All posts

Two GitHub certifications in 24 hours: How I studied for GH-900 and GH-300

Illustration of the GH-900 and GH-300 certifications completed within 24 hours

GitHub gave me 90 days to pass the Foundations (GH-900) and Copilot (GH-300) certifications as part of onboarding. I finished both in my second week, less than 24 hours apart, after roughly a week of focused study around work and onboarding.

I knew about the requirement from day one and started almost immediately, before it could become a day-89 problem. The plan was to take the exams one at a time. After passing Foundations, though, I’d already done enough Copilot prep that another practice assessment suggested I was ready. The idea of finishing both within 24 hours appealed to me, so I booked Copilot for the next day.

I came to GitHub with years of experience in technology, software development, Git, GitHub, and AI-assisted development. Studying gave me a reason to check the parts of the platform that daily work doesn’t necessarily teach you. It also helped me validate what I knew while building credibility at a new company. Passing an exam doesn’t make someone an expert or replace experience, but I was pleased to have filled some gaps and finished the requirement early.

What I needed to know for each exam

Foundations covers Git and GitHub basics, repositories, collaboration, modern development practices, Projects, security and administration, and the community and open-source ecosystem. Branches, commits, pull requests, issues, repository settings, and the usual collaboration workflow were familiar. The amount of ground covered was the challenge.

Knowing Git doesn’t mean knowing the whole GitHub platform. It’s possible to work in pull requests every day without touching Projects, to use permissions without knowing every role distinction, or to use open-source software without much reason to learn InnerSource, repository templates, Marketplace, and community features. I wanted to check that broader knowledge alongside the parts I already used.

GitHub permissions cheat sheet used during Foundations prep

I used this cheat sheet to get more precise about permissions I already understood in practice.

Copilot is narrower but goes deeper into the product. I’d used it and other AI tools before studying, which helped, but generating code is only part of the exam. It covers responsible AI, Copilot in the IDE and command line, agent workflows, prompt engineering, context, testing, developer productivity, architecture and data handling, privacy, content exclusions, safeguards, and organizational controls.

Newer capabilities are there too: Agent Mode, Copilot Edits, MCP, agent sessions, sub-agents, Spaces, code review, reusable instructions and prompt files, and policy and audit controls. I needed to understand how context is built, how data is handled, how the different Copilot experiences fit together, and what administrators can control. Some of that is hard to pick up through ordinary use if you don’t administer Copilot for an organization.

I planned my study around those differences and the official objectives:

The resources I used

I started with Microsoft Learn for both exams, worked through the official material and objectives, and took practice assessments for an early baseline. This helped me work out where to spend my time without trying to memorize every paragraph. Returning to those assessments later showed whether the studying was helping. I’d start the same way again, before spending much time in question banks, and take the first assessment before I felt ready.

LinkedIn Learning and the official video training mostly ran in the background while I did other things. I wasn’t treating every video as a class that required notes. Listening reinforced concepts I’d met in Microsoft Learn and kept the terminology familiar without needing another full study session. I wouldn’t prepare with video alone: understanding an explanation while listening didn’t always mean I could answer a question about it later.

At my computer, my favorite way to drill was the Certification Study tool, built by my colleague Greg Gingo with help from Katie May. It presents one question at a time, grades the answer, and puts missed questions back in the queue. I couldn’t read a correction, think “yeah, that makes sense,” and forget about it; eventually I had to answer it correctly. It saves progress between sessions and tracks first-attempt accuracy separately from overall accuracy. The source is on GitHub.

I also used the community-maintained ghcertified question banks extensively. They provided more practice with plausible multiple-choice answers, especially for Foundations’ many product details. With Copilot, I had to remember that community banks can lag behind both the product and the certification blueprint. Even a large bank can miss topics, so I checked its coverage against the official objectives, particularly for newer features.

ChatGPT and Muse fit into the shorter gaps in my day. I already used both, and there was no planned division of work between them; I picked whichever suited the moment. Five or ten minutes with my phone was enough for another set of questions. Being able to start with so little setup kept me studying without having to find a free hour. At my desk I could return to Microsoft Learn or Greg’s tool.

I could also ask the AI tools for more questions on a weak topic or change the format when an explanation wasn’t sticking. ChatGPT made custom infographics and visual cheat sheets, while Muse helped me condense material into cram sheets near the exams. The resources served different purposes, so I got more out of combining them than trying to choose a single “best” one.

First page of a GitHub Copilot certification cram sheet

Near the exam, I used sheets like this to jog my memory about material I’d already studied.

Letting mistakes decide what to study next

In previous certification prep, I’d worked through modules, videos, and reading in order until I’d completed everything. That spent too much time on material I already knew. This time, once I understood the basics, I moved quickly into questions, looked for clusters of mistakes, and studied those topics before testing them again. Several misses on permissions meant more permissions practice; uncertainty about Copilot content exclusions or administration sent me back to those subjects.

Reading about repository roles can feel convincing until you’re choosing between four plausible permission scenarios. Answering questions tested whether I could retrieve and apply the material, which helped me more than rereading it. Greg’s tool brought back missed questions automatically; ChatGPT and Muse let me try new questions on the same concepts, or switch to a visual or condensed explanation if I still wasn’t getting them.

One Copilot quiz was especially useful. I’d done well on broad practice material, then scored badly on a small set deliberately focused on my weaker areas. The lower score made sense: this quiz left out the topics I already knew. It gave me a short list to work on, which was more useful than protecting a good average with another familiar test. Once I was doing well overall, ten hard questions on a weak subject could tell me more than another broad assessment.

The study feedback loop: learn, test, find gaps, target gaps, retest

I wanted to find the surprises before exam day. I expect to use the same approach outside certification prep, to catch the gaps that routine work can leave.

Some early AI-generated quizzes had such obvious wrong answers that I could pass by elimination without knowing the subject well. I asked for harder questions, better distractors, and more multi-select answers. Multi-select practice helped particularly with Foundations, where deciding whether two or three similar statements are correct demands more precision than spotting one likely answer.

Deciding when to book

If I were consistently scoring around 70%, I’d keep studying and investigate the missing 30%. Mistakes spread across every domain would call for more general review; mistakes concentrated in two or three areas would call for focused practice and another test. Permissions, Projects, administration, content exclusions, prompting, and security are useful categories to check. The pattern tells me more than the percentage alone, and it doesn’t require starting the whole course again.

Booking was easier because I could take the exams at no cost as a GitHub employee. After passing Foundations and doing well in Copilot practice, I was comfortable scheduling GH-300 immediately. If I’d been paying for each attempt, I probably would have waited longer. I’d consider whether the gaps were shrinking, how comfortable I was with those that remained, and the cost of another attempt, rather than treating one practice score as a universal threshold. I didn’t need to know everything, but I did need to know where I was still weak.

The last 48 hours

I’d use the same process again, more deliberately from the start. In the final couple of days, I’d concentrate on weak areas instead of adding more material. Once I could answer a troublesome topic consistently, I’d move on rather than review it for the tenth time. The visual summaries and condensed sheets were useful here because I could scan them for reminders without relearning everything.

I wouldn’t start a newly discovered course the night before the exam. By then I’d rather do a few targeted questions, review the short notes, and sleep. I’d take a rested attempt with 95% of my planned studying done over another late hour trying to cram in an obscure detail. Eventually the extra study helps less than being able to think clearly.

Taking the exams

Both exams were online. Scheduling and check-in can make an exam feel imposing, but once the questions appeared, the work felt familiar. A week of practice had made me used to reading carefully, weighing plausible answers, and taking time to check wording and multi-select questions. There were few surprises.

I finished each in under 30 minutes and passed both comfortably. I wasn’t trying to be fast; I could move on when I knew an answer and take time when I wasn’t sure. Repetition helped more than trying to memorize every possible fact. The exams felt much like the practice sets I’d been working through, which made finishing them so early in my new job especially satisfying.