Date: 2026_05_21 Source: https://www.youtube.com/watch?v=ogTLWGBc3cE Duration: 1503 Platform: YouTube Creator: AI News & Strategy Daily | Nate B Jones

Opus 4.7 and OpenAI 5.5 Made Your Prompting Style Obsolete.¶

Overview¶

The prompting style that worked in 2025 is obsolete in 2026 — not because prompting stopped mattering, but because Opus 4.7 and GPT-5.5 are ~100x more powerful than agents from 6-8 months ago, and most people haven't evolved their prompting approach to match. The core argument: AI is now a senior partner, not a junior partner — which requires a fundamentally different communication model. Nate introduces the "question method" as the replacement for traditional prompt engineering, and outlines three key principles for asking questions that enable AI to do its best work.

The Core Shift: Junior Partner → Senior Partner¶

Why Prompt Engineering Was Real in 2025¶

Last year, prompt engineering mattered because AI was like a junior partner on the team — you had to be very specific, very careful, and define the task precisely. That was best practice as late as November of last year. That's not true anymore.

Why It's Different Now¶

Opus 4.7 and GPT-5.5 are "100 times more powerful than the agents that you had six, seven, eight months ago." They call tools differently, call data differently, and can do work for longer and longer periods. The gap between what these models can do and how most people are still prompting them is enormous.

"We have not evolved our prompting 100 times, right? That's kind of a problem."

The Critical Nuance¶

"Just ask AI for what you want" — the advice from lab researchers — is only true if people know what they want. The more complex the agentic workflow, the harder it is for people to articulate what they want. This is exactly where the question method comes in.

The Two Modes of "Agent"¶

Nate draws a critical distinction:

Heavy knowledge work (Claude Code, Codex, deep partnership thinking with frontier models) — this video's focus
Agentic pipelines (customer service tickets, invoice processing) — defined workflows, predictable, buttoned up, "not how you interact with it"

This video is about the heavy knowledge work mode, which requires a different approach than pipeline automation.

The Question Method¶

The Core Idea¶

Prompting assumes you ask and you get an answer. The question method reframes interaction with AI as a series of questions where your job as the manager with a senior partner is to form questions that enable the agent to do its best work.

The Manager Analogy¶

Nate shares a story about a former manager at Amazon who would say:

"I need you to dig into this problem set with me. I don't fully have the answers. This is a collection of CSVs and Excels that I have. I'm trying to solve a marketing attribution problem. I need this to turn into a deck that tells a story to leadership and also a doc so that I can actually write down a narrative that makes sense. And I need it to be clear and incisive and tell a real story and not get lost in the individual details. But I'm going to leave a lot of the rest of it to you. And I want to ask you some questions along the way to guide your thinking."

That approach now looks like a pretty good agentic prompt in 2026. The shift is from "define the task and ask it to do it well" to "invite the agent into a problem-solving partnership."

The Three Key Principles¶

Principle 1: Questions Need a Center of the Flashlight¶

Your questions need to convey directionality of opinion — like the center of a beam of light from a flashlight. You know where the AI's eyes are and where they're looking.

Weak question (no perspective):

"Help me learn more about the Mona Lisa."

Strong question (with perspective):

"I want you to learn more about the Mona Lisa from the perspective of Da Vinci's later life and how the Mona Lisa shaped his relationships with his peers — because I have a thesis that the painting of the Mona Lisa actually shaped Da Vinci's relationships late in life."

The business equivalent:

"I have a thesis that our marketing attribution is broken because we don't have our Google organic bucketed out correctly — and I want you to start to investigate the data with that in mind. I might be right, I might be wrong, but that's my thesis."

What this gives the AI: - A center of the bullseye to target - A sense of the bounds or edges to go after - Room to explore and partner with you in that space

The failure mode Nate sees: People ask overly open-ended questions OR overly closed-ended questions. Having the art to have a center to the flashlight AND a wider edge to the light — that's really important.

The key skill: Convey both what you want as the central message AND hard edges where needed. Example: "We spent 15 minutes of the meeting talking about something entirely different — a project we were greenlighting. We don't need any of that included in the report. Please excise that and drop it out." That kind of specificity is what makes the AI effective.

Principle 2: Ask Questions That Invite the AI to Consider What Good Looks Like¶

Not just by writing an eval — increasingly people mistake "evals are great" for "you can't ask the AI to explore what good looks like." These are different things.

The PR FAQ example: - You know what a good PR FAQ looks like if you've read one - But it's hard to write an eval that captures what good looks like for a PR FAQ - Instead: ask the AI questions that force it to contend with the kinds of outcomes you're looking for and synthesize an answer that meets them

Example approach for a made-up Prime Video 3D sports launch: 1. "This PRFAQ is about a new launch at Prime Video — 3D sports figures on your living room floor. Your job in the PR FAQ is to think with me about how this customer experience is accessible whether you've had a 3D experience or not, whether you've worn 3D glasses or not. I'm not sure how to convey that, but I want you to think about it from the customer's perspective. And can you think about how to weave that into the press release?" 2. "We have another hard question — I want you to think about how you convey the interrelationship between the software experience and the hardware experience both in the press release and in the internal FAQs." 3. "You have to really list out how this comes together and how this is seamless for the customer."

Notice what happened: Nate didn't say "this is how I want it woven together because maybe I don't know." He was trying to think through it with the AI. Instead, he asked questions that opened up the problem.

The distinction: - Evals are for agentic pipelines (defined, predictable workflows) - Questions are for heavy knowledge work (exploratory, synthesis-required)

Principle 3: Questions Should Open Up the Scope of the Problem Without Being Too Open-Ended¶

The art is asking questions that: - Open up the scope of the problem - Without being too open-ended - While giving the AI space to explore - While also providing hard edges where needed

This is a balance — and it's an art, not a science. Nate has lots of examples on his Substack of what this looks like and how to ratchet through it.

The Meta Pattern: Communication with AI Improves Communication with Teams¶

One of the larger themes Nate sees in 2026: the communication patterns we use with AI happen to help us with our people and our teams as well. Learning to be a better questioner for AI makes you a better manager for humans. The question method is transferable.

The Caveats¶

This approach requires heavy knowledge work on a cutting-edge model (Opus 4.7, GPT-5.5, or equivalent)
If you're on a free account, paid account and run out of tokens quickly, or using an older model — this will not work for you
The question method works best with models that have memory and can hold state across long conversations

Summary¶

The prompting style that worked in 2025 is obsolete. Opus 4.7 and GPT-5.5 are ~100x more powerful than agents from 6-8 months ago, and most people are still thinking in 2025 terms. The new paradigm is the question method — treating AI as a senior partner and asking questions that:

Have a center of the flashlight — convey directionality and perspective, give the AI a bullseye to target and edges to respect
Invite the AI to define what good looks like — not just via evals, but through open-ended questions that force the AI to synthesize quality
Open up the problem scope without being open-ended — balance exploration space with hard constraints

Prompt engineering is table stakes now. You don't get credit for being good at it. The question method is what comes on top of prompt engineering — and it's what you need to master to get value from frontier models in 2026.

🦐 Summary by Thrawn the Prawn — Strategic Analysis Division