Custom software was always an option. In the late 1990s and early 2000s, many enterprises took it seriously. They hired development teams, scoped projects for months, and launched initiatives that were supposed to deliver competitive advantage through proprietary systems. Most of those initiatives arrived late, cost more than projected, and solved a version of the problem that had already moved on by the time the code shipped. The lesson the market absorbed from that era was not subtle: building your own is expensive, slow, and organizationally draining in ways that don't show up in the initial budget.

Salesforce, Workday, and a generation of software as a service companies built their businesses on that lesson. The pitch was straightforward. Stop maintaining infrastructure. Stop managing upgrade cycles. Stop hiring specialists to keep a system running that your vendor's support team can handle at scale. Subscribe, configure, and focus organizational energy on the work the software enables rather than the software itself. It worked. Enterprise software became a subscription economy, and the SaaS model became the default assumption for any new category.

The market is now betting that AI reverses this. That premise deserves scrutiny.

The sell-off is pricing disruption in the wrong direction

ServiceNow is down roughly 40 percent year to date in 2026. Adobe and Salesforce have each shed about 31 percent. Workday is off more than 45 percent. On April 23, ServiceNow logged its worst single trading day on record, an 18 percent drop, after first-quarter earnings showed deal slippage in certain geographies. The broader software sector moved with it. The iShares Expanded Tech-Software sector exchange-traded fund fell roughly 6 percent that day and is down about 19 percent for the year.

The narrative driving that sell-off is that general-purpose AI, ChatGPT, Microsoft Copilot, and AI coding tools collectively will hollow out the enterprise software category. Why pay for ServiceNow's workflow automation when an AI agent can orchestrate the same process? Why pay for Adobe's creative suite when image generation tools exist? The market is treating these as live, near-term threats rather than potential pressures requiring evidence.

That framing confuses the prototype layer with the production layer.

Vibe coding is not a staffing model

AI-assisted development, the kind enabled by tools covered in this post from two weeks ago, has genuinely compressed the cost and time required to build functional software. That compression is real and it matters for the economics of software development. What it has not changed is the organizational requirement that follows the first working version.

Running enterprise software in production means data governance, model drift management, audit trails, access controls, compliance documentation, incident response, and continuous evaluation as the underlying models change. A team that "vibe codes" a procurement workflow tool over a weekend still needs someone accountable for what that tool does when an edge case surfaces at 2 a.m. or when a regulatory audit arrives. In the 2000s, the hidden cost of custom development was the maintenance burden after launch. AI-assisted development does not eliminate that burden. It accelerates the path to inheriting it.

The organizations that moved to SaaS did so partly because they recognized they were not in the software business. Most enterprise buyers still are not. The decision calculus has not changed as much as the current market narrative implies.

The real disruption is inside the category, not against it

What the market is getting wrong is the direction of disruption. The threat to incumbent enterprise software vendors is not that enterprises will build their own replacements using AI coding tools. The threat is that vendors who build AI with real governance, embedded guardrails, continuous model advancement, and auditable outputs will widen the gap on vendors who have bolted a chatbot onto a 2019 data model.

That is a meaningful distinction for enterprise buyers evaluating vendor roadmaps, and it is a meaningful distinction for investors reading earnings calls. ServiceNow's seat-based pricing model faces real pressure as AI lets organizations do more with fewer users. That is a legitimate question about revenue model durability. It is not the same question as whether enterprises will stop needing workflow orchestration software.

Adobe's position in creative and document workflows faces pressure from AI-native alternatives in specific use cases. That is also real. It is not the same as the creative and document workflow problem going away.

The vendors who will lose market position are those whose AI capabilities are cosmetic. The vendors who hold and grow position are those whose AI is doing real operational work inside systems enterprises already depend on. That distinction is not visible in a year-to-date stock chart that treats the entire category as one trade.

The infrastructure layer is getting the credit the application layer is not

AI infrastructure, compute providers, networking, model training platforms, is performing differently in the market. That divergence reflects something worth examining. Investors are comfortable assigning durable value to the picks-and-shovels layer of AI because the demand signal there is clear: more AI activity requires more infrastructure regardless of which applications win.

The application layer is being treated as inherently at risk. Some of that skepticism is warranted. Some of it is market sentiment overshooting in response to a narrative that AI replaces software rather than AI being delivered through software.

Enterprise buyers are not swapping ServiceNow for ChatGPT. They are asking whether ServiceNow's AI is substantive enough to justify the contract, and whether the vendor's pace of AI advancement will keep them ahead of what an internal team could assemble. That is a different and more tractable question than the one the market is currently pricing.