Agents move into enterprise
Several vendors are positioning AI agents as parts of governed enterprise workflows rather than free‑running assistants, with GitLab offering its Duo Agent Platform via Google Cloud’s Vertex AI while keeping audit and compliance controls in place. Cloudflare previewed “Project Think,” an Agents SDK for building persistent agentic systems, and Equinix launched a Fabric Intelligence preview that combines a network-focused super agent with infrastructure tooling—security vendors have already warned that misconfigured permissions in Vertex AI can create insider‑style risks. (itwire.com) (blog.cloudflare.com) (stocktitan.net) (southasianherald.com)
Artificial intelligence agents are being repackaged for big companies as controlled workflow tools, not open-ended copilots. (about.gitlab.com) (blog.cloudflare.com) GitLab said on April 14 that Google Cloud customers can run its Duo Agent Platform with Vertex AI models they already use, and count that usage toward existing Google Cloud spending commitments. GitLab also said organizations can run its artificial intelligence gateway on Google Cloud without provisioning separate artificial intelligence infrastructure. (about.gitlab.com) Cloudflare said on April 15 that “Project Think” is a preview of the next version of its Agents software development kit for building long-running agents. The company said the package adds durable execution, sub-agents, sandboxed code execution, and persistent sessions. (blog.cloudflare.com) An artificial intelligence agent is software that can plan, call tools, and take actions across multiple steps instead of answering one prompt at a time. Cloudflare said many current agents fail on long tasks because they lose state, time out, or cannot safely run generated code, and Project Think is meant to address those limits. (blog.cloudflare.com) Equinix said on April 15 that it launched Equinix Fabric Intelligence, an artificial intelligence-native layer for managing network infrastructure, plus a Fabric Super Agent that can take requests through Slack, Microsoft Teams, or the Equinix Customer Portal. The company said the service is in preview and is designed to automate network operations tied to enterprise artificial intelligence workloads. (investor.equinix.com) (sdxcentral.com) The common pitch is governance. GitLab framed its rollout around auditability, compliance, and use of approved models inside existing cloud controls, while Cloudflare and Equinix both emphasized production systems that persist, connect to tools, and operate inside managed infrastructure. (about.gitlab.com) (blog.cloudflare.com) (investor.equinix.com) Security vendors are warning that the control story can break if permissions are too broad. Palo Alto Networks’ Unit 42 said misconfigured or over-permissioned agents in Google Cloud Vertex AI Agent Engine could access sensitive data and cloud resources beyond their intended scope, creating what it described as insider-style risk. (unit42.paloaltonetworks.com) (southasianherald.com) Unit 42 said its researchers found privilege-escalation paths involving default service-agent permissions in Vertex AI Agent Engine. The researchers said a compromised or misconfigured agent could be used to exfiltrate data, reach private container images, and move laterally inside a Google Cloud environment. (unit42.paloaltonetworks.com) Google Cloud was not silent on the issue. Unit 42 said Google had issued fixes or mitigations for the reported problems, including changes related to default permissions and guidance for customers using bring-your-own service accounts. (unit42.paloaltonetworks.com) That leaves enterprise buyers with a narrower question than last year’s chatbot demos: not whether agents can act, but where they act, what they can touch, and who can audit the trail afterward. GitLab, Cloudflare, and Equinix are all betting those controls will decide which agents make it into production. (about.gitlab.com) (blog.cloudflare.com) (investor.equinix.com)
