Key Takeaways
- IDC predicts 70% of software vendors will shift from seat-based to consumption or outcome-based pricing by 2028
- AI agents can draw on data from many systems at once — blurring the traditional boundaries between CRM, ERP, and other enterprise software categories
- A MuleSoft survey of 1,050 IT leaders, published in February 2026, found organisations use 12 AI agents on average, and 83% said most or all of their teams had adopted agents
- New job titles emerging: Agent Supervisor, Agent QA Lead, AI Ops Manager, Chief AI Officer
- Traditional enterprise software will not disappear — but the pricing models that fund it are being fundamentally restructured
The Facts
The enterprise software industry is confronting a structural challenge that has been building for three years but is now forcing board-level decisions: AI agents do not pay software seat licences. A seat licence model assumes that a human employee is the unit of software consumption — you buy one seat per person using the system. When an AI agent performs the tasks previously done by a human employee, the seat licence model captures none of that value.
Disclosure: RECATOOLS is written with the assistance of Claude, made by Anthropic, which is also named in this article. Readers should weigh that interest.
IDC's FutureScape: Worldwide Agentic AI 2026 Predictions report quantifies the expected response: by 2028, 70% of software vendors will have refactored their pricing strategies around consumption metrics, business outcomes, or organisational capability rather than user seats.
The more fundamental disruption may be structural rather than pricing-related. AI agents can read from a CRM, an ERP, a document management system, and a real-time data feed simultaneously and synthesise that information into action. This capability makes the traditional software category boundaries — the separation between CRM and ERP, between marketing automation and customer service — less relevant than it has been for four decades.
Technical Deep-Dive
The headless software model is the technical enabler of this transition. Traditional enterprise software was designed around human-facing interfaces — dashboards, forms, reports. The interface was the product; users learned to navigate it, and switching costs were partly learning costs. Headless architecture exposes all software functionality through APIs and CLI commands, enabling agents to access and act on data without the human interface layer.
Salesforce Headless 360, announced alongside Agentforce Operations, exemplifies this approach: the full CRM platform is accessible through API endpoints that agents can call from any surface — a Slack bot, a Claude conversation, an automated pipeline — without requiring agents to "navigate" the Salesforce interface as a human would.
Model Context Protocol (MCP) accelerates the headless transition by standardising how agents connect to any tool or data source. By late 2025, more than 10,000 public MCP servers had been deployed, meaning the majority of popular enterprise software tools had already created agent-accessible interfaces. The standardisation eliminates much of the bespoke integration work that previously made enterprise software connectivity expensive and slow.
The ASEAN Perspective
ASEAN enterprises are at an interesting position relative to this transition. The region has significant adoption of major global SaaS platforms — Salesforce, ServiceNow, SAP, Microsoft 365 — but also a large base of smaller, locally-developed software that lacks the engineering resources to rapidly build agentic capabilities.
For Singapore, Malaysia, and Indonesia's technology procurement teams, the pricing transition creates a near-term opportunity: as vendors shift to consumption-based models, the ability to negotiate pricing based on actual AI agent usage (rather than peak headcount) may reduce total software costs for enterprises that are aggressively automating routine processes with agents.
The emergence of ADLC roles — Agent Supervisor, Agent QA Lead — will create specific talent demand in ASEAN's technology labour markets. Singapore's SkillsFuture initiatives are well-positioned to develop training pathways for these roles, particularly given Singapore's existing strength in enterprise technology management.
RECATOOLS Verdict
The SaaS model is not dying — but it is being repriced. The transition from seat-based to consumption or outcome-based pricing is genuinely disruptive to software vendor revenue models, but the underlying value of enterprise software functionality does not disappear when agents start using it instead of humans.
For ASEAN enterprises, the practical implication for 2026 is contractual: when renewing enterprise software agreements, explicitly negotiate for consumption-based pricing options and avoid multi-year seat-count commitments that assume human headcount does not shrink as AI automation scales.
Frequently Asked Questions
SaaS seat licensing charges per human user. AI agents perform human tasks without consuming seats, meaning vendors capture no revenue from agent-driven usage under current models.
IDC predicts consumption-based pricing (per API call or task completed), outcome-based pricing (per business result achieved), and organisational capability will replace seat models.
Software architecture that exposes all functionality through APIs rather than human-facing interfaces, enabling AI agents to access and act on data without navigating a traditional UI.
A standard that allows AI agents to connect to any compatible tool or data source through a consistent interface, reducing bespoke integration work.
A MuleSoft survey of 1,050 IT leaders, published in February 2026, found organisations use 12 AI agents on average, and 83% said most or all of their teams had adopted agents.
Correction, 18 October 2026: An earlier version included a quotation attributed to Forrester analyst Kate Leggett and a quoted phrase attributed to Salesforce, neither of which we could find in the published sources; both have been removed. The 12-agent figure comes from a MuleSoft survey, not a Google report, and has been corrected.
View author profile → · Editorial policy
