At TechXchange 2025, its annual event for developers and technologists, IBM unveiled a suite of capabilities – some already available and others in preview – with a clear goal: moving from AI pilots to production and unlocking productivity across the entire technology lifecycle, from development to operations and business workflows. The company frames the announcements along four main themes: agent orchestration, governance and observability in hybrid IT, developer productivity, and open options for adopting AI models and tools without vendor lock-in.
“AI productivity is the new business speed,” summarized Dinesh Nirmal, Senior Vice President of Products at IBM Software. “These enhancements elevate developer efficiency, agent orchestration, and infrastructure intelligence to a new level.”
Open “Black Box” Agents: Watsonx Orchestrate Adds AgentOps, Standardized Flows, and Visual Integration
At the core of IBM’s agent framework is watsonx Orchestrate, which consolidates over 500 tools and domain-customizable agents created by IBM and its ecosystem. The approach is tool-agnostic and portable across any environment so that clients can deploy and govern agents at scale.
The star feature is AgentOps, an integrated layer of observability and governance covering the entire agent lifecycle: real-time monitoring, policy-based controls, action traceability, and anomaly detection to correct deviations on the fly. IBM’s classic example is an HR onboarding agent: without AgentOps, compliance and sensitive data handling are only checked post-issue; with AgentOps, every step is monitored and governed.
Additionally, IBM announced:
- Agentic workflows (general): standardized, reusable flows that sequence agents and tools with consistency—a robust alternative to fragile scripting that breaks at scale.
- Integration with Langflow (in tech preview, general availability expected late October): drag-and-drop visual builder enabling non-technical teams to set up agents in minutes.
The roadmap includes bringing these capabilities to mainframe with the new watsonx Assistant for Z: agents for IBM Z capable of understanding conversational context and automating operational processes with security and compliance, evolving from reactive resolution to proactive system management. Following the launch of IBM z17, IBM promises a redesigned experience that optimizes mainframe user productivity.
“InfraGraph”: An Intelligent Control Plane to End Hybrid Cloud and IT Silos
Post-acquisition of HashiCorp, IBM introduced Project InfraGraph, aimed at replacing fragmented tools and manual processes with a unified, intelligent control plane for observability. In multi-cloud and hybrid environments, tool proliferation causes information silos and manual cycles.
The use case IBM highlights: when a critical CVE emerges, typical response involves coordinating via email with multiple teams and tracking manually in spreadsheets to ensure all instances are patched. With InfraGraph, organizations get a single view of their infrastructure estate and security posture, both in HashiCorp Cloud Platform (HCP) and external environments, with near real-time data. Users can drill down to clusters or containers (e.g., within a VPC) and see affected components without relying on CSV reports.
Delivery and scope: InfraGraph will be offered as a capability within HCP, with plans to expand integrations toward IBM’s portfolio: Red Hat Ansible and OpenShift, watsonx Orchestrate, Concert, Turbonomic, and Cloudability, to unify infrastructure, security, and applications based on consistent data and policies. HashiCorp has opened sign-up requests for the private beta, scheduled to start in December 2025.
“Project Bob”: An AI-Powered IDE for End-to-End Modernization, Testing, Deployment, and Security
Another significant announcement is Project Bob—currently in private technical preview—an AI-native IDE aiming to transform the software development lifecycle (SDLC) in enterprises, starting with application modernization. Unlike current code assistants, Bob orchestrates complex tasks such as writing, testing, updating, and securing while maintaining context across sessions and artifacts.
Key capabilities:
- Scalable modernization: system updates and framework migrations with multi-step refactoring and architecture awareness in large codebases.
- Smart generation and review: assistance aligned with enterprise patterns, security requirements, and compliance obligations.
- End-to-end orchestration: covering development through testing, deployment, and maintenance, with chained tasks and persistent context.
- Shift-left security: early scanning, accelerated FedRAMP hardening, and cryptographic migration—embedded in workflows.
Instead of a single model, Bob coordinates among leading LLMs (Anthropic Claude, Mistral, Llama, IBM Granite) based on task requirements. IBM is opening access requests for interested organizations.
“Choice and Flexibility” as Policy: Partnership with Anthropic and Vendor-Neutral AI
For IBM, a common obstacle in AI adoption is inability to deploy according to organizational requirements: where, how, and with what technology to operate, without vendor lock-in. The company is expanding its partner ecosystem to provide integrable AI tools and services within existing environments.
In this vein, IBM announced a new alliance with Anthropic: integrating Anthropic’s LLMs directly into select IBM products, starting with Project Bob. It has also created a verified guide—“Architecting Secure Enterprise AI Agents with MCP”—focused on the agent development lifecycle: designing, deploying, and managing secure enterprise agents.
Why It Matters: Moving from Testing to Production with Control
The overarching narrative of the announcements emphasizes that operationalizing AI requires three foundations:
- Governable agents: without observability, policies, and auditing, automation with agents won’t scale in regulated sectors. AgentOps and standardized flows aim to close this gap.
- Unified operational data: if infrastructure, security, and applications remain in silos, incident response and change management become manual and slow. InfraGraph envisions a single visibility and action plan.
- Developer productivity aligned with enterprise goals: an IDE that modernizes, tests, deploys, and secures with context reduces delays and errors compared to disconnected toolchains.
The underlying technical rationale is clear; the challenge is execution: ensuring that AgentOps and Orchestrate are usable by business and IT teams, that InfraGraph truly connects with heterogeneous ecosystems, and that Bob delivers measurable value in mixed teams (legacy apps, microservices, infrastructure as code).
What Companies Can Expect (Adoption Scenarios)
- AI Center of Excellence needing to move to compliance: AgentOps for auditing agent decisions and enforcing policies (PII, segregation of duties, retention).
- Large-scale hybrid operations: InfraGraph as a single source of truth (HCP and external) for closing patch gaps and security posture.
- Continuous modernization: Bob as a lever for updating frameworks, refactoring, and migrating components with shift-left security and planned quantum-safe cryptography.
- Mainframe + AI: with watsonx Assistant for Z, mainframe teams automate operational tasks with agents capable of understanding context and compliance controls.
FAQs
What problem does AgentOps solve in watsonx Orchestrate?
It makes visible and govern what the agent does: monitors actions in real-time, applies policies, records decisions, and alerts on anomalies. It’s key to trusting agents in production and in regulated sectors.
What is Project InfraGraph and how will it be delivered?
It’s a unified control plane—with near real-time data—for observability and infrastructure inventory in hybrid/multicloud IT. It will be available as a capability in HCP, with planned future integrations including IBM/Red Hat products (Ansible, OpenShift, watsonx Orchestrate, Concert, Turbonomic, Cloudability). The private beta is scheduled for December 2025.
How does Project Bob differ from a code assistant?
Bob is an AI-powered IDE designed to modernize and orchestrate the entire SDLC: writing, testing, deployment, maintenance, and security with persistent context. It coordinates multiple LLMs (Claude, Mistral, Llama, Granite) and integrates security (shift-left, FedRAMP, quantum-safe cryptography).
How does IBM Z fit into this agent strategy?
With watsonx Assistant for Z, IBM plans to develop mainframe-specific agents that transition from reactive incident response to proactive management of operational processes, maintaining security and compliance in Z environments.
via: newsroom.ibm