Get Your Engineering Out of the Middle Ages: A Practical Guide to AI-Powered Programming

Programming is entering a renaissance

My favorite analogy? We’re moving from Middle Ages-style feudal agriculture

Reference:: From Manors to Markets: The Economic Role of Agriculture in Medieval Europe

to a more modern, industrial way of doing things!

Reference: Agriculture

We are not there yet, but it’s so much nicer, so much fun and so enjoyable to program with AI! It’s still programming but with better abstractions! From machine code → assembly → C → high-level languages like Python & JS → AI-assisted programming. At every step, abstractions improved. Now we write with AI. It’s still programming, but faster, better and much more efficient.

Show me how you do it!

You can chat and theorize about this all the time, but first-hand experience is irreplaceable to make a correct judgment!

Let me show you how I work day-to-day with AI. My stack is minimal: Anthropic Pro, ChatGPT Pro and Cursor Pro – total ~$240 + API usage.

Cursor Pro – just use it! If you have any doubt -> don’t! It has strong competitors – so if another tool works better, great. The point is the workflow:

• Ask – My most-used feature. It’s a persistent chat that can edit multiple files with line-level precision. Some of my sessions run for weeks. I debate code, tweak logic, request alternatives, apply best practices – all like I’m pairing with a sharp teammate. Can’t imagine coding without it anymore.
• Autocomplete/Tab – Nice for simple tasks. But honestly? Just table stakes these days.
• Agent – I rarely use it. It creates extra files, derails easily. Reminds me of “Thoughts On A Month With Devin” blog. Still, tools like Claude Code, OpenAI Codex, and Devin are evolving fast and becoming much more accessible.
• Cmd+K – Perfect for small, focused edits. I mostly use it for quick grammar fixes.

How I use Cursor with LLMs:

• For small codebases / early projects: throw everything into context – it just works.

• Reuse prompts? Move them to Rules

• Repeatable workflows? Build MCP servers

• New APIs/tools? Document context in llms.txt

When it gets messy – treat it like mentoring a junior dev:
→ Reduce scope.
→ Isolate problems.
→ Add more context.

There is a lot of nice feedback about other editors: windsurf, zed, and it looks like VS Code with Copilot is trying to catch up, but I’m all-in on Cursor.

UI-based tools like Bolt and Lovable are great for low-code workflows or non-engineers. I tried Bolt Pro, but ended up back in Cursor for serious code work.

Those tools are great, but probably for non-tech folks mostly and eventually you end up in code editor anyway. Some other tools I use are: warp as terminal and graphite for PR review.

Recommendations for companies

Here I want to give actionable recommendations on how to take advantage of LLM-based coding tools if you are not doing this already.

🏗 Level 1 – Basics

Start simple.
Give every engineer:

• Cursor Pro
• API keys for OpenAI / Anthropic

Run an onboarding session or a few pair-programming rounds if needed.
This gets you 80% of the value for 20% of the effort.
Anything else is harder.

🔧 Level 2 – Beyond Basics

Start building:

• A library of your team’s best prompts and Rules
• MCP servers for your most-used workflows (DBs, observability, etc.)
• Add 2 more tools: AI code review platform and an AI-enhanced terminal

🚀 Level 3 – Advanced

Go all in on building your own setup.
Custom models and custom infra:

• Use platforms like Coder for custom infrastructure
• Deploy custom or self-hosted LLMs for large codebases (DIY Personal Copilot or managed solutions like Poolside)

🌀 Bonus Meta-Level

This one’s just for fun (kind of):
Start publishing defeatist blog posts like:

“Engineers are doomed.”
“Coding is solved.”
“It’s over.”

Then promote them like hell – so your competitors stay confused while you scale!