This post is for information and diary purposes. Frankly, considering that the majority of my day job is now centered around reading, reviewing, and debugging English instead of code, the last thing I want is to spend hours hammering out content that can potentially be overshadowed by AI content slop found on every major platform.
My shift
If you followed the developments of AI over 2025, you probably know the breakthrough moment came somewhere around the release of Claude Opus 4.5 at the end of November.
Let’s be real, the meaning of breakthrough in AI had a different meaning to different groups of people. For me, being a tech-focused founder who gets his hands dirty in code, this definition was, at the time, based purely on how well the model could perform technical coding tasks to effectively clone me. Moreover, nailing the coding part was exactly the challenge that excelled model providers such as Claude and created a massive influx of new users.
Up until that point, the AI used in our products was still heavy, but never outweighed the work around integration of systems, data plumbing, and daily technical tasks that consumed a good chunk of my time.
Using the majority of models prior to Opus 4.5, including previous versions of Opus itself, produced mediocre results, helping mainly in structuring UI components across a suite of products, a task quite frustrating for someone focused on data and backend systems. The generated code could not be trusted, nor could the complex data queries, as every attempt to create core functionality usually meant more work to correct later, and in many cases the results of such coding assistance would break the system.
Fast forward to today
Nothing surprising really, you hear this now all the time, over 90% of my code is written with AI, the majority of which requires just a quick run through followed by “Merge Pull Request.” Don’t worry, the tests are included. It makes little sense to go and explain how good the models got in the coding space past the metric given above.
The superpower that came to me in the form of existing models and coding tools, paired with 22 years of experience, translated into insane productivity and growth across the products I work on. A huge portion of that went into TalentHub Partners, a platform so massive it would otherwise be impossible for our small team to deliver this much value on such short timelines.
The winning processes are:
- development in all forms, from prototyping to months of backlog features built in a week, to nearly automated bug-fixing processes
- MCPs that expose application data directly into LLMs to not only be cool and chat with data, but infinitely expand the relationship between data sets, enrichment, and 👇
- agents and automation across the majority of internal and client-specific day-to-day tasks
The use cases are so vast that, on a daily basis, there isn’t a task that doesn’t get automated. I am not talking about soccer practice calendar planning either, this is real business and operational work, the kind of bottlenecks that, once automated, let a small team tackle even bigger opportunities.
At some point, I want to write up my now daily work cycle, which has been transformed from several linearly structured tasks to a vortex of prompts that excites and overwhelms at the same time.
Biggest struggle - sleep
I was never a heavy sleeper, but nowadays my sleep resembles that of a heavy drinker, short and troubled. That said, the velocity and information flow about AI rolling in on us from every channel is, in a way, an addiction. The wide variety of tasks and speed of their execution may and does create mental exhaustion, sometimes by the time you get the first cup of coffee ☕
Good thing for me, I don’t have many readers to wake up to a ton of comments on how I should manage my time, and for anyone tempted, it’s a personal character trait, not a lack of a productivity system. The nature of heavily integrating AI into your work process today, firing asynchronous tasks of various complexities and contexts, and processing the aftermath of models’ output, isn’t best suited for the human brain, it becomes Colin Cherry’s cocktail party experiment. This does present an interesting and new way of finding priorities and aligning to them best, something that I am doing more of now since the start of this journey.
This article piece is my by-product of having time in between daily tasks and a way to take my mind off things, not away from AI, how can I, but to reflect the behavioral changes in getting work done day to day. Every thought in it came from my own “model”, the one that should get a bit more attention in the age of artificial intelligence.
P.S. Spell-checked by AI powered by Opus 4.6, because I speak 3 languages and none of them well.