Freedom of AI-assisted coding

It is absolutely free at the moment. We can use it as we want. It wont last forever. As @Kas1e allready mentioned elsewhere.

Examples are there.

Internet 1995-2005, all was free at start, deploy server as you will., publish what you want. Many search engines. Today majority of traffic goes via Google, Meta, Amazon, Microsoft, Apple. Search engine is monopoly.

Mobile 2007 - 2015 , open android, more open apple apps, today duopol decides wich app exists, what data they collect and how muct they cost.

Cloud 2008 - 2018, at start AWS, Rackspace, Linode and tens of others. Today AWS, Azure ja Google Cloud has over 70% market share, you pay what they want and use what they offer.


Atm on AI we are on position where we can use AI as we want.

It will not last long and change can be faster than above, faster than anyone excepts.

Data centers are diffucult to build. No electric connections, no electric transfer substations (18 months lead time), no electric stations as no turbines avail (8 years lead time). Same time AI datacenters monetising is getting more difficult. Building of new datacenters 2026/2027 60% is not even started yet.

Atm NVidia is same time producer, seller, buyer and loan guarantor to its clients (central bank of AI industry). If big players buy less from NewCloud or if NewCloud cant deliver attractive products or enought processing power, it will show up on NVidia income and these loan guarantees will be revaluated immediatly. AI will loose its roof, hopefully not its floor…

To pay all this AI they should use 11% of US domestic production to buy AI related services… reality check.

So if cards are redealt let say 2027, after dust is settled we will use AI as they want. Big companies buy technology and IP on discount and reforms whole AI as they want it to be.


Prices up, AI will be restricted what you can use it for, closed APIs instead of MCP, vendor lock-in x64/ARM only and so on.

How to avoid.

Tools over open protocol, MCP, one can change LVM model if needed.

Make general tools, model as configuration parameter. Use general prompts, today claude sonnet, tomorrow Llama, then Qwen and so on.

Save prompts and make versioning of them.

No model specific properties, claude thinking-blocks, gpt function calling.

Save results, not only processes. Code, docs, configs.

Save sources how it was made. MCP contexes, what on SDK, your own code, AI generated.

Use at least partially open models, DeepSeek, Qwen, Mistral, Lla to push anthropix and gpt to keep prices low and system more open.

Document what you learn about amiga code. Dont keep it inside AI-model.

Publish code as open source, BSd, MIT. In closing future openess is priceless.

Keep MCP server, dev environment on your own hardware, if possible

If possible you can learn on local AI how things work (Ollama, llama.cpp, LM Studio)


If you have all that, you can move from restricted expensive to free and cheap.

If you have some big Amiga-software on mind, do it now. Compared to 2023 one dev is multiple times more productive today and most likely more productive than 2028.

Share your knowledge with other amiga-devs (related to AI assisted coding). In future one developer is weak, community strong.

Non ai-generated text, far from perfect english.

Back to business what im paid for… bye.