With Bob, IBM is diving into an already saturated market of AI assistants for software development. Bob can draw on decades of experience to distinguish itself.
There is no shortage of artificial coding assistants today. Anyone wanting to develop an application only needs to ask GitHub Copilot, Claude Code, or ChatGPT Codex. Recently, that has also become possible with Bob, IBM’s coding assistant that was released to the world a few months ago.
Marty van de Korput, CTO for IBM in Belgium and Luxembourg, and Bilal Soussi, Technical Sales, invite us to the headquarters in Evere, located across from NATO, to introduce Bob. Don’t just call Bob a ‘coding assistant’, van de Korput immediately makes clear. “Bob is a development partner for the full cycle, from planning to testing, execution, and maintenance. We are evolving from AI-driven coding to delivery.”
Polyglot
The road to Bob began back in 2021, van de Korput outlines. “We started with Project CodeNET for code autocompletion. That evolved in 2023 into independent coding with watsonx Coding Assistant, and with Bob, we now want to support the entire trajectory. Internally, Bob has already proven itself at scale: it is already being used by 100,000 IBMers.”
Bob, whose name has no link to the well-known cartoon Bob the Builder, aims to distinguish itself with a deep knowledge of more than a hundred programming languages, including languages that are less common today. Van de Korput: “60 percent of coding work consists of modernization. Knowledge of old programming languages is disappearing due to a shortage of developers. AI can help to understand the code.”
With Bob, IBM wants to bridge the gap between ‘old’ technologies like mainframes and new technologies like AI agents. “The mainframe market remains stable. Companies don’t want to move, especially in the context of sovereignty, but they will have to modernize mainframes. Translating Cobol is not that difficult in itself, but continuing to use the platform optimally is much more important.”
Commoditization of code
The role of programming languages has changed with the arrival of AI-driven tools, Soussi notes. “Code is becoming commoditized. Even with limited knowledge of code, you can achieve a lot. We need to extrapolate skills from the work itself. The choice of programming language is no longer decisive, just like the choice of LLMs, because models evolve quickly. It’s more about bringing the right tools to developers and focusing on business outcomes.”
“Programming languages aren’t going to disappear suddenly; they will continue to support applications,” van de Korput chimes in. “But for general applications, traditional languages will become less important. There is a misconception between code and modernization. A different programming language should also enable you to utilize the benefits of a platform.”
Paying in Bobcoins
Soussi is eager to show Bob. The tool is accessible through various licensing formulas, for large enterprises and smaller companies. The different modes range from a simple chat interface to a planning mode and an agent mode that also executes the work.
To prevent your AI bill from skyrocketing or your credits from running out immediately, Bob promises transparency regarding costs. Each license includes a certain amount of ‘Bobcoins’, where one Bobcoin corresponds to approximately 50 euro cents. You don’t buy bread with them at the supermarket, but tokens. After every prompt, Bob calculates the cost, which can also be monitored via a dashboard.
“The costs of AI tools are a major point of discussion for large companies today. Tokens are difficult to quantify. You see where the money is going. There is no ‘black box’,” Soussi explains.
“The multi-agent model should also make Bob more cost-effective than competing tools. Bob reasons which model is required for each query. This allows for smarter model usage and cost optimization across different models,” van de Korput adds.
Human at the controls
What stands out during the demo: there is almost no code to be seen on the screen. You have to search through the menus to view code and possibly edit it manually. This fits into a broader evolution of the work of software developers, Soussi believes.
“As a developer, you will hardly sit in front of the code anymore. You work at an architectural level. You draw up a plan with Bob, and Bob then executes that plan.” It remains important, however, that you as a developer stay at the controls. IBM does not position itself as an apostle of vibe coding.
Soussi agrees: “I think we are starting to feel the limits of vibe coding. For a hobby project at home it’s fine, but at a corporate level, there are more problems bringing it into production. The technology is good but needs clear controls and governance. If you just roll out AI, you create a mess.”
Bob has only just been launched, so Soussi and van de Korput are not yet making any grand statements about the rollout. “On average, we achieve a 45 percent productivity gain across the entire cycle, and in internal projects even up to 70 percent. The first layer of the software cycle will feel the impact first, but we want to go as broad as possible,” says Soussi.
“The enthusiasm is there. People who are ‘pushed’ to follow our training sessions are often the most convinced afterwards,” van de Korput concludes with a quip. IBM draws on 115 years of software experience to take on the ‘new’ generation of AI companies. Does the old dog still have some tricks up its sleeve?