Phase 1: Veteran Coder
Early teens: Geekin on 8-bit computers; Vic 20 and C64, making games.
1995-2017: Full-stack freelancing: Mostly web/database SAAS.
Phase 2: Machine Learning Pioneer
2018-2021: Codification of cognitive empathy combined with SQL and huge amounts of automated web scraping, plus manual input of data yielded a mimicry of clinical cognitive empathy in the form of a chat bot.
Phase 3: Vibe coder-ish
[dark ages of LLMs]
2022-2023: Cautious fascination, limited trust, limited use.
2024-early-2025: Dove in. Scoffed arrogantly at the label “vibe coding”. Only in hindsight will I admit I was sorta doing it. I say “sorta” because to me “vibe coding” was reserved for those who had not been software engineers for 30 years.
“Hey AI, build this big thing…”
“Hey AI, do these 5 things…”
“Hey AI, fix this bug…”
Gee, it seemed to work after a bit of tweaking.
Then I looked at the code it wrote. The naming conventions it chose. The file sizes! The repetition and redundancy! The rare generalization; it built a specific function for every single feature! Yeah, for someone who prided himself on writing applications that could be handed off to anyone for easy groking, expanding and maintaining, this was a nightmare! Throw in headaches from security vulnerabilities, databases with hundreds of unnecessary tables and columns, weird or non-existent normalization, and, yeah… a lot of reverts and even just starting over clean.
Fast forward to now…
Phase 4: AI Team Lead
Early-mid-2025: I discovered an extension for VS Code that was head and shoulders above Github Cockpilot. I swear this is not a commercial for Roo Code. Anyway, it was free and you plugged in as many API keys as you wanted so you could use the latest LLMs for the “modes” (really, agents) included, so the only limits you had were whatever LLM provider you chose. Want no limits? Use OpenRouter and fill your credit wallet there with money!
*Then* I discovered how to write custom instructions for Roo Code’s modes/agents and how to make my own! I quickly built my own little dev team! I expanded “Architect” into a 3-planner team that brainstormed with user, logged its progress, and so much more! For the beautiful gory details, check out the repo: https://github.com/ScotterMonk/AgentAutoFlow
*Then* the Roo devs added **Skills** (yes the replacement for shitty MCPs), wonderfully lightweight and efficient! With that addition (a couple weeks ago?), I modified my team so that each mode had clearly defined low-context-window-using tools!
So here I am, **loving** using LLMs to help me rise up to a more architectural level, every day finding ways to improve my team’s instruction sets.
*Now, I rarely write any code and that’s fine with me*
Much more fun and productive to keep adapting my team to the quickly changing LLMs available and their capabilities. Am I more productive now? Including the time I spend improving the team while they build stuff for me, as well as constantly asking for refactors, constant rebuilds of each project’s knowledge base, and methodical studying of the thought processes revealed by the LLMs as they plan and build for me, I’d guess I’m doing about x5 net efficiency compared to what I could do 4 years ago. And an important part for me: **IT’S FUN**!
I LOVE being able to swim in this “mid/high level” (creating / planning / guiding / organization)! OK now I want to write an article about this, heh. Thanks for the inspiration!
Watch the animated musical story / prediction of where things could be going and how we get there. Trigger warning: It starts out dark.
Recent Comments