Introducing Workflows and Sub-Agents in IBM Bob v2

2026-07-15 · Source: Niklas Heidloff · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, short

Summary

IBM Bob v2 introduces workflows and sub-agents to enhance reliability, predictability, and efficiency in AI-driven software development. Workflows allow developers to define precise sequences of steps, blending deterministic code for structured processes (e.g., filesystem scans, shell commands) with AI reasoning for non-deterministic tasks like code generation or analysis. This hybrid approach ensures predictability, efficiency by using zero AI tokens for deterministic steps, and improved debuggability. Specialized workflows, such as the Java Unit Testing Workflow with eight sequential steps, are available through Premium packages like the IBM Bob Premium Package for Java Modernization. Additionally, Bob v2's sub-agents enable parallel execution of isolated tasks by spawning independent agent instances, each with its own clean context window. This significantly reduces wall-clock time for large tasks and maintains context hygiene, making them ideal for modular tasks like generating unit tests for multiple classes simultaneously.

Key takeaway

For AI Engineers developing agentic software, adopting a hybrid workflow approach with IBM Bob v2 can significantly improve reliability and cost-efficiency. You should structure your tasks to utilize deterministic steps for known processes, reserving AI for creative reasoning. Utilize sub-agents to parallelize independent tasks, reducing wall-clock time. This also maintains clean context windows for predictable outcomes.

Key insights

Hybrid AI agentic development combines deterministic steps with AI reasoning for efficiency and control.

Principles

• Deterministic tasks should avoid AI tokens.
• Isolate AI context for complex tasks.
• Define explicit steps for agent predictability.

Method

Workflows define a precise sequence of steps, alternating between local deterministic execution (e.g., file scans, shell commands) and AI-driven model invocations for generation or analysis, often incorporating interactive user gates.

In practice

• Use sub-agents for parallel test generation.
• Implement interactive steps for developer control.
• Structure tasks to minimize AI token usage.

Topics

• IBM Bob v2
• AI Agents
• Workflows
• Sub-agents
• Java Modernization
• Unit Testing
• Parallel Execution

Best for: AI Architect, AI Engineer, Machine Learning Engineer, MLOps Engineer

Related on AIssential

• Move faster, ship safer: RedMonk’s analysis on how leading engineering teams use agentic AI — Agentic AI enables software teams to automate routine tasks while maintaining control over quality and critical decisions.
• Why we built ADK 2.0 — ADK 2.0 combines AI agents with deterministic workflows for reliable, efficient, and secure production-grade AI applications.
• Use subagents and custom agents in Codex — OpenAI Codex now offers subagents and custom agents for modular, specialized AI-driven development tasks.
• 😸 Practical Workflows for working with Claude Cowork — Claude Cowork transforms AI from a chatbot into an active desktop agent capable of complex workflow automation.
• Blueprint First, Model Second: A Framework for Deterministic LLM Workflow — Decoupling LLM decision-making from workflow execution via code blueprints ensures deterministic, verifiable agent behavior.
• From assistants to agents — Agentic AI autonomously executes goals, shifting human-AI collaboration from reactive assistance to proactive, independent action.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Niklas Heidloff.