IBM launches Bob AI platform for SDLC governance

IBM launched Bob, a cloud-based AI platform that embeds across the software development lifecycle to reduce delivery costs, enforce security and compliance controls, and log automated changes. The platform is available now as a software-as-a-service product with a 30-day free trial. An on-premises version is planned for customers with strict data-residency or compliance requirements; IBM said existing watsonx Code Assistant customers will receive support while they plan migration paths.

Bob operates as an agentic development partner that can run in persona-based modes, call developer tools, and keep humans in the loop to enforce standards. The platform maps dependencies in legacy architectures before making changes, then coordinates specialized agents across testing, documentation and continuous integration pipelines to carry out refactoring and modernization tasks.

IBM designed Bob to limit risks when generative coding tools touch complex systems such as mainframes and tightly coupled corporate databases. The system evaluates a change request, assigns it to an appropriate model based on complexity and latency, applies security and policy checks, and only then performs edits or automated actions.

IBM piloted Bob internally in June 2025 with 100 developers and has expanded access to more than 80,000 employees. Internal surveys reported an average 45 percent productivity gain on tasks including new feature development, security remediation and modernization. IBM reported the Maximo team recorded a 69 percent time saving on complex refactors, and Instana reported roughly 70 percent reductions on specific assignments, about 10 hours saved per engineer each week.

External customers reported similar efficiency gains. Cloud provider Blue Pearl compressed a typical 30-day Java upgrade into three days and reported zero post-deployment defects on the affected platform. APIS IT used Bob to overhaul government systems across mainframe and .NET environments, producing architecture analysis and documentation ten times faster and achieving full accuracy on legacy JCL and PL/I assets. Veran Pokornić, solution architect at APIS IT, reported Bob migrated complex .NET services “in hours instead of weeks.”

To control costs and match model capability to task needs, Bob uses dynamic multi-model orchestration that routes requests to models based on complexity, latency and operational cost. IBM says the engine can call models including Anthropic Claude, open-source models from Mistral and IBM’s Granite, along with fine-tuned variants for edit prediction and security screening. The platform applies pass-through pricing and provides usage visibility so teams can align AI spend with production outcomes.

Security and auditability are built into the platform. Bob performs prompt normalization, sensitive-data scanning, real-time policy enforcement and automated red-team checks. Engineering leads can set manual approval gates or enable auto-approvals by task type. The BobShell command-line interface generates self-documenting agent processes so each automated decision and code change is traceable from initiation to deployment. Dinesh Nirmal, senior vice president at IBM Software, noted the platform is intended to let enterprises move quickly while keeping governance and security in place. Neel Sundaresan, general manager of Automation & AI at IBM Software, added that developers need a system that understands context and can act on it.