There's no doubt that AI coding tools are improving rapidly. Andrej Karpathy, co-founder of Open AI and former AI director at Tesla - the person who coined the phrase vibe coding - has admitted that his growing reliance on them means his own ability to write code manually is beginning to atrophy. He even credits AI for removing the drudgery from programming, making it more fun.
But the technology is not standing still. New capabilities are being introduced at a rapid pace, including sophisticated AI-assisted automation to support software engineers and architects working in enterprise teams.
One of the most exciting in this area is Project Bob, which IBM unveiled in preview in October last year. Billed as an AI-first integrated development environment (IDE), Bob is built as a fork off VS Code, it lives inside the IDE, making it easy to ask, design, debug, and refactor without breaking flow.
In this article, I explore what IBM and others are saying about Project Bob.
Going beyond AI coding
Most importantly, IBM emphasizes that Bob is designed to go beyond the capabilities of a simple AI coding tool that provides code suggestions or autocomplete facilities. Rather than being IBM's response to AI coding assistants like Cursor AI or GitHub Copilot, Bob stands out through its clear, enterprise-focused approach.
Yes, it uses AI to help generate code and responds to natural language prompts, but it also uses AI to help developers address enterprise tasks, like moving from Java 7 to Java 18 or becoming RedRAMP compliant. At the same time, it's able to understand enterprise architecture, security requirements, and compliance obligations. It's been described as an AI-powered software development partner that understands your intent, repo and security standards.
Agentic architecture to tackle enterprise challenges
Project Bob comes within an agentic architecture that enables AI agents to help developers address the complex challenges in real-world enterprise software development lifecycles. It supports agentic workflows that that break down complex developer tasks and coordinate specialized agents across coding, testing, documentation and pipelines.
Two important agentic modes that developers can switch between, depending on where they are in the development cycle, are Architect (or Plan) Mode and Code Mode.
Architect Mode allows developers to step back and receive high-level assistance before they get into implementation details. It helps plan tasks, analyze requirements and design implementation steps.
Code Mode, as the name suggests, provides more hands-on coding help to help developers iterate and work faster. And all the time Bob is aware of the entire folder structure, the active terminal and hidden dependencies.
Other modes include an agentic review mode which understands developer intent, flags issues early and explains its reasoning.
Alongside IBM's own Granite AI models, Bob can switch between industry-leading LLMs, including Anthropic Claude, Mistral AI and Llama. Crucially, it automatically selects the right model in real time, balancing accuracy, latency and cost.
Understanding the enterprise
As you'd expect from IBM, Project Bob is built to understand and support enterprise platforms, including IBM Z and IBM i. It's optimized to work with heritage programming languages like COBOL, PL/I, Assembler, REXX, JCL and RPG and to explain, refactor and generate code for systems that rely on them.
Bob is designed to tackle the everyday challenges that enterprise developers face. Like automating Java upgrades and supporting major framework migrations and multi-step refactoring projects, all the time being context-aware across large-scale enterprise codebases running to millions of lines of code.
Security is built-in
While many AI assistants like GitHub Copilot or Cursor AI are cloud-based, Project Bob can be deployed on premises or within an enterprise's own cloud infrastructure to meet strict data residency and governance requirements. It helps ensure that the enterprise's sensitive intellectual property and confidential data never leaves their controlled environment.
Moreover, Bob understands the standards and compliance requirements of organizations operating in highly regulated sectors, with built-in expertise for FedRAMP, HIPAA and PCI to help development teams deliver secure, production-ready code.
AI-powered secrets detection is built in to ensure code is continuously reviewed for vulnerabilities and credentials leaks as developers write it, rather than having to be checked afterwards. Similarly, built-in Semgrep security scans and fix suggestions allow security issues to be addressed without developers having to leave the IDE.
Building AI-driven workflows
Project Bob's agentic capabilities include those that help developers plan projects, write and rewrite code as well as identify and fix bugs and write tests to ensure it works.
On top of this, development teams can also use Bob as an AI development partner to build their own AI-driven workflows. Using watsonx Orchestrate (IBM's platform for deploying and managing AI agents) and Model Context Protocol (MCP), they can create agents that access external tools and integrate them directly into development workflows (for example, this tutorial that discusses how Bob can set up a Python project, build and validate MCP tools, deploy the tools and create an agent that uses these tools in watsonx Orchestrate).
The proof of the pudding is in the eating
According to IBM, Bob represents the next evolution of software engineering, helping teams modernize legacy systems, build new applications and maintain enterprise-grade standards with ease.
And it appears to be driving impressive gains already. In line with IBM's client zero strategy, Bob has been proven to work internally before being introduced to the world. Over 6,000 IBM developers have been using Bob in areas such as modernization, security and new-app development projects, delivering average productivity gains of 45%.