What 6 Months of Posting About AI Taught Me (the boring, useful parts)

I'm a year in writing about AI tools on DEV, and I keep getting DMs asking the same questions:
"How do you pick tools?", "What stays useful after the hype?", "Is prompt engineering still a thing?"
Instead of answering them one by one, here's a brain dump.

1. Most "AI tools" are CRUD wrappers

The "AI X" product where you upload a CSV and get a chart?
90% of them are 200 lines of code + an LLM call + a Stripe webhook.

If you can write Python, you can rebuild most of them in a weekend. So either
build it (fun) or pay for the UX (also fine) — but don't think you're locked in.

2. Prompt engineering is now 10% prompt, 90% context

• What changed: stuffing the prompt is dead. Stuffing the context window with curated snippets is what works.
• What didn't: writing clear specs is still the highest-leverage skill. If you can write a 10-line brief an LLM can execute, you're 10× the dev who can't.
• What I use daily: a PROJECT.md at repo root with goals / non-goals / glossary. Cheap, high ROI.

3. Local models are good enough now

• Llama 3.2 3B on a 2015 laptop does summarization fine.
• Qwen 2.5 7B on an M2 Mac answers code questions well.
• Use them for: redaction, classification, low-stakes chat. Don't use them for: anything where a wrong answer costs you money.

4. The thing nobody wants to say out loud

Half the value of "using AI" is having an opinionated peer in the room.
The other half is not trusting it when it's confidently wrong.

Tools come and go. The habit of "ask, verify, ask again" is the actual skill.

5. My actual stack (June 2026)

• Daily driver: Claude / ChatGPT for reasoning, Gemini Flash for cheap classification.
• Editor: Continue.dev in VS Code, Codeium for autocomplete.
• Local: Ollama + Qwen 2.5 7B.
• Tracking: I just keep a markdown file. I tried Notion, Airtable, Obsidian — markdown won.

---

If you're starting out: pick one tool, use it for 30 days, then write what you learned.
That's the post I'd want to read.

What stack are you on? Curious what others are finding sticky — drop a comment.