🎢 “We’re Entering an era where intelligence will no longer be a limited resource." - Reid Hoffman

Hello Reader,

Last week at AgentCamp in Christchurch, hosted by Steve Knutson at Stratos Technology Partners, I spoke about a shift that is quietly transforming how people work with AI.

Most people still think the secret to good AI output is writing a clever prompt, but the deeper truth is that prompting is only a small part of the equation.

The real leverage comes from context engineering—everything you frame up for a LLM, prior to the prompt.

Antonio Gullí described context engineering as "the discipline of designing, constructing, and delivering a complete informational environment to an AI Model prior to generation."

This includes the prompt, of course, but also conversation history, instructions, goals, knowledge sources, structured outputs, and sometimes persistent memory.

When those elements are arranged deliberately, the model performs far better because it understands the situation it is operating inside. Instead of guessing what you want, it can reason from a well-prepared landscape.

The easiest way to see this difference in action is through progressive prompting. Rather than asking for a final result all at once, you build the work in stages.

Compare these two types of prompts, without and with context, in 2 different browser tabs, with the same LLM open in each:

In Tab 1:

In Tab 2, enter these one at a time, sequentially:

You may notice that progressive prompting creates output that is far more coherent and thoughtful than single-shot prompting. This is because the first prompt has to dilute its token budget over choosing the topic, organising it, and composing the speech. The second set of prompts uses context engineering to triple the token budget, and focus on each aspect of the process individually.

This approach also reveals a subtle constraint that many people overlook: context quality matters more than context size.

Large context windows are impressive, but they do not automatically produce better results. When too much irrelevant or poorly structured information accumulates in the conversation, the model begins to lose clarity.

Some practitioners call this “context rot.” The AI becomes distracted by outdated instructions, conflicting details, or irrelevant data. The solution is not simply adding more information; it is curating the right information and pruning what no longer serves the task.

How can you get a handle on context engineering? Come to my next webinar.

On March 20, I will be running a live webinar inside the AI Coaching Academy called Context Engineering: The Key to Agentic Design.

In this session, we will explore how to design better AI workflows, how progressive prompting improves results, how agentic memory systems work, and what practical guardrails help keep AI systems reliable and secure.

My goal is not just to explain these ideas, but to demonstrate them, so you can see how these techniques translate directly into everyday work.

If you are curious about the direction AI is heading and want to develop the skills that will matter as these systems become more capable, I would love to have you join us.

The future of AI will not belong to the people who simply type the best prompts. It will belong to the people who understand how to design the context, structure the workflow, and guide intelligent systems toward meaningful goals. And like any skill worth having, that capability grows through deliberate practice.

​Register here.​

📰 New AI News This Week

• OpenAI released GPT-5.4, the first general-purpose model with computer-use capabilities
• Eubiota.ai uses agentic AI to discover microbiome therapeutics in the human gut
• Dyna.ai raises millions to bring agentic AI into financial services

Are you ready to grow your Ai capabilities?

Join me in the Ai Coaching Academy:

Friday, March 20 at 2:00 PM GMT+13 Context Engineering: The Key to Agentic Design Context engineering prepares everything prior to the prompt. Even the best prompts will fail without the right context. Giving AI backstory and reference material will create a landscape that is suitable to your scenario. But you can't just give an AI agent infinite data; you have to understand memory as a stateful context, so you don't overwhel... RSVP

Keep on growing!

- Caelan Huntress ​[email protected]​

​​​​​