If you have seen "IBM Bobathon" mentioned in a student group chat or a careers newsletter and want to know what it actually involves before you commit an evening to it, this is the guide. No filler, just what you do, what it takes, and what you get if you pass.
What the Bobathon actually is
The IBM Bobathon is an exciting challenge built with IBM Bob, an agentic AI software development tool from IBM. "Agentic" means Bob does not just answer questions in a chat window. Point it at a codebase and it can read the code, plan a fix, write the changes, run tests, and modernise old code across an entire repository, the way a very fast, very literal junior developer would, if you supervised it closely.
The challenge itself is a single scenario called The Legacy Fix. It is a constructed training scenario, not a real company's real data, built so that everyone who attempts it faces the same problem under the same conditions. That matters more than it sounds: a fixed, shared scenario is what makes the result comparable across every student who attempts it, and it is what makes the certificate at the end mean the same thing for everyone who earns one.
The scenario: your first week as a junior engineer
You are dropped into the role of a junior engineer in your first week at a German car-parts company. An old internal program has been quietly reporting the wrong numbers, and nobody has had time to work out why. You are handed your own copy of the legacy repository and asked to sort it out.
From there, the work has four parts:
- Point an AI agent at the repository and have it read through the existing code to understand what it does and where it is likely to be wrong.
- Fix and modernise the code, using the agent to do the heavy lifting while you decide what actually needs to change.
- Catch the agent's mistakes. This is the part people underestimate. An agentic tool will confidently make changes that look right and are not, and part of the challenge is noticing when that happens instead of accepting every suggestion.
- Verify your fix using the repository's own acceptance script, the same check a real engineering team would run before merging anything, then hand in your public GitHub link.
What you actually need
A laptop, an internet connection, and roughly two to three hours. The whole challenge runs in English and is built for beginners, so you do not need prior repository experience or a computer science degree to start it. What you do need is patience to actually read the code the agent produces rather than skimming past it.
You will need an AI coding agent to do the work with. Students can use a 30-day free trial of IBM Bob for this, though the challenge itself is tool-agnostic: any capable AI coding agent works, Bob's trial is simply the fastest way to get one running if you do not already have one set up.
It also helps to know, going in, that "beginner-friendly" here means the starting point, not the standard you are held to. You are not expected to already know the legacy codebase or have used an agentic tool before. You are expected to read carefully, question what the agent hands you, and check your work against the acceptance script rather than against how confident it sounds. That distinction is the whole challenge in one sentence.
What you walk away with
Pass the challenge and you earn an IBM SkillsBuild certificate of completion. It costs nothing to attempt, but it is not handed out for showing up: about three in ten participants do not pass, based on ProoV's own grading data. That failure rate is worth sitting with for a second, because it is exactly what makes the certificate mean something. A credential that everyone gets on completion tells a recruiter nothing about your work; one that three in ten people do not clear tells them you cleared a real bar.
Alongside the SkillsBuild certificate, you also get a ProoV scorecard, a breakdown of how you performed against the grading criteria, tied to a public link anyone can open to verify it. That link is the part worth keeping. A line on your CV that says you fixed a legacy codebase is a claim. A link to a scorecard and a public repository showing exactly what you changed is proof.
How to actually start
Go to the IBM Bobathon page, read the brief so you know what the acceptance script is checking before you touch any code, and set aside an uninterrupted two to three hour block rather than starting it between lectures. If you do not already have an AI coding agent set up, the Bob free trial takes a few minutes to get running.
One habit that separates a pass from a near miss: before you accept anything the agent proposes, ask yourself what it would take to be wrong about this specific change, and check that against the acceptance script rather than against how confident the explanation sounds.
