Agentic AI boom drives network security market surge and infrastructure shift

Three software vendors launched coding agents designed for team collaboration in the first week of June 2025, marking a shift from single-developer use to enterprise-scale agentic AI deployments. The releases signal that the industry is moving beyond isolated AI assistants toward agents that interact with shared infrastructure across development teams.

The network security market rose 14 percent in the same period, driven largely by buyers seeking consistent AI guardrails across users, applications, clouds and branch locations, according to a Dell'Oro Group report. The simultaneous jumps in agent tooling and security spending reflect a common driver: organizations are deploying AI agents beyond sandboxed experiments and must now secure them at production scale.

Coding agents move from solo to team infrastructure

Vendors shipped agents that operate on shared Git repositories, collaborate on pull requests and coordinate with CI/CD pipelines. These agents differ from earlier coding assistants in that they persist state across sessions and can act on behalf of the team, not just an individual developer. The shift requires new access control, audit logging and identity management because a compromised agent could alter code or pipeline configuration across multiple projects.

• Agents now require scoped permissions to repos, secrets and deployment environments
• Teams must log agent actions at the same level as human commits
• Git platforms are adding agent-specific service accounts and rate limits
• Observability tools need to trace decisions from model call to merged code

Cost and privacy drive model switching

Lindy, an AI agent startup, switched its inference provider from Anthropic to DeepSeek and reported saving millions of dollars annually. The move illustrates a broader trend: as agentic loops grow longer and more expensive, inference cost has become the main barrier to sustainable deployment. Lindy's switch underscores that organizations will favor cheaper models when agents must run hundreds or thousands of calls per task.

Apple is taking a different approach. At its June 2025 developer conference, Apple emphasized on-device processing and contextual privacy for its AI features. The company is courting developers who need agents to handle sensitive user data without sending it to cloud APIs. Apple's pitch is that agents running locally can enforce privacy constraints that cloud agents cannot guarantee, a selling point for regulated industries like healthcare and finance.

GitHub abandoned its flat-rate Copilot subscription model earlier this year, moving to usage-based pricing. The change reflects the same cost reality that Lindy acted on: agents consume variable compute, and flat pricing does not align with the economics of long-running autonomous tasks.

What comes next for agentic AI infrastructure

The convergence of team-oriented agents, rising security spend and model-switching cost pressures points to a maturing market. Infrastructure vendors will need to deliver identity-based agent access controls, granular cost attribution and privacy-preserving execution environments. The next six months will likely see more providers fragmenting their model offerings by task, routing simple lookups to cheap models and complex reasoning to premium ones, all while maintaining a unified security policy across agents, users and API gateways.