How fast can we go with AI development?

This morning, I read an announcement by one of the companies I work with that they have matured the AI enterprise coding internally. This statement did not surprise me too much, as I had the chance to meet some of the technical team and the CEO - all bright, smart, and goal-oriented people. At the same time, I am using a very similar approach to develop most of the software solutions we offer at ITWS, with the difference that they work on existing, well-known software, while we craft new, often specific projects. 

But this is not that important - the sense I felt when I read this was that technology advances homogeneously, and while adoption may lag for some companies, others that are flexible and open-minded can excel at pulling ahead. So I decided to make a retrospective of the last few years and see how things evolved.

[2018 - 2020] This was the year I first encountered AI in depth and trained a model. Back then, I worked with another team to develop software for internal and client use, leveraging a cutting-edge self-trained model trained on a large dataset. Still, it was actually much, much smaller than the ones we faced with the rise of OpenAI and their models. Respectively, the constant effort to develop a platform with only a few hardcore developers at hand lasted almost two years and was fascinating. We teamed up, shipping feature after feature - testing, debugging and most of all being at the top of the wave, coming up with something unique in 2020 - tremendous work. As the head of R&D for that team, we were always on the edge of innovation, and the product was worth it! But development was a large and costly initiative.

Around the same time, I was personally struggling to push further a project that was a lot on my heart - the GoutBye concept to help gouty arthritis patients. The design was ready wayyy back in 2015, and since then, I've attempted to find someone to build an MVP. Main issue: I wanted the app's stack to be Vue 3, not React / React Native.

Yes, since day one of modern frameworking, I've been using Angular way back in 2015 (made a PhoneGap hybrid app with Angular for a US client). Then Google changed the Angular model after version 1.5, and it became awful (personal opinion). A few people tried to convince me how cool React is, but it is actually a technology I disliked at first sight. The way CSS and HTML were injected into their templates really breaks something in me. Thus, to this day, I don't use React and avoid projects with it. Yeah, it is just JavaScript with some fancy extras yet - no. And every developer I came across on Freelancer.com or a similar site proposed doing it their way, most often React. The prices were just wow, but the projects they showed me were, in my opinion, below average. So I decided to go myself and develop it personally, the way I want.

[2021 - 2022] My focus towards the desire to learn a good framework that I like and can operate fast with, led me to adopt Vue3 as a flexible, robust and fast framework. In 2020, after another round of desperation for GoutBye and an offer to build it on Flutter, I started learning Vue3 from scratch. Using the GSAP animation library and Vue components, I quickly realized this is love at first sight. I started to build SPAs for different clients - games, small platforms with serverless architecture. MongoDB Atlas with the ex-Stitch API was something super cool. I was happily flexing my muscles and jumped straight into Capacitor + Vue3 app architecture, which was great. Day after day, I spent some of my time preparing assets and incorporating design and other features into the new mobile app. Registered App Store / Google Play accounts and slowly started to see the light at the end of the tunnel. MongoDB was very flexible for authentication, security and as a document DB for this app, and the HIPAA compliance gave me peace of mind. All was joy and fun.

At the end of 2022, OpenAI released the first LLM, and I remember I was blown away. The capabilities this miracle was trained on were outstanding! 

[2023]  Stepping into 2023, I solved with OpenAI one crucial functionality that I really needed for my app - AI-assisted recipes. The concept is long, and I will cover it in another shard, but the general idea was to train my own AI model to help with healthy, gout-friendly recipes. The large language models solved this for me. There wasn't much assisted coding, and I entirely wrote most of the projects I did in 2022 and 2023. I remember slowly shifting away from Stack Overflow for advice, as OpenAI has consumed it along with Vue documentation, and I could find answers to problems much more quickly than by searching the net.

One old client and friend from the Prezi era approached me in the summer of 2023 to handle a complex video production project with extensive creative work. I jumped on the ship and really enjoyed the project. As a result, the balance between the never-ending agency projects and this specific huge task (literally a year and a half of effort) took its toll on the development of my app, and in general, I put these aside. An additional heartbreaking moment for the GoutBye app was that, in March 2023, MongoDB Realm was put to sleep, killing my authentication and a plethora of serverless functions on which the app was built.

[2024] However, in the summer of 2024, I learned about the possibility of presenting the app on the Japanese market through the IANMSP initiative and a trade mission for IT companies from Bulgaria. Oh boy, I was approved and soooo excited! 

There was a small problem, though. The app was not working - thank you, MongoDB! And another client came to me around the same time and wanted a specific system that would allow them to proactively promote activities, including games with receipt registration, SMS notifications, and complex draw winner logic. I think this was the key moment I rearealizedhis project is great for levelling up and getting back in the game. I started exploring Nuxt to build a complex system with an integrated frontend and backend, and I was literally blown away! Nuxt was mature, solid, fast, serverless, kick ass in so many different ways! With all the excitement, I was still on the risky side and a bit nervous. I had from EOM August to mid-September to develop a complex system, and then just a couple of weeks to prepare for Japan. I was jumping into Nuxt, a Vue framework with its quirks, but I didn't have a working app because of MongoDB's authentication and serverless functions, both now deprecated. 

I remember then I told myself - c'mon, you can do it - use any leverage possible to pull ahead. I bought the highest Cursor plan and started to adapt it heavily - learning and coding at the same time. I also switched the GoutBye app from Vue 3 to Nuxt, linked it to Supabase authentication, and decided to go all-in on an on-device database, implementing secure SQLite. This was the a-ha moment, understanding that having a clear vision, a clear goal, and enough programming and project management experience can untie my hands. I built a comprehensive, secure system for my clients, handling thousands of registrations in September and October. I went to Japan after a wisdom tooth operation and migrated the app to Nuxt 3.

[2025], Realising that this approach is really powerful, I was leveraging and comparing Cursor, Windsurf, Cline, Co-pilot, and other exotic options to enjoy coding and speed up development on my end. This approach worked well for several projects, and my focus since March has shifted to on-premises solutions for process automation and AI integration. I was again a bit out of the development sphere. Nuxt4 matured, and I updated my apps. Still, in the summer, I started talking with a couple of friends in SEO and marketing optimisation (GTM/GA4/BigQuery) about a project they had in mind. Claude Code was released in the spring of 2025, but it came into my focus around September. I bought an account to accelerate the development of my friends' startup idea, and I was stunned to find that it actually allows agents to be orchestrated to develop complex features. 

I was immediately on board, trying to build a programmatic, agile process from idea to re-realisation. After a few rounds of practice around November, I had the chance to talk with my friends from the company I mentioned at the beginning of this post and get onboarded to their Claude Code process. One thing they said stuck in my head: "Why are you trying to guardrail so heavily the AI model? It is intelligent enough to fulfil a simple task." I switched the very same night and the day after into building Nuxt-oriented SAT (Synthetic Agentic Team), around this idea and with much greater focus on getting a Feature Request right, and laying down one consistent documentation for FRs and their tasks and reviews, than heavily controlling, and making semi - automated JS validators. 

The effect: for less than a week, I pushed in my evening features with complexity that would take me two weeks to a month. I rearranged, optimised, and expanded the SQLite database for my GoutBye app, finalising some missing functionality with no problems or security issues. And pushed a rather complex task for my SEO/Marketing guys overnight - crazy December.

[2026] Today is the 22nd of January. I have developed, tested, and completed more work on my projects that would otherwise take me up to the end of this quarter. The code is consistent, up to date, secure and tested. I am truly hyped that, by continuing to optimise my process, I will soon be able to ship solutions much faster than most of my competitors, and not only - the era of self-developed solutions is now here, as I can build something, automate the content generation, and productize the system for consistency. 

To wrap this up. You can go incredibly fast with the development if you know what you're doing. Fixing the stack, following best practices, making very consistent and consequential requests, and, of course, having a very solid testing and review process unlock the true potential of the new CLI tools. I have yet to test Codex and similar tools in depth, but Claude Code is fantastic, and for the price I am paying, it more than justifies the investment.

[Future] I plan to write another shard on the topic but It will be around March when I fully leverage Claude Code and the upcoming AI hacks that come to town. 🫡