One thing I've noticed over the past year is that the conversation around AI coding assistants is changing.
A while ago, most discussions were about how quickly an AI could generate code. Today, the more interesting questions are:
- How do teams control what AI agents can do?
- How do they integrate into existing developer workflows?
- How do organizations enforce standards and governance?
The latest IBM Bob update seems focused on those questions.
One addition that stood out to me is the ability for administrators to lock Bob configurations across managed developer machines. Teams can disable auto-approval, control updates, and enforce required hooks, helping ensure AI-assisted development follows organizational standards instead of becoming a collection of custom local setups.
Another interesting enhancement is expanded hook support. Hooks can now run at multiple stages of the agent workflow, including:
- SessionStart
- UserPromptSubmit
- PreToolUse
- PostToolUse
- Stop
That opens up possibilities such as running linters automatically, triggering tests after changes, or preventing actions that don't meet project requirements.
IBM Bob also now supports the Agent Client Protocol (ACP), allowing it to work inside ACP-compatible clients such as IntelliJ IDEA, Neovim, and Zed while maintaining session management, guardrails, streaming, and tool integrations.
A smaller feature that I think could have a surprisingly large impact is support for MCP Elicitation. Rather than guessing when information is missing, the agent can involve the developer mid-workflow to gather additional context before proceeding.
The update also expands beyond coding tasks. Bob can now read, create, and edit Microsoft Office documents including Word, Excel, and PowerPoint files. As AI agents increasingly become part of day-to-day work, moving between code and documentation without switching tools feels like a natural next step.
For me, the most interesting theme across this release is not a single feature. It's the continued focus on making AI agents fit into real-world development environments through governance, workflow automation, and integration.
What do you think is the next capability AI coding agents need before they become a standard part of every engineering team's workflow?
Further reading: IBM Bob Release Notes