Capgemini Exceller AI-Assisted Coding Round 2026: A Guide

The Capgemini Exceller AI-assisted coding round scores how you work with an AI assistant, not just the answer you reach. You solve a coding problem by directing the tool, and the process itself is being measured.

That reframing is the whole game. A candidate who pastes a vague request and copies the output will score below one who frames the problem, prompts with intent, and checks the result. This round is distinctive to Exceller, with almost no equivalent in other India IT drives, so a little targeted practice goes far. The full stage sequence is in the Capgemini Exceller framework guide.

What the AI-assisted coding round tests

Capgemini runs Exceller as its campus hiring route for technical graduates, described on the company’s technical graduates page. This round checks a modern skill: using an AI assistant effectively to solve coding problems quickly, rather than reaching the answer any way possible. The stated focus, per the framework shared with candidates, is the collaboration itself.

Put plainly: the assistant can write code, so the test is whether you can direct it well.

The scaffolding flow, step by step

The round follows a step-by-step logic the framework calls a scaffolding flow. You give clear, correct, and complete inputs at each step to reach the next, and the final step produces the code. A practical way to move through it:

• Read and frame: understand the problem, its inputs, outputs, and constraints before typing anything.
• State the plan: tell the assistant your approach in plain terms, not just the goal.
• Guide the build: ask for the solution in steps, giving the logic and edge cases you want handled.
• Review and correct: check the generated code, fix what is wrong, and confirm it handles the boundaries.

Skipping the framing step is the fastest way to stall, because a weak first input produces weak steps after it.

What is being evaluated (four things)

The framework names four scored dimensions. Knowing them tells you where to spend effort:

Dimension	What it means
AI literacy	Understanding the question and responding meaningfully
Prompt quality	Structured, relevant prompts over random one-liners
Problem-solving	Choosing the right approach and guiding the tool to it
Review and adapt	Checking and improving the code, not copy-pasting

Notice that two of the four are about judgement, not typing. That is deliberate.

Dos and don’ts

The framework is direct about behaviour, and it lines up with common sense:

• Do read each instruction carefully, think first, then tell the assistant your plan.
• Do explain the logic, inputs, outputs, and constraints in your prompts.
• Do review and edit the generated code before running or submitting it.
• Don’t type vague prompts like write the code with no explanation.
• Don’t copy sample values or numbers into your solution as if they were the constraints.
• Don’t rely blindly on the output or submit code you do not understand.

Why Capgemini scores the process, not the answer

The design choice here is deliberate, and it reflects how software work is changing. AI assistants can now produce working code from a clear description, so the scarce skill is no longer typing a solution from memory. It is knowing what to ask for, judging whether the result is correct, and adapting when it is not. Capgemini notes in its recruitment process guidance that it looks at how candidates approach challenges, not only what they already know. This round puts that idea into a test.

That is why a candidate who understands the problem deeply will usually beat a faster typist. If you can state the inputs, outputs, constraints, and edge cases in plain language, your prompts almost write themselves, and the assistant has enough to produce a solution you can trust. If you cannot, no amount of prompting rescues a fuzzy plan.

The review step carries more weight than most candidates expect. Generated code can look right and still mishandle an empty input or a boundary case. Reading it critically, spotting the gap, and correcting it is exactly the judgement the round is built to measure. Copy-pasting without that check is the clearest way to score low, even when the assistant did most of the work.

There is a mindset shift in all of this. You are not being tested on whether you can avoid using AI, and you are not being tested on whether AI can replace you. You are being tested on whether the two of you together produce good, correct work efficiently. That is the skill Capgemini is hiring for.

How to prepare

Practise the habit, not a script. Take a few standard problems and solve them by writing structured prompts: state the problem, the constraints, your plan, then ask for a stepwise solution and review it. Time yourself, since efficiency is part of the round. Build the reflex of reading generated code critically, which is the same instinct trained by the Capgemini debugging assessment.

A concrete drill helps more than reading about it. Take five standard problems you already know how to solve by hand, such as reversing a string, finding duplicates in an array, or checking a balanced parenthesis. For each, write a full prompt from scratch: state the problem, the inputs and outputs, the constraints, and the edge cases you want handled, then ask for a stepwise solution and review what comes back. The point is not the code, which you could write yourself. The point is training the muscle of describing a problem so precisely that the solution is almost forced. Time yourself, and by the fifth problem you will notice your prompts getting shorter and sharper.

For context on why Capgemini has moved toward AI-aware hiring at all, the 2026 pseudocode round and GenAI hiring context is a useful companion. Start by practising the framing step, because a clear problem statement is what makes every prompt after it land.