<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Roman Epifanov on Medium]]></title>
        <description><![CDATA[Stories by Roman Epifanov on Medium]]></description>
        <link>https://medium.com/@romanepifanov666?source=rss-eb8df7ccb8e3------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*N34IkexwrL2zR6ybaMdCvQ.jpeg</url>
            <title>Stories by Roman Epifanov on Medium</title>
            <link>https://medium.com/@romanepifanov666?source=rss-eb8df7ccb8e3------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 10:02:21 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@romanepifanov666/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Management books | Useful advice or just more words?]]></title>
            <link>https://medium.com/@romanepifanov666/management-books-useful-advice-or-just-more-words-848b2a3fb2cd?source=rss-eb8df7ccb8e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/848b2a3fb2cd</guid>
            <category><![CDATA[information-technology]]></category>
            <category><![CDATA[book-review]]></category>
            <category><![CDATA[managment]]></category>
            <category><![CDATA[books]]></category>
            <category><![CDATA[teamwork]]></category>
            <dc:creator><![CDATA[Roman Epifanov]]></dc:creator>
            <pubDate>Tue, 27 Oct 2026 12:01:02 GMT</pubDate>
            <atom:updated>2026-10-27T12:01:02.506Z</atom:updated>
            <content:encoded><![CDATA[<p>Management books | Useful advice or just more words?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lsJa1W9eks2Z7zeDBIOc8g.jpeg" /></figure><p>Because of the AI shift in the IT industry, I decided to spend more time improving my soft skills. One area that has interested me from the beginning is managing teams, people, and companies. So I started looking online for answers, and I was surprised by how difficult it was to find a good book with meaningful information. It feels as though I missed a whole shift in the book industry. These days, it seems that anyone who has spent even a couple of years in Big Tech decides to create “content” in the form of videos, posts, or even books. In reality, these sources often contain 99% words and only 1% genuinely useful advice or tactics for handling specific situations. Or worse, they’re AI slop published as books. The ratings are rigged. It feels like everyone is publishing a book except you.</p><p>Still, I got a few recommendations from my coworkers, along with one book that seemed to come up in almost every interview with a prominent CEO.</p><p>The first is <a href="https://www.amazon.com/dp/109814824X?lv=shuf&amp;channelId=500&amp;plpRedirect=mhFallback"><em>Leading Efficient Engineering Teams</em></a>. Overall, the book is fine, though its price seems a bit high for the value it offers. To be honest, I didn’t find much that felt original or especially useful. Most of it is a compilation of existing writing and research, without a meaningful recipe for addressing problems that might come up on your team. For that reason, I wouldn’t recommend buying it if you have other options. For me, the ratio of useful content to everything else was about 30 to 70, with much of the book referring to previously published work or research.</p><p>The second book I read was <a href="https://www.amazon.com/dp/908338182X?lv=shuf&amp;channelId=500&amp;plpRedirect=mhFallback"><em>The Software Engineer’s Guidebook</em></a>. Although it doesn’t meet my original need, I would still recommend it to people starting a career in software or those who have been in the field for a few years. From the perspective of a junior software developer, it would have been a very good book to read ten years ago, at the start of my career. It helps explain what each person in an organization can and should do. It also provides statistics about the market and discusses how to think about the current AI shift across industries.</p><p>The last book in this article is <a href="https://www.amazon.com/dp/B015VACHOK?lv=shuf&amp;channelId=500&amp;plpRedirect=mhFallback"><em>High Output Management</em></a>. I’d say it’s the cheapest and the best of the three. When I started reading, I put bookmarks on almost every page. Partway through, I realized there was no point: every page contains useful insights and approaches that still apply today. The book helps explain why Intel succeeded at the end of the last century: people knew what they were doing and what was expected of them. Published in 1983, it describes approaches that many modern IT companies still don’t use, which is surprising. Examples include OKRs and the distinction between output and outcomes. It’s easily one of the best books I’ve read!</p><p>In general, I’d say that everything new is well-forgotten old. That seems especially true today, when people rephrase old ideas and sell them for two or three times the price without adding meaningful work of their own.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=848b2a3fb2cd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How I failed the Google interview]]></title>
            <link>https://medium.com/@romanepifanov666/how-i-failed-the-google-interview-771849b771ab?source=rss-eb8df7ccb8e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/771849b771ab</guid>
            <category><![CDATA[google]]></category>
            <category><![CDATA[interview]]></category>
            <category><![CDATA[leetcode]]></category>
            <category><![CDATA[big-tech]]></category>
            <category><![CDATA[coding-interviews]]></category>
            <dc:creator><![CDATA[Roman Epifanov]]></dc:creator>
            <pubDate>Tue, 10 Feb 2026 15:31:01 GMT</pubDate>
            <atom:updated>2026-02-10T15:31:01.629Z</atom:updated>
            <content:encoded><![CDATA[<h4>10 tips which can help you to pass</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*egXvPxxv3RnnGetnWPVawQ.png" /><figcaption>My LeetCode profile</figcaption></figure><h4>How it started</h4><p>On one of the day I received an unexpected message via LinkedIn. From the beginning, I was very suspicious, because I had sent a lot of applications via the Google career site a few years ago. But never had even a rejection letter.</p><p>After the short screening with the recruiter, I got the information about the interview process and the number of stages. There should be a minimum of 3 stages in the overall 6 interviews, based on the position.</p><h4>First round</h4><p>In the first interview, I was surprised, but it looks like I got lucky because an interviewer didn’t show up for the interview. It gave me a few additional weeks for preparation.</p><p>The Googliness interview was not a surprise for me at all. Almost every piece of information that you can find on the Internet is the truth. Also very helpful was preparing a list of situations based on the <strong>STAR </strong>technique.</p><p>At the start of January, after the Christmas break, I finally got my coding interview. The overall experience was very nice. The interviewer was eager to help, and the task itself was very easy. Anyway, I made a mistake of not checking my solution with edge test cases.</p><p>Based on the metrics, which you can find in <a href="https://www.youtube.com/watch?v=-btCHmVMHro">that video</a>, I suppose I had such results:<br>Algorithms: <strong>4</strong><br>Coding: <strong>4</strong><br>Communication: <strong>3</strong><br>Problem Solving: <strong>3</strong><br>Total: <strong>Learning Hire | Hire</strong></p><h4>Second round</h4><p>After a working week, the recruiter has connected with me and shared that I have passed to the next stage. Also, it was highlighted what I should improve. As I expected, it was the edge test cases. Fortunately, it was only one mistake. I have passed to the next stage. At that stage, due to my personal needs, I already wanted to end the process because it takes a lot of energy, and it is not guaranteed that they will have the position that I need. The main target was to relocate to a new country. I scheduled the interview for the same week as soon as possible, because I decided to end this process and not waste time on it.</p><p>In the first interview, from the beginning, the interview started very strangely. The guy was cold, and there was a feeling that he didn’t want to be here. Under the pressure from his side, it was also difficult to understand what he wanted because he was not a native speaker. In addition, I noticed that on the 3rd follow-up, he didn’t know how to ask it properly and started changing the subject of the task. In the end, the performance was below average. I have failed to give a proper time complexity for the algorithm. <em>(In the feedback, I got information that I don’t know the data structure, which, in my opinion, is hilarious, because the solution was completed without mistakes.)</em></p><p>Algorithms: <strong>2</strong><br>Coding: <strong>4</strong><br>Communication: <strong>1</strong><br>Problem Solving: <strong>3</strong><br>Total: <strong>Learning Not Hire</strong></p><p>On the second interview, it felt fine; the problem itself was linear. But I have solved it in a non-optimal way. Also shared the time complexity with a mistake and got a hint to use specific tests. Overall, I would say it was an average performance.</p><p>Algorithms: 3<br>Coding: <strong>4</strong><br>Communication: 4<br>Problem Solving: <strong>3</strong><br>Total: <strong>Learning Hire</strong></p><h4>The rejection</h4><p>After a working week, I received a call with feedback. As I expected, it was rejection. Nevertheless, it was an amazing experience and an opportunity to test my knowledge and find gaps. It was difficult to imagine because in my daily routine, there are not many challenges that the average big tech engineers are facing. I definitely will try after the freeze period.</p><h4>What can help you</h4><p>To tell the truth, what I read in the forums, such as Reddit, Blind, and LinkedIn, was not helping at all. And based on what I got on my interview, it was the opposite.</p><ol><li>Be morally prepared to spend at least a few hours every day on training, and parallelly you need to work. And you need to do it for at least a few months.</li><li>You have the possibility to schedule interviews on separate weeks, which can help you to prepare and have a break.</li><li>Starting from the code docs, I would say more accrued implementation had <em>LeetCode. </em>And yes, it is not Google Docs or Microsoft Office Word.</li><li>The tasks were bound to the real-world problems. They were not like most of the <em>LeetCode </em>problems. 3 out of 3 tasks I need to implement some classes that were doing specific business logic by utilising proper data structures and algorithms.</li><li>Also, be prepared that your interviewer is not a native speaker and can have a bad mood; try to be clear, polite, and professional. Also, there was a recommendation: if you see it is not working for you, you must contact the recruiter immediately and share the situation, and there is a chance they will reschedule the interview.</li><li>After solving 222 problems, I could say that the <em>LeetCode </em>problem could distract you because there, you have tunnel vision on the problem, the description is fully written, and you have a lot of test cases. No timer and destruction when you need to explain everything to the interviewer.</li><li>Also, a very positive thing about the interview: you are kind of working in a pair with a colleague. Although sometimes it can be difficult to explain everything, sometimes the discussion can lead to some hints. In addition to that, you can observe the mimic of the interviewer; it may help you to see if you are going in the proper direction.</li><li>Checking preparation materials always can help, specifically for round two; they shared all possible data structures and algorithms, which you need to know to pass the stage. You will be able to see that some of them are highlighted and repeated several times.</li><li>Exercise the time complexity and memory complexity calculation in difficult solutions when you have multiple parameters and aspects that can distract you.</li><li><a href="https://testing.googleblog.com/">The little article</a> on how to test the solution properly</li></ol><blockquote><strong>Disclaimer:</strong><br>This is my personal experience with the Google interview process. Results and experiences may vary. This content is not affiliated with or endorsed by Google.</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=771849b771ab" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cursor vs Copilot my experience]]></title>
            <link>https://medium.com/@romanepifanov666/cursor-vs-copilot-my-experience-bd61f776a41e?source=rss-eb8df7ccb8e3------2</link>
            <guid isPermaLink="false">https://medium.com/p/bd61f776a41e</guid>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[github-copilot]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[cursor]]></category>
            <dc:creator><![CDATA[Roman Epifanov]]></dc:creator>
            <pubDate>Sat, 06 Sep 2025 12:35:52 GMT</pubDate>
            <atom:updated>2025-09-06T12:35:52.951Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*1a8dALTNxpFEyNzc" /><figcaption>Photo by <a href="https://unsplash.com/@pockethustle?utm_source=medium&amp;utm_medium=referral">TheMarketingHustle.com</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><h3>Preambular</h3><p>A few words about me, because I am new here. I am a software engineer with more than 7 years of experience in IT and 4 years in enterprise. I am mostly focused on backend (C#) but also worked with frontend (JS/TS). There won’t be any big words or live coding examples. I just want to share my experience with both tools as a person who tried them in private projects and in enterprise. I will share my experience of using Cursor and Copilot with Claude Sonnet.</p><h3>My experience</h3><p>To tell the truth, my first contact with the Cursor was very skeptical; from the beginning I was thinking it was another tool that was born in this “AI” bubble. When anyone can add in its company report the word “AI”, and investors will buy its stocks or invest in the company no matter what. But when I started my work, I was a little bit surprised how easy it is to work with that if we compare it with Copilot. I think a huge game changer here is the context window and that the IDE can easily help LLM to find proper context, also the<a href="http://spec.md/"> spec.md</a> file and Cursor’s rules. Also, I would recommend reading about what is under the hood of the Cursor.</p><h4>Cursor</h4><p>From my perspective, it is a very raw IDE tool, which has a lot of disadvantages if we compare it with such IDEs as VS 2022 or Rider. Sometimes there are some bugs that I never saw in the VS Code itself. This is my list of what I have noticed.</p><ol><li>The huge problem for me right now is it is impossible to easily integrate with server languages such as C# or Java; every time you need to switch back to VS 2022 in order to recheck the code, build it, and execute tests in the test viewer.</li><li>Sometimes you need to add these rules in the context manually; sometimes the IDE can add them automatically even if you are always on.</li><li>Faced hallucinations of LLM with primary constructor; sometimes it tries to rewrite them to basic ones if you are using Cntrl+K.</li><li>It was really strange, but there is a new bug that was not in the VS Code. Sometimes it is difficult to check long lines; it always hovers over it with a tooltip.</li><li>Sometimes it changes code without allowance; for example, polished and ready code was again regenerated fully. To tell the truth, in Copilot it was even worse.</li><li>Because there is no metadata information for some C# classes, the Cursor can’t validate contracts.</li><li>I don’t know what happened; it was a limitation of learning data or something else, but I had faced hallucinations of LLM with a primary constructor due to utilizing cntr+k.</li><li>Sometimes I caught myself on the idea of not checking code a few times. I think it is harmful.</li><li>Also, there were several times when an agent was frozen on command execution.</li></ol><h4>Copilot</h4><p>The main problem with Copilot for me is the context window and how it works with some files. However, after the last try of Copilot utilization on a daily basis after 3 months, I little changed my mind, because at the start of 2025, with full context, it couldn’t understand the logic of the classes contracts and was generating very poor unit tests. But compared with the Cursor now, there is still a very big gap between how the IDE works with Agent.</p><h3>Conclusion</h3><p>In general, the tool itself is very raw, similar to an IDE for the backend. But I liked the agent for the possibility to use context and react to the code changes/test errors. Overall, I would say it improved my performance up to 30% by initializing base classes (mechanical work of copying and replacement). And only left the business logic changes.</p><p>I recommend using a Cursor only for senior engineers in order to prevent performance and knowledge degradation in the long run for the developers. I would recommend reading the article from Microsoft.</p><p>Not only that, but I tried to avoid vibe coding as much as possible; nevertheless, I caught myself on the idea of not checking code a few times. Furthermore, I think it is dangerous and can lead the user to the situation when you start blindly trusting the Cursor. Also, we need to take into account that the IDE itself is a very raw product, and I would love to see the Cursor agent integrated with VS2022 or VS Code Rider.</p><p>With few words of potential, it is very easy to create an MVP with a different OOP language or for a different platform. You don’t need to hire a new developer who has been working with a specific platform or language for this purpose. Reading opportunity for startups to start projects that were not profitable.</p><p>Nowadays, when the double is still floating, there are a lot of buzzy worlds, but we need to understand what is really useful. Definitely the LLMs, agents, and integrated LLMs into IDE are here, and they won’t disappear, and we need to learn how to work with them. Of course there is a huge risk of learning material degradation, and the GPT-5 has proved it. It looks like we are limited right now by the processing power, cost of electricity, and other infrastructure problems that humans need to solve before discovering real AI. Although there are still open problems and questions that the IT community needs to solve, such as</p><ol><li>Problem with learning for new comers.</li><li>Will we have a new generation of programmers if they are afraid to spend five or even six years in university to become an intern?</li><li>How to deal with vibe coding and guys who just blindly copy code and shout they are programmers?</li><li>Limitation of learning material for LLMs?</li></ol><h4>Links:</h4><ol><li><a href="https://www.microsoft.com/en-us/research/wp-content/uploads/2025/01/lee_2025_ai_critical_thinking_survey.pdf">The Impact of Generative AI on Critical Thinking</a></li><li><a href="https://www.dol.gov/agencies/wb/data/high-demand-occupations">High demand occupations by US Department of labor</a></li><li><a href="https://roman.pt/posts/cursor-under-the-hood/">The cursor under the hood</a></li><li><a href="https://www.builder.io/blog/cursor-vs-github-copilot">Cursor vs Copilot</a></li><li><a href="https://medium.com/realworld-ai-use-cases/is-cursor-better-than-vs-code-with-copilot-absolutely-and-its-not-close-180b08d163f8">Cursor better vs Copilot</a></li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bd61f776a41e" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>