Enterprise Software's Agentic Shift: What's Changing in 2026

Enterprise Software Watch | September 15, 2026

AI Agents Reshape the Enterprise Stack — Salesforce Dreamforce, Google Cloud, SAP and Agent Governance

Market Signal

AI Agents Move From Pilots to Production — and Governance Becomes the New Battlefield

AI agents entered a new phase this week as Salesforce, SAP, Google Cloud, Workday, and Microsoft moved from agent announcements to agent governance, observability, and outcome-based pricing. Salesforce opened Dreamforce with a new portfolio of named Agentforce agents and a multi-model platform strategy. Google Cloud and Accenture launched a 1,000-engineer Gemini Enterprise group. SAP doubled down on SAP's autonomous enterprise vision, and Workday confirmed agent adoption rose 35% in a single quarter.

The strategic throughline: agent governance and orchestration — not model capability — have become the new platform battleground. Gartner forecasts 40% of enterprise applications will embed task-specific AI agents by the end of 2026, up from under 5% in 2025. Yet only 17–31% of organizations have deployed agents to production, and 86–88% of agent pilots fail to graduate — a striking gap between feature embedding and operational deployment.

Enterprise Software Market This Week

Salesforce Opens Dreamforce with Seven Named Agentforce Agents

Salesforce unveiled seven job-ready agents — Hunter (B2B sales), Piper (inbound pipeline), Carter (shopper), Casey (help desk), Fin (customer service), Page (IT/HR), and Marshall (supply chain) — paired with an upgraded Agentforce Co-worker featuring multi-agent orchestration and Agent Script for constraining agent behavior. The company's Long Horizon Runtime lets agents execute multi-step workflows across weeks-long deal cycles.

Why this matters: By naming and packaging agents as discrete job functions, Salesforce is making agent adoption legible to enterprise buyers — and setting up an agent-level monetization model that decouples revenue from human seats. The company is pivoting from human seat licensing to action-execution pricing, monetizing Data Cloud and autonomous action credits to offset seat compression.

Google Cloud Named Gartner MQ Leader; Accenture Group Scales

Google Cloud was named a Leader in Gartner's inaugural 2026 Magic Quadrant for Enterprise AI Assistants, placing in the Leaders quadrant for both Completeness of Vision and Ability to Execute. The company's Gemini Enterprise platform offers a unified "AI front door" with open connectivity to Microsoft 365 and third-party software, built-in governance, and simple economics — chat and search included in the base SKU. A new Accenture Gemini Enterprise Business Group will deploy 1,000 engineers to help clients scale.

Why this matters: Google is positioning Gemini Enterprise as the interoperable control plane for heterogeneous agent environments — an explicit contrast with Microsoft's bundling strategy and Salesforce's walled-garden approach.

SAP Signals Only Complex Agents, Not Simple Tasks

SAP has signaled that it will release only AI agents capable of solving the most complex enterprise problems — not simple tasks any large language model can handle. The company is investing €100 million to accelerate customer adoption of autonomous workflows, and its Knowledge Graph gives agents a structured map of business entities and processes that general-purpose agents lack.

Why this matters: SAP leadership has acknowledged a structural headwind: 60% of SAP's on-premises customers have not yet migrated to the cloud — a prerequisite for the company's agentic roadmap. SAP's bet is that business context, not model capability, determines enterprise agent success.

Workday Agent Adoption Up 35% in a Single Quarter

Workday has shifted its agentic strategy from volume to value — consolidating simple agents into more capable agents targeting complex HR and finance processes. More than 5,500 customers now use one or more Workday agents.

Why this matters: Workday's experience suggests enterprises want fewer, more powerful agents that create measurable ROI — not a proliferation of narrow tools. This is a direct challenge to vendors that have pursued agent quantity over agent quality.

Airrived Launches Agentic Observability for Enterprise AI Agents

Airrived launched Agentic Observability for its enterprise Agentic OS, targeting companies deploying autonomous AI agents at scale. The product provides a full trace from data ingestion to business outcome — showing who created an agent, what permissions it has, whether human approval is required, and what sensitive data it touches.

Why this matters: Traditional software monitoring is insufficient for systems that reason, decide, and act autonomously. Governance and observability are rapidly becoming purchasing criteria in enterprise agent deployments.

AI Agents Reshape the Application Layer

For decades, enterprise software waited for humans to log in, navigate UIs, enter data, and trigger the next workflow step. AI agents dismantle that model by perceiving objectives, planning multi-step actions, and executing across applications autonomously.

The progression is now well established: traditional SaaS → AI-assisted SaaS → AI-native software → autonomous AI agents. Each phase changes the fundamental unit of value. In the copilot era, AI made existing interfaces easier to use. In the agentic era, users delegate outcomes, not tasks.

The agentic stack splits into three layers:

Frameworks for building agents: LangChain, LangGraph, and CrewAI provide components for orchestrating complex, stateful workflows.
Platforms for controlling agents: ServiceNow AI Control Tower, Salesforce Enterprise AI Harness, SAP AI Agent Hub, and AWS Multi-Provider Generative AI Gateway offer governance, observability, and cost management.
Enterprise products for deploying agents: Glean (enterprise search), Moveworks (employee support), Sierra (customer-facing agents), UiPath (automation with AI-driven exception handling).

Salesforce's Agentforce Co-worker now supports multi-agent orchestration — the ability to decompose complex tasks and dispatch them to specialized sub-agents. "Agent Script" gives enterprises a programming language to constrain agent behavior and reduce hallucination risk in critical workflows. This is not a feature update. It is a new software architecture.

The End of the Per-Seat Software Model?

Per-seat pricing worked when humans were the only users. Now AI agents outnumber employees 25-to-1 in some deployments, and the traditional software budget model is breaking. A Cruxy survey of 300 SaaS CEOs in April 2026 found 97% plan to retire seat-based pricing within two years — yet 94% said seat-based pricing currently aligns with their product's value.

Pricing Model	Billing Unit	Best Fit Workload	Where It Breaks
Per-seat	Human user or licensed seat	Chat assistants and productivity tools used directly by employees	Bot-to-bot workflows and fleets managed by a small team
Usage-based	API calls, tokens, tasks, minutes	Variable consumption workloads	Unpredictable costs and budget overruns
Outcome-based	Results achieved (e.g., tickets resolved, revenue generated)	Well-defined, measurable outcomes	Difficult to attribute causality
Hybrid	Base platform + variable usage	Enterprise platforms supporting many users and agents	Layered charges that obscure total cost

Salesforce shifted Agentforce to outcome-based pricing in August 2026, charging based on revenue gains or cost savings. HubSpot dropped its Customer Agent pricing to $0.50 per resolved conversation — half Intercom's price and a third of Zendesk's committed rate. OpenAI is testing performance-based pricing with major enterprise clients, charging only when AI completes a specific task successfully.

But the transition carries execution risk. Only 4 of 65 enterprise software companies analyzed by AlixPartners have fully adopted outcome-based pricing, and more than half still rely primarily on per-seat models. Attribution disputes loom large — Stripe has warned that sales conversions "may result from product changes, marketing campaigns, or seasonal factors" rather than the software itself.

Andreessen Horowitz estimates AI app companies spend 20–40% of revenue on inference and per-customer fine-tuning, compressing gross margins to 50–60% versus the 70–85% that SaaS built its valuations on. By 2030, Gartner predicts seat-based vendor revenue share will decline from 21% to 15% as agent- and outcome-based models expand.

Incumbents vs. AI-Native Challengers

Incumbents hold substantial advantages: customer relationships, proprietary data, integration ecosystems, and process logic embedded over decades. SAP's Knowledge Graph gives its agents a structured map of business entities and relationships that startups cannot replicate. Salesforce's customer data and Workday's HR records provide the context agents need to be effective.

But AI-native challengers have architecture on their side. An AI-native startup, Hang Ten Systems, backed by $32 million in seed funding, is building an "AI-native delivery model" using agentic code generation and reusable skills libraries — positioning itself as a direct alternative to traditional enterprise software implementation.

The company's founder has argued that every enterprise will be transformed by AI, and that the gap between leaders and laggards is widening.

The emerging consensus is a hybrid outcome: incumbents will retain the systems of record and the data, but the execution layer — where work actually gets done — is being rebuilt. Salesforce has acknowledged this tension directly, saying the company is following startups rather than leading the change on pricing innovation.

Microsoft, ServiceNow & the Agent Orchestration Layer

Microsoft: Multi-Model Copilot and Agent 365 Consumption Pricing

Microsoft is turning Copilot into a multi-model platform, adding OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1 to Copilot Cowork and Copilot Studio within four days of each other. Agent 365 at $15 per user per month and Copilot Credits priced from $0.01 each position Microsoft's agent registry as the enterprise control plane — bundling orchestration into M365 and Microsoft Azure commitment agreements.

ServiceNow: Open MCP Servers for Claude, Copilot, and Gemini

ServiceNow launched out-of-the-box MCP servers for Claude, Copilot, Gemini, and other AI clients on September 10, opening its platform to external agents. The company also reimagined its AI Agent Studio with the September 2026 release, offering a faster path from ideation to deployed agent. ServiceNow AI crossed $1 billion in annual contract value, with net new AI ACV growing more than 40% sequentially.

Why this matters: ServiceNow's structural advantage is that its agent platform operates directly on data, workflows, and business rules already in ServiceNow — avoiding the integration work external agents require. Opening its platform to external agents positions ServiceNow as the orchestration layer rather than the agent layer — a hedge against agent commoditization.

ERP, CRM, HR & Workflow Automation

SAP: Joule Work and the Autonomous Enterprise

SAP's AI strategy centers on Joule as the unified AI engagement layer, with Joule Work for intent expression, Joule Assistants for cross-functional coordination, and Joule Agents for specific multi-step tasks. Instead of navigating individual applications and entering data across screens, users interact primarily with Joule, which orchestrates the right combination of workflows, data, and agents to complete the outcome.

Oracle: Fusion Agentic Applications Across ERP and HCM

Oracle has launched 12 new Fusion Agentic Applications across ERP and supply chain, with 20+ total agentic applications planned. The company is moving HR from reactive talent systems to an "outcome-driven, proactive" operating model with new Fusion Agentic Applications and AI agents for hiring, manager coaching, and workforce management.

Workday: Sana Enterprise as an AI Workbench

Workday's Sana Enterprise is an AI workbench for HR, finance, and IT to build, orchestrate, and run AI across the entire enterprise. Purpose-built agents include a Talent Acquisition Agent for recruiting workflows and a Payroll Agent for payroll teams. Workday's collaboration with Google Cloud focuses on extending enterprise AI agents into the tools users already know, rather than asking customers to switch applications.

AI-Native Startups vs. Incumbent SaaS

A solid set of ready-made agentic AI products now serves organizations that don't want to build agents from scratch:

Glean: Enterprise search and internal knowledge agents
Moveworks: Employee support and internal workflow automation
Sierra: Customer-facing agents for support and sales
Cognigy & Kore.ai: Conversational automation at scale
UiPath: Existing automation base extended with AI-driven exception handling

However, incumbents retain structural advantages. ServiceNow's agent platform operates directly on data, workflows, and business rules already in ServiceNow, avoiding the integration work external agents require. SAP's Joule Agents execute within SAP's transactional systems as native capabilities rather than external add-ons. The moat is data and process logic — not model access.

Data, Security & Governance

Most agentic AI implementations are failing, but leading organizations that reimagine operations and manage agents as workers are finding success. Governance must happen at three levels:

Code level: Verification checkpoints embedded throughout the pipeline — security scanning, license checking, and policy validation must run continuously on agent-authored output.
Access level: Agents need scoped permissions aligned to the principle of least privilege.
Accountability level: Human accountability for every agent action is non-negotiable; code provenance tracking and scoped agent permissions are not optional.

Enterprise AI agents often stall at the first permission wall. What makes an agent enterprise-grade is the ability to operate autonomously after deployment while respecting guardrails, audit trails, and compliance requirements.

Salesforce's Enterprise AI Harness includes a context layer of data, metadata, and semantics; an AI reasoning and orchestration component; user-defined governance guardrails; and security tools that enforce identity, privacy, and permissions policies. SAP AI Agent Hub (via LeanIX) gives enterprises a single control plane to discover, govern, and monitor agents across all vendors, with full general availability planned for Q3 2026.

The New Enterprise Software Economics

AI infrastructure cost optimization has become a 2026 board issue as $487 billion in spend pressures SaaS margins, cloud contracts, and investor diligence. Inference spending in AI-optimized infrastructure as a service is set to surpass training for the first time in 2026, reaching $23.3 billion compared to $19 billion.

Company Type	AI's Role	Average Gross Margin
AI-augmented SaaS	Internal copilots, minimal customer-facing AI	~80%
AI-enabled SaaS	AI features inside a traditional product	60–79%
AI-native	The model is the product	50–59%

The operating metric that matters is gross margin after model, retrieval, storage, logging, and orchestration costs. Companies must track core software gross margin (target 80%+) alongside blended AI-inclusive gross margin (target 50–65%+ if AI is core to the product).

M&A & Strategic Moves

Notable 2026 acquisitions shaping the agentic stack:

NVIDIA → Hugging Face: $12.93 billion, the largest AI acquisition of the year, securing the open-source platform for AI and machine learning.
Salesforce → Fin: ~$3.6 billion, an AI-based customer service and operations software developer.
ServiceNow → Sweep: Hundreds of millions, adding cross-platform AI agents that work across ServiceNow, Salesforce, and HubSpot.
Palo Alto Networks → Console: ~$500 million, an AI-native platform for building agentic workflows and automating IT helpdesk tasks.
Okta → Permiso Security: ~$200 million, an AI identity security startup.
Cyera → Oasis Security: ~$1 billion, a "non-human" identity security provider.
SoundHound AI → LivePerson: $304 million true cost, bringing together OASYS voice AI and 1 billion monthly enterprise messages.

These deals describe a map of the AI stack from compute to corpus to agent to workflow, explaining where the weight of the industry is actually falling.

Strategic Analysis

The enterprise software industry is undergoing a transformation more profound than the shift from on-premises to cloud. We are moving through three phases:

Phase 1: Systems of Record (pre-2025) — Software that stores data and waits for human input.
Phase 2: AI-Embedded (2025–2026) — AI capabilities integrated into workflows, with agents that can execute defined tasks autonomously within a single application.
Phase 3: Agentic Platforms (2026+) — Multi-agent systems that coordinate across applications, data systems, and human processes to execute complex workflows end-to-end.

The winners will be platforms that control:

Enterprise data: Snowflake, Microsoft Fabric, and SAP Business AI Platform provide the foundational data layer upon which agents depend.
Agent orchestration: ServiceNow AI Control Tower, Salesforce Enterprise AI Harness, Microsoft Agent 365 bundle orchestration into platform agreements.
Infrastructure runtime: Datadog, NVIDIA, and hyperscalers monitor agent performance, security, and cost in production.
Governance and identity: SAP AI Agent Hub, Okta, Cisco (Astrix acquisition) secure non-human identities and enforce access policies.

Value is shifting away from thin wrapper software user interfaces toward underlying platforms that control enterprise data, security posture, and cross-system agent orchestration. The next major platform shift in enterprise software will be won by companies that provide the data, governance, and orchestration layers upon which intelligent agents depend.

What to Watch Next

Outcome pricing attribution disputes: The first major enterprise contract dispute over AI-attributed outcomes will test whether outcome-based pricing can scale — or whether it remains a premium-tier option alongside hybrid models.
Incumbent seat erosion: Watch for the first publicly reported decline in per-seat revenue at a major SaaS vendor attributable to AI agent adoption. Salesforce's FY2027 guidance and Workday's net new seat additions will be the leading indicators.
AI-native startups reaching enterprise scale: Sierra, Fin, and Hang Ten Systems are early tests of whether AI-native companies can win large enterprise contracts against incumbents. A single Fortune 100 deployment replacing a legacy system would validate the challenger thesis.
Agent-to-agent orchestration standards: MCP has emerged as the de facto standard for agent-to-tool connections, but agent-to-agent protocols remain fragmented. The vendor that establishes the dominant orchestration layer will capture disproportionate value.
Q4 2026: SAP Joule Work and Joule A2A (agent-to-agent) capabilities reach general availability, enabling multi-agent workflows across SAP's ecosystem.
Consolidation continues: Expect more acquisitions of AI-native startups by incumbent SaaS vendors seeking to fill agentic capability gaps.

The CODEW Analysis

The next enterprise software battle is not about interfaces. It is about who controls the AI agents executing the work.

For two decades, enterprise software value accrued to systems of record and the user interfaces built on top of them. AI agents break that model. When software perceives objectives, plans multi-step actions, and executes across applications autonomously, the interface ceases to be the product. The control plane becomes the product.

That is why this week's strategic moves matter. Salesforce named seven agents and shipped a governance harness. Google Cloud earned Gartner MQ Leader status and scaled with Accenture. ServiceNow opened its platform to external agents through MCP servers. Workday reported 35% agent adoption growth. Airrived launched agentic observability. The common thread is not AI features — it is agent orchestration, governance, and identity becoming the new foundation of enterprise software. Value is migrating from thin application layers to the platforms that own enterprise data, enforce security posture, and coordinate agents across systems. The winners of the next platform shift will be the companies that provide those layers — not the ones with the prettiest interface.

Sources

The CODEW — Enterprise Software Watch: Salesforce: Agentforce as the New Growth Engine & Microsoft: The Universal Enterprise AI Control Plane
The CODEW — The AI Reset of Enterprise Software
BaseRadar — AI Agents 2026: Enterprise Shift
IDC — The Agent Takeover: What Happens When AI Becomes the Primary User of Enterprise Software
Calcalistech — Workday Q2 2026 earnings coverage
Investing.com — SAP at Goldman Sachs Communacopia + Technology Conference; Palo Alto Networks / Console; Okta / Permiso; Cyera / Oasis
Futunn — Workday Q2 earnings call transcript
kblip — Palo Alto Networks paid $500M for Thrive-backed Console
Iskylar — 40% of enterprise apps to embed AI agents by 2026
Buttondown — Field Guide: The State of AI Agent Adoption 2026
Agentry — Gartner: Only 17% of orgs deployed AI agents so far
Analytics Insight — Best Agentic AI Tools, Platforms and Frameworks in 2026
AI Agent Store — AI Agent News, September 2026
InformationWeek — How AI Agents Broke Traditional SaaS Pricing
mpost — How AI Agents Are Transforming Enterprise Software
sokko.ai — AI Agent Pricing Comparison
Skydive — AI Agent Pricing Models
CloudZero — AI Gross Margin
CRN — The 10 Biggest Tech M&A Deals in 2026 So Far
Oracle Blogs — Fusion Insider Roadmaps
SAP Architecture Center — Reference Architecture
aiqod — Agentic AI Statistics
FutureIoT — New AI Agents Reshape Roles, Learning and Workforce Planning
Digital Thought Disruption — SaaS Pricing Reset: AI Agents Renewal Strategy
SiliconANGLE — Workday Reimagines Workflows with Enterprise AI Agents and Google Cloud AI Agents in Action
MarketIntel — AI Infrastructure Cost Optimization Becomes 2026's Boardroom Test
Deloitte — 2026 Tech Trends: Agentic AI Strategy
hrtechEdge — Workday Pushes Agentic HR as HCM Platform Wins Gartner Leader Status
uk.investing.com — Workday Q2 2026 earnings call transcript
CRN — Google Cloud Named Leader in Inaugural 2026 Gartner Magic Quadrant for Enterprise AI Assistants
Airrived — Agentic Observability Launch Announcement
Business Wire — Google Cloud and Accenture Launch Gemini Enterprise Business Group

The CODEW Stat

40% of enterprise applications will embed task-specific AI agents by the end of 2026, up from less than 5% in 2025. Yet only 17–31% of organizations have actually deployed agents to production, and 86–88% of agent pilots fail to graduate to production. Meanwhile, 97% of SaaS CEOs plan to retire seat-based pricing within two years — yet only 4 of 65 enterprise software companies have fully adopted outcome-based pricing.

The CODEW