The legacy tech company is positioning Bob as a way to bring structure and guardrails to the growing use of AI agents in software development. As organisations move from small pilots to real-world deployments, they face new risks around security, orchestration, and reliability when agents start interacting with live systems and data.
Bob is already widely used inside IBM. According to the company, the platform began with around 100 internal users in the summer of 2025 and has since expanded to more than 80,000 IBM employees globally.
Rather than turning development over to fully autonomous agents, Bob adds a structured layer to the software lifecycle that deliberately pauses for human intervention. IBM describes a system in which AI handles “agentic” tasks such as writing and testing code, but progress is gated by human-led checkpoints throughout the workflow.
The company says this approach has delivered significant efficiency gains for some internal teams. On selected tasks, IBM reports time savings of up to 70%, which it equates to an average of about 10 hours saved per week.
By baking human approval and review steps into the orchestration, Bob is meant to address two concerns that are increasingly top-of-mind for enterprises experimenting with AI in development:
Security and reliability: Systems that look promising in controlled pilots can behave unpredictably when connected to real-time data and production environments.
Auditability and control: Organizations need to understand and document how AI-driven changes are made, and ensure humans remain accountable for decisions.
The result is a more guarded form of automation that emphasizes oversight rather than full hands-off operation.
Bob is built to work with multiple AI models rather than relying on a single large model or orchestration framework. The platform supports:
IBM’s own Granite series of models
Claude from Anthropic
Selected models from French AI company Mistral
Other smaller distilled models
Notably, the roster does not currently include Alibaba’s Qwen or other fully open source models, according to the information provided. Instead, the mix reflects a curated, enterprise-focused set of options designed to fit within IBM’s broader AI and automation strategy.
This multi-model routing is part of a broader shift in how large organizations are thinking about AI development tools. Rather than using a single model to generate code snippets or applications in isolation, enterprises are looking for systems that can:
Orchestrate complex, multi-step workflows across the software lifecycle
Combine different models for different tasks, based on their strengths
Keep humans in the loop to review, approve, and adjust AI-generated work
IBM’s framing of Bob reflects that trend. The platform is presented less as a standalone code assistant and more as an AI-enabled partner embedded in the development process, from writing and testing code to preparing it for production.
The company also positions Bob as a way to address “orchestration failures” that can emerge when AI agents are chained together without sufficient structure. By enforcing a sequence of checks and human sign-offs, IBM aims to reduce the likelihood that an error in one step propagates unchecked through a larger workflow.
As enterprises accelerate their use of AI agents inside the software development lifecycle, tools like Bob are likely to be judged not just on raw productivity gains, but on how well they manage risk, maintain transparency and fit into existing governance frameworks. IBM’s emphasis on human checkpoints and curated model choices underlines that this launch is squarely targeted at those enterprise concerns.
Paul Balo is the founder of TechBooky and a highly skilled wireless communications professional with a strong background in cloud computing, offering extensive experience in designing, implementing, and managing wireless communication systems.
We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.
