# Back to Writing: My Claude Certification Week > Source: > Published: 2026-09-06 16:01:01+00:00 Four Claude certifications. Five days. Zero third-party study guides. That’s the short version of the week that pulled me back to writing after eighteen months of silence, and the long version is worth telling because of what it took to get through the hardest of the four. **In this article:** I haven’t published anything here since February 2025. That’s not a dramatic story. Family needed attention for a stretch, the kind of thing that quietly reorders your priorities without asking permission. Around the same time, work stacked up with multiple projects running in parallel, and I was spending most of my spare hours upskilling instead of writing about what I already knew. Writing was the thing that got dropped, because it was the only thing that could be dropped without anyone else noticing. Nobody was waiting on a blog post. Deadlines don’t work that way for a side project, so it kept sliding to next week, then next month, until eighteen months had gone by and I hadn’t noticed the gap until I checked my own profile. What eventually pulled me back wasn’t a resolution or a plan. It was Anthropic’s [Claude Certification Program](https://anthropic-partners.skilljar.com/page/partner-certifications), which expanded from a single exam to four in July 2026. I’ve been writing about Claude since [Claude 3 first launched](https://medium.com/gitconnected/claude-3-next-generation-ai-models-0c39410a72af), so a certification program felt like a natural reason to start again. I signed up for all four almost on impulse, more curious than committed at first, and then had to actually follow through once the study material was already in front of me. I sat all four exams within the same week in August 2026, in this order: Associate — Foundations, Developer — Foundations, Architect — Foundations, then Architect — Professional. Each one is proctored through [Pearson VUE](https://www.pearsonvue.com/us/en/anthropic.html), either online or at a test center, and each is scored on a 100 to 1,000 scale with a passing cut of 720. The fees range from $99 for Associate up to $175 for Architect Professional (I got them free because of partner network), and every credential stays valid for 12 months before it needs renewing. Stacking them back to back wasn’t part of some grand strategy. I’d already committed the study time, and I didn’t want a two week gap to let any of it go stale before the next exam. In hindsight, doing all four in one week meant the concepts overlapped in my head in a useful way. Tool design questions from the Developer exam were still fresh when I hit similar territory in Architect Foundations two days later, and that continuity probably saved me more than any single study session did. I didn’t touch third party prep courses or mock exam sites for this. I went through Anthropic’s own [Academy prep courses on Skilljar](https://anthropic-partners.skilljar.com/page/claude-certification-exam-prep-courses), and I read every official Exam Guide PDF for all four certifications cover to cover before scheduling anything, starting with the [Architect Foundations guide](https://everpath-course-content.s3-accelerate.amazonaws.com/instructor%2F6nizmqk8tpzpfjvt6qmmav7rh%2Fpublic%2F1783542750%2FClaude+Certified+Architect+%E2%80%93+Foundations+Exam+Guide.pdf) since that one turned out to matter most. Each guide lists the exact domain weightings, sample questions with explanations, and a set of hands on preparation exercises, and the guides are explicit that they are the authoritative reference for what gets tested. That approach worked because the guides read like internal test specs rather than marketing material. Each one names the topics that are explicitly out of scope, so I didn’t waste time studying things like model fine tuning or token counting algorithms that were never going to appear on the exam. I treated the “How to Prepare” and preparation exercise sections in each guide as a checklist, working through them in order rather than skimming for whatever looked interesting. If I’m honest, this was also the first time in a while that studying felt like the same kind of focused work as writing used to feel. That’s probably part of why it pulled me back into publishing again. The four certifications map to different jobs, not different skill levels of the same job. Associate is aimed at people guiding customers toward the right Claude use cases without necessarily writing code themselves. Developer is built for engineers working directly with the Claude API, Claude Code, and Model Context Protocol. The two Architect tiers sit at a different altitude entirely, testing full production system design: deployment choices, agentic architecture, evaluation strategy, and safety tradeoffs. Architect Foundations, as one concrete example, breaks down into five weighted domains. Agentic Architecture and Orchestration carries 27% of the exam. Claude Code Configuration and Workflows and Prompt Engineering and Structured Output each carry 20%. Tool Design and MCP Integration sits at 18%, and Context Management and Reliability rounds it out at 15%. That weighting alone tells you where most of your prep time should go. Over a quarter of the exam rests on multi agent orchestration and coordinator subagent design by itself, which is a heavier lift than it sounds when you’re used to thinking about single agent workflows. That Prompt Engineering and Structured Output share is worth pausing on, since it lines up closely with [an earlier piece I wrote on prompt engineering best practices](https://levelup.gitconnected.com/prompt-engineering-best-practices-ed6cac3f343d). Few-shot examples and explicit output schemas, the same ideas from that post, show up again here as tested skills rather than general advice. Out of the four, Architect Foundations gave me the most trouble, and not for a single obvious reason. Part of it was fatigue. It landed third in the week, after two exams had already worn down my focus. But most of it came down to the exam design itself. The scenarios blend developer level and architect level judgment in the same question set, so you’re rarely choosing between one obviously right answer and three obviously wrong ones. Several of the answer options are structurally similar to each other, differing only in a small mechanism: one option leans on a prompt instruction to enforce a rule, another leans on a programmatic hook or hard gate to enforce the same rule. On paper both sound like they would solve the stated problem. The difference only becomes clear when you think through which one fails silently under load, and which one fails loudly enough that someone actually notices before it causes damage. I remember sitting with a couple of those questions longer than I expected to, reading the same two options back to back, before realizing the tell was buried in a single word about reliability rather than in the scenario’s headline problem. That pattern repeated enough times across the exam that by the end I’d started deliberately looking for the word that signaled “guaranteed” versus “probably,” which is exactly the kind of instinct the exam seems designed to build. That’s a deliberate design choice on Anthropic’s part, not a flaw in the test. The official exam guide is explicit that questions are scenario based and drawn from realistic production situations, which means picking between two reasonable sounding answers is the actual skill being measured, not simple recall of a term or a tool name. It’s less “do you know what MCP stands for” and more “can you tell which failure mode you’re actually protecting against.” A quick honesty note here: the exact wording and moment I described above is a reconstruction of how that kind of question felt to work through, not a verbatim account of a specific numbered question. Anthropic’s exam content is confidential under the certification’s non disclosure terms, so no draft of this post reproduces or paraphrases specific exam items beyond what appears in the publicly available sample questions in the official exam guide. If you want this section to reflect a more specific memory of your own, that’s worth rewriting in your own words before publishing. Now that I’m back, the plan is to write more regularly about agentic architecture, enterprise integrations with Claude, and the broader shifts happening across AI as they land, rather than waiting for a big personal milestone to justify hitting publish. Some of it will come directly out of what I studied for these exams. Multi agent orchestration patterns and MCP tool design are both things I want full pieces on, not just a paragraph mentioned in passing here. There’s also a practical reason to keep writing about this specific corner of the field. The certification program itself is still young, having only existed in its current four exam form since July 2026, and the guidance around building production Claude systems is still being written in real time by the people doing it. That’s a reasonable niche for a blog that already leans toward AI and agentic architecture. If you’ve been through any of these exams yourself, I’d like to hear which one gave you the most trouble, and whether it was the content or the exam format itself that made it hard. *Note: exam fees, passing score, domain weights, and validity period cited above come directly from the sources listed. Personal details such as my study experience, timeline, and the difficulty of specific concepts are drawn from what you told me and are not independently sourced. The narrative passage about the Architect Foundations exam questions is a plausible reconstruction, not a verbatim memory, and is flagged as such in the text for you to personalize or remove before publishing.* If you enjoyed the article and wish to show your support, make sure to: [Back to Writing: My Claude Certification Week](https://pub.towardsai.net/back-to-writing-my-claude-certification-week-14107faca592) was originally published in [Towards AI](https://pub.towardsai.net) on Medium, where people are continuing the conversation by highlighting and responding to this story.