When Anthropic in February unveiled plugins able to execute tasks such as automated reconciliation and contract review, markets quivered: shares in established software vendors fell on the prospect that intelligent agents could directly perform work that enterprise software once orchestrated. That moment crystallised a question for CIOs and investors alike — if an AI agent can call data and complete a task end‑to‑end, do companies still need complex SaaS stacks? Wang Wenjing, chairman and CEO of Chinese ERP and enterprise‑software vendor Yonyou, answered that question bluntly at Yonyou’s global ecosystem conference in Beijing on March 7: "AI will not kill software."
Wang’s argument is rooted in how AI capabilities and enterprise IT actually interact. He traces a technical trajectory from large models trained on public internet text, to models enriched with professional knowledge, and now to models conditioned on real business data inside companies. That last stage, he says, makes the enterprise environment a primary training and validation ground for AI, elevating the value of on‑premise and customer data rather than rendering it redundant.
For Wang, the defining change is not an extinction of software but a shift in its driving logic: enterprise systems will move from function‑and‑workflow centric architectures to ones driven by data and models. The new shapes include intelligent agents, domain ontologies and "skill packs" that can reason, coordinate and act; older process‑centric modules will remain because many tasks — precise accounting, deterministic production execution, high‑stability transaction processing — are best handled by rule‑based software.
He describes a dual‑mode architecture as the practical future for enterprise IT: deterministic, process‑execution software working alongside AI decisioning systems. In this arrangement the legacy systems are not victims but feeders — they generate the structured data that intelligent applications need to learn and operate. The implication is that rather than being displaced, SaaS will become an ingredient in a broader, hybrid stack where models and agents add decisioning layers on top of conventional transactional systems.
Wang also pointed to a fast‑moving adoption curve inside Chinese companies: from the initial phase of learning about AI in 2022, to experimentation, and now to a cohort of firms that are "mastering" AI in production. He argued that China’s combination of scale, dense business scenarios, plentiful data, and supportive policy and infrastructure gives domestic vendors an edge in building commercial enterprise AI at scale. That, in turn, makes China an important battleground for the business‑grade AI wave.
The commercial consequences are already visible. Wang notes that many of the fastest‑growing AI companies globally are focused on B2B offerings, suggesting an industrial window opening for enterprise AI. For incumbents with long customer relationships and accumulated datasets, the prize is a durable moat: ownership of business contexts and validated operational data will make it harder for generic AI agents to supplant tailored enterprise systems.
This convergence of models, agents and legacy systems does not resolve thorny practical problems. Integrating agentic services with established ERP, CRM and manufacturing systems raises questions about data governance, security, explainability and liability for automated decisions. Equally important are business model shifts: value capture may move from per‑user subscriptions to outcome‑based pricing, skill‑pack marketplaces and model‑as‑a‑service charges, upending traditional SaaS metrics and investor expectations.
Wang’s message is a strategic one for executives and investors: treat AI as an evolutionary force for enterprise IT rather than an immediate extinction event. Companies that have invested in stable, process‑oriented software can leverage those systems as the data backbone for more intelligent layers, while vendors must adapt by combining deterministic modules with model‑driven decisioning and agentic orchestration. The software industry will change shape, but it will still exist — more as a nervous system of data, models and services than as a collection of discrete, form‑fixed applications.