AI is compressing the cost of building and maintaining software. That threatensens some “coding machine” jobs — and it also reopens niche custom apps that budget once killed. A practitioner’s note for freelancers, small agencies, NGOs and organisations, with Umbraculum (docs) only as an illustrative stack example (not an endorsement).
OpenPLC is open source. That is not a slogan for this brewery project — it is how we reverse-engineered the Editor project layout, the Runtime upload path, and the generated ST so AI-assisted work could stay grounded in real code. Here is what that looked like in practice.
AI in the IDE raises GDPR-shaped questions: what leaves the sandbox, who processes prompts, and when on-prem DeepSeek on EU GPU (e.g. Contabo) is the honest answer for agencies and privacy-focused merchants.
Garrison Lovely’s Obsolete (OR Books) frames AI as a capital project to replace human labour. Here is a practitioner’s take on the ethical stakes and the technical habits that keep AI useful without surrendering judgment.
AI authoring is useful when it sits inside a discipline: explicit rules, runnable procedures, specialist agents, and gates that refuse bad work. Here is how that looks across complex development work (Magento, PLC, ERPs, native apps) — with orchestration, analysis, and tests ahead of vibes.
For the first time in open source, I feel something chilling — that we could have surrendered our intellectual strength to the big tech giants. I could be wrong. I do not like this feeling. PhotoCraft’s “you already know” is where that doubt starts.
How much can you trust an AI IDE on well-known logic and technical detail when the chat context periodically resets? I answered that by forcing a pamphlet on electromagnetic fields — a domain I am not expert in — inside an unrelated PLC project workspace. The surprising part was not the fluency: it was how little context was lost when the job stayed gated and organised.
A technical companion to how we run Cursor 3 — plugins, AGENTS.md, Ansible, flows, and criteria-based testing — opened with labour-market caution (Acemoglu/Ricardo, not a sales pitch). Adaptation is survival pressure, not proof conditions improved; 5× delivery can worsen the market we already see freezing.
Some people address the assistant with words the like of “schiavetto” (Italian — roughly “little slave” / disposable help). Others overlay a digital whip when the answer is wrong. That is not a joke about “posture” — it is abusive language and gesture. Words are important. If someone rehearses mastery over a servant in the chat, ask how they would behave toward others when nothing stops them.
With the rise of AI I wonder if the internet’s dream of a better world is sunsetting — AI x-risk talk paired with AI as a deterrence weapon. Marat’s meanest individual, Skokie, then Germany, Scotland, Assange. Soft doubts, hard sources.
