Am presenting using cursor, v0 and a claude (free account). Along with a few local ai tools like ollama, llama, and openweb ui ~ but am struggling. My background includes, founder level, enerprise mangement and design. Have degree in engineering but zero experience coding and will never be a “real” coder. Cursor, v0 and Claude.ai make it possible to create strong protoypes and have made decent progress ~ but not production grade.

Am facing platform decisions and a commitment to practices including; TDD, etc., Projects are website/ app based and include tool building and two B2B projects. Technology includes; Python, fastAPI, react.js, typescript and Sqlite.

“Claude Code” seems potentially lethal at my skill level; it moves too fast and on too many fronts and am not equipped to evaluate its production nor contain and protect my machine. A Claude Pro subscription “should” be sufficent; as my time is mostly invested in; writing requirements, prototying and admin. Coding is slow and limted to 2x per week at about 6 hrs per session.

Am leaning toward a subscribtion to Cursor and v0. Cursor provides access to model options include Claude. Does this choice make more sense than a dedicated claude (pro, api or claude code) account? Should other options be considered (cline, jetbrains, bolt or other tools (at my level)? Thank-you for the post and for considering my query.

]]> https://codeaholicguy.com/2026/02/28/my-experience-in-agentic-engineering/#comment-19604 Wed, 04 Mar 2026 01:20:45 +0000 http://codeaholicguy.com/?p=9854#comment-19604 !!</p> <!-- /wp:paragraph -->]]> Hi Hoang,

I am a final-year student in Computer Science in Vietnam, I have been impressive with your sharing and perspectives about Software Engineering and Computer Science several years, about both techniques and our mindset. This is the first time I leave a comment in your post because I really want your opinion about my case.

Fortunately, I had a little “dream” moment before AI to develop my fundamental skills in engineering in 2022 without caring about agents or anything of AI. However, two last years become a more difficult period for us, the beginners, because people confirm that AI is an invitable skill for SE and we cannot survive without learning how to use them. I have to change to adapt the market requirements but it leave me a question.

“How can CS students learn in this era? How can they handle both fundamental knowledge and market required skills in a limited time?”

Apparently, AI is necessary, but for engineers with many YOE, they can use it confidently while it can make students with limited experience pass basic knowledge such as OS, performance, algorithms, etc. by forcing them to follow market’s trends.

The conflict is that a student or a fresher is put in a massive requirements of the market just because a reason “You already have AI so you should know this” while they do not have enough time to creat solid knowledge that will create a new generation lacking of understanding about fundamental knowledge compared to generations before AI (before 2020, I guess that).

Your sharing is really valuable and helpful for me to shape my career path. I genuinely hope a response from you, honestly.

Thanks for your consideration, have a good day 😊😊😊!!

It is not really about the tool. It is about how you instruct the model to think and structure the technical design so it becomes executable.

For me, the key is forcing the model to reason clearly and define scope, constraints, architecture, APIs, edge cases, rollout plan, and break everything into implementable steps. If you do that well, most strong models can produce a solid technical solution.

Personally, I prefer Opus for this phase because it handles deeper reasoning and trade-offs better. But regardless of the tool, I usually run the design through AI DevKit review to challenge assumptions, clarify ambiguities, and iterate with the model until the document reaches a stage where engineering can execute without guessing.

Cảm ơn em, chúc mừng năm mới nha

Yes, I did, spec-kit is a good tool, I like how it handles spec generation independently. For ai-devkit, I want to go a bit deeper into the full workflow and integrate tightly with tools, so it feels native to the tools that engineers are using.