IBM launches Bob, an AI coding platform with human checkpoints built in
IBM announced the global launch of Bob, an AI-powered software development platform that writes and tests code while requiring human approval at key workflow stages. The platform is already in use by more than 80,000 IBM employees after a pilot that started with 100 users in summer 2025.
Bob differs from other AI coding tools by inserting structured checkpoints where humans must review and approve the agent's work before it proceeds. This approach reflects how enterprises want to balance automation with control-moving away from fully autonomous systems toward workflows that keep developers in the decision-making loop.
IBM says some teams have saved up to 70% of time on selected tasks, averaging 10 hours per week. The platform supports multiple models, including IBM's Granite series, Anthropic's Claude, and Mistral, rather than relying on a single model.
How Bob differs from other coding tools
Tools like Cursor and Claude Code put the developer at the start of the process-writing prompts, chaining steps, and debugging. Bob pre-structures the development lifecycle into role-based stages where agents check in with users for approval as a natural workflow checkpoint.
Neal Sundaresan, general manager of Automation and AI at IBM, said the distinction comes down to design philosophy. "Model capability alone isn't enough," he said. "How you deploy it, how you structure context, and how you keep humans in the loop is what determines whether AI actually delivers."
The shift reflects a broader divide in how enterprises approach AI development: some prioritize flexibility and experimentation, while others demand reliability and auditability. Bob targets the latter group.
The broader debate over autonomous agents
Enterprises continue to experiment with how much autonomy to give AI agents. Some providers, like Nvidia with NemoClaw and Kilo with Kilo Claw, have added security layers around autonomous agent systems. OpenAI's updated Agents SDK includes sandbox implementations that limit agent actions.
Sundaresan acknowledged that fully autonomous agents may eventually become standard, but he argued for caution. "If you tell me that the final answer will be OpenClaw, then we will get there," he said. "But it's better to open the gate slowly than say, 'oops, how do I close it now?'"
Pricing and tiers
Bob uses a credit system called "Bobcoins" for usage tracking. One Bobcoin equals $0.50 USD. Users consume credits by generating code, running commands, and performing file operations.
IBM offers four individual subscription tiers:
- 30-day free trial: 40 Bobcoins
- Pro plan: $20 per month, 40 Bobcoins
- Pro+ plan: $60 per month, 160 Bobcoins
- Ultra tier: $200 per month, 500 Bobcoins
All individual plans include access to specialized agentic modes, literate coding, the Bob Shell for CLI workflows, and Model Context Protocol integration. An Enterprise plan is available through sales contact, offering team management, flexible roles, and distributed credits across an organization. Enterprise subscribers also get priority support and usage dashboards.
Bob is now available globally wherever IBM conducts business.
Developers interested in AI-powered code generation can explore Generative Code Courses or broader AI Coding Courses to understand how these tools fit into development workflows.
Your membership also unlocks:
