that make that distinction well will be better placed to decide what to build, what to buy and how quickly to change course as the economics of enterprise software continue to move.