IBM releases Bob AI coding assistant after testing on 80,000 employees, claims 45% productivity gains

IBM has released Bob, its AI coding assistant, to general availability following internal testing with 80,000 employees who served as early users. According to IBM, teams using Bob saw average productivity gains of 45% across complex, multi-step workflows.

The release coincides with the launch of IBM Bob Premium Package for Z, which integrates IBM watsonx Code Assistant for Z to target enterprise mainframe applications. The Z package is currently available as a no-cost private technical preview.

Technical architecture and pricing

Bob uses a combination of frontier large language models, open-source models, small language models, and IBM's Granite SLM family. The platform automates workflows across the full software development lifecycle, from discovery and planning through design, coding, and testing.

Pricing operates on a "Bobcoin" credit system, where each Bobcoin is valued at approximately $0.50:

• Pro tier: $20/month with 40 Bobcoins
• Ultra tier: $200/month with 500 Bobcoins

The credit-based pricing model follows GitHub Copilot's recent shift away from flat-rate pricing after the company found complex prompts unprofitable under its previous structure.

Mainframe focus and claimed results

The Premium Package for Z includes two specialized modes. Architect mode helps teams understand application structure, dependencies, and change impact before updates. Code mode generates and refactors code using "Z-aware context" for mainframe environments.

IBM reported internal testing on its RevTech platform delivered "10x project-based ROI," automated 300,000 payloads in testing scenarios, and reduced monitoring setup time from months to hours. The company positions Bob as particularly valuable for analyzing systems with technical debt and limited documentation—common characteristics of legacy mainframe installations.

Security concerns and competitive positioning

In January 2026, researchers disclosed vulnerabilities in Bob, finding the CLI could be manipulated to execute malware and the IDE was susceptible to AI-specific data exfiltration vectors.

Kate Holterhoff, senior industry analyst at RedMonk, noted Bob's multi-model approach as a differentiator. "This is a double edged sword, as developers can be suspicious of black box tools, but it also eliminates the paralysis of choice that comes from switching models between tasks," she said.

What this means

IBM's approach of testing Bob on 80,000 internal employees mirrors strategies from PwC and Accenture, which have tied employee progression to AI platform adoption. The credit-based pricing model represents an industry shift toward consumption-based billing for AI coding assistants, following GitHub's acknowledgment that unlimited usage models are financially unsustainable for complex AI tasks. IBM's focus on mainframe modernization targets a specific enterprise pain point where institutional knowledge is scarce and systems remain critical, though the platform's security vulnerabilities and the effectiveness of its claimed productivity gains remain to be independently verified.