Hi, it's Carl. Before I get into it, there's a free gift at the end of this section. Stay with me.
Here's the question I get from clients or recruiters: "Is prompt engineering still worth learning when models keep getting smarter?"
Short answer: yes, but not as a standalone job.
In 2024, "prompt engineer" was a real role. People made $1,800–$2,000/month writing prompts that got GPT-4 to produce clean, structured answers. That era is over. Claude, ChatGPT, Gemini, Grok, DeepSeek — they're all smart enough now to handle lazy prompts and still return something decent.
But "decent" and "production-ready" are different planets.
Try this right now. Open two tabs of your favorite LLM and paste these prompts separately:
Prompt A (the lazy one):
Build me a Make.com automation for a marketing agency. We take data from Google Sheets, push it to an LLM to create a cold pitch, then push to Gmail.
Prompt B (the engineered one):
ROLE: You're an AI Engineer specializing in Make.com automations.
TASK: Build an automated lead enrichment workflow.
TRIGGER: New form submission (Google Forms or webhook).
PROCESSING: Extract Name, Email, Company. Send to OpenAI API to (1) classify industry, (2) generate a short company description, (3) assign a lead score 1–10. STEPS: Use HTTP module to call the API → parse JSON → format enriched data. STORAGE: Save to Google Sheets or HubSpot.
NOTIFICATION: Slack message if lead score > 8.
TONE: Senior AI Engineer explaining to a mixed technical/non-technical audience.
Both prompts describe "an automation." The output is not even close.
Prompt A gives you a vague tutorial. Prompt B gives you modules, data schemas, branching logic, and tone-matched documentation. Same model. Same task. Different world.
Here's the real shift: models got smart, but they still don't know your context, your constraints, or your user. That gap the one between what the model can do and what your product needs is where prompt engineering lives now.
It's no longer a solo skill. It's a pillar of AI engineering, alongside evals, retrieval, tool use, and orchestration. The prompt engineers who survived 2024 didn't quit. They became AI engineers.
If you're hiring: stop asking for "prompt engineers." Start asking for people who can design prompts inside a system with fallbacks, evals, and structured outputs.
If you're building: stop typing one-liners and hoping. Write prompts like you write code.
The takeaway: prompt engineering isn't dying because models got smarter. It's evolving because the surface area got bigger.
Free Resource of the Week
The Universal Prompt Template (with documentation)
This is the exact template structure I use across for automations, code review, lead scoring, you name it. It covers role, context, task, constraints, output format, and tone, with fill-in sections and examples.
Grab it here: The Universal Prompt Template →
Duplicate the Notion page, swap in your variables, and adapt the sections to your use case. Works with Claude, GPT, Gemini, or any frontier model.
Sign-off
That's it for this week. If prompt engineering is something you want me to go deeper on specifically how to embed prompts inside software and fine-tuning pipelines reply and tell me.
See you next Thursday nerds.
— Carl
P.S. If you know a CTO or founder who's still confused about where prompt engineering fits in 2026, forward this to them. That's how this newsletter grows.