Enterprise AI shifts to agents
The enterprise AI market is moving from chat interfaces and novelty demos to agentic systems that orchestrate tasks and to pricing models that reveal commercial strategy. (reuters.com) Vendors are now publishing formal rate cards and usage tiers, and analysts say buyers care more about workflow execution, integrations and governance than raw model performance. (help.openai.com) AI agents for customer service are being pitched as the next practical layer—able to call tools and complete multi-step workflows—so security and sandboxing are becoming core procurement questions. (techcronus.com)
Enterprise buyers are moving from chatbot pilots to software that can finish tasks across systems, and vendors are now pricing that shift in public. (money.usnews.com) (help.openai.com) OpenAI’s strategy change is now part of the investor story. Reuters reported on April 14 that some OpenAI backers are questioning the company’s $852 billion valuation as it pivots toward enterprise customers and responds to competition from Anthropic. (money.usnews.com) An agent is software that does more than answer a prompt: it plans steps, calls tools, keeps state and completes multi-step work. OpenAI’s developer docs say the Responses application programming interface is recommended for new projects, and its Agents software development kit is built for applications that own orchestration, tool execution, approvals and state. (developers.openai.com 1) (developers.openai.com 2) The commercial model is changing with the product. OpenAI’s help center says ChatGPT Business and ChatGPT Enterprise adopted flexible pricing on April 2, 2026, added a new Codex-only seat type, and cut the price of standard ChatGPT seats by $5 per month. (help.openai.com 1) (help.openai.com 2) (help.openai.com 3) That matters because enterprise buyers are buying workflow coverage, not just model access. Google Cloud’s architecture docs describe agentic systems as orchestrators that connect disparate enterprise systems, while Microsoft says autonomous agents in Copilot Studio are meant to process updates, triage events and trigger follow-up actions within policy boundaries. (docs.cloud.google.com) (learn.microsoft.com) The procurement questions are also shifting from benchmark scores to controls. Microsoft’s agent governance guidance says agents that access business data and act across systems must be managed with security, governance and lifecycle controls, and Gartner says regulated industries will not accept black-box behavior without explainability, auditability and logged decisions. (learn.microsoft.com) (gartner.com) Customer service is one of the clearest early use cases because it already runs on defined workflows such as refunds, order changes and account checks. Gartner said on March 5, 2025 that agentic artificial intelligence would autonomously resolve 80% of common customer service issues by 2029 and cut operational costs by 30%. (gartner.com) The safety problem gets harder when an agent can take actions instead of drafting text. Anthropic said effective sandboxing needs both filesystem isolation and network isolation, because an agent compromised by prompt injection could otherwise leak sensitive files or download malware. (anthropic.com) The market is also getting more crowded around the same pitch. ServiceNow said in February 2026 that Claude would become the default model for Build Agent on its platform, and Google and Microsoft are both publishing enterprise architecture and governance guides aimed at production deployments rather than demos. (anthropic.com) (docs.cloud.google.com) (learn.microsoft.com) The next phase of enterprise artificial intelligence looks less like a chat window and more like a supervised worker with permissions, logs and a billable seat. Vendors are now selling that package in rate cards, software kits and governance controls instead of novelty demos. (help.openai.com) (openai.com)
