The Year of Context Engineering

Summary

---

Podcast Summary:

MarTech Podcast ™ – "The Year of Context Engineering"

Host: Benjamin Shapiro
Guest: Scott Brinker (Marketing Technology thought leader, "Godfather of MarTech")
Date: January 8, 2026

---

Episode Overview

This episode examines a crucial emerging theme for marketers in 2026: "context engineering." Host Benjamin Shapiro and guest Scott Brinker discuss how marketers must move beyond prompt engineering—previously the focus of AI-powered marketing—toward harnessing and curating the right context for large language models (LLMs) and agentic AI systems. The discussion covers best practices, pitfalls, and practical advice for balancing context in AI-powered processes, drawing on industry analogies and real-world martech experience.

---

Key Discussion Points & Insights

1. Setting the Stage: Yearly Trends in AI for Marketers

• Scott Brinker reflects on the evolution of AI in marketing:
  • 2024 as "the year of prompt engineering"
  • 2025 as "the year of agentic AI"
  • 2026 positioned as "the year of context engineering"
    (01:15)

2. What is Context Engineering?

• The guest compares the transition from SEO (Search Engine Optimization) to AEO (Answer Engine Optimization) with the shift from prompt engineering to context engineering.
• Definition:
  • Context engineering is not just about crafting instructions (prompts) for AI; it also involves supplying it with the right data access and appropriate tools, especially as agentic AI (AI that can take actions) becomes mainstream.
  • This involves bundling instructions, relevant data, and tool permissions together so AI can work effectively and safely.
• Notable quote:
  • "[Context engineering] is that art, that practice of saying, okay, now when I ask the AI to do something, I want to bundle up instructions. I want to bundle up access to the right data for what it might want to need to do. I want to like point it at the tools it can use and executing that. And that's what context engineering is all about."
    (Podcast Guest / Martech Expert, 02:38)

3. The Diminishing Value of Prompts Alone

• Scott Brinker shares insights from Nicholas Holland (Head of AI at HubSpot), noting that LLMs have become good at deciphering intent even from poorly written prompts.
• Without proper context to filter and focus the AI's attention, outputs become generic or "dumbed down to the norm."
• Challenge: Determining the right scope and boundaries of context to provide.
  (03:08–04:01)

4. The Human Analogy and Real Organizational Challenges

• The guest analogizes assigning work to humans versus AI: you must provide clear objectives, relevant data, and access to necessary tools in both cases.
• Pitfall: Most organizations fail not because of AI limitations, but because they lack clarity on the "job to be done" and the process itself.
  • Many struggle to define the process or data required for a task, which translates into poor outcomes when automating with AI.
• Notable quote:
  • "...if you really want to get value out of these things, you got to step back from the AI piece of it and say, okay, what's the job to be done? What is the specific task that I want this thing to accomplish? What information, what tools is it going to need to execute..."
    (Podcast Guest / Martech Expert, 04:11)

5. The 'Goldilocks Problem': Not Too Much, Not Too Little Context

• Scott discusses challenges faced while building their internal podcast production infrastructure ("podcast OS"):

  • Too much context (e.g., huge strategy documents, full transcripts, every LinkedIn post) overwhelms the AI and leads to failures or meaningless output.
  • The task is not to simply provide more data, but to limit and curate context so that it is relevant and digestible.

• Notable quote:

  • "The problem isn't how do I give more rich information and let the machines sort through it and figure it out, it's how do I limit the information so it's only what's relevant so then the LLM can digest it. And I think that that's the problem with context engineering. It's not just give the LLM everything. You actually have to find a balance and, you know, not too little, but also not too much. We have a little bit of the Goldilocks problem."
    (Scott Brinker, 05:47)

• The guest wholeheartedly agrees and emphasizes the importance of balance:

  "Balance 100% I more eloquent than anything I could have said."
  (Podcast Guest / Martech Expert, 07:04)

---

Notable Quotes & Memorable Moments

• "There's probably like a, a parallel here of what AEO is to SEO is a little bit what context engineering is to prompt engineering."
  (Podcast Guest / Martech Expert, 01:40)
• "Without context as a filter, everything gets dumbed down to the norm."
  (Scott Brinker, 03:30)
• "If you really want to get value out of these things, you got to step back from the AI piece of it and say, okay, what's the job to be done?"
  (Podcast Guest / Martech Expert, 04:11)
• "We have a little bit of the Goldilocks problem. It's like, not too little, but also not too much."
  (Scott Brinker, 05:56)

---

Important Timestamps & Segments

• 01:15 – Framing 2026 as the "Year of Context Engineering"
• 01:40–03:08 – Defining context engineering and its connection to AI in marketing
• 03:08–04:01 – Limitations of prompt engineering and the growing need for curated, well-defined context
• 04:01–05:47 – Practical analogy: giving tasks to humans vs. AIs; why businesses struggle with AI implementation
• 05:47–07:04 – The Goldilocks challenge: achieving balance in contextual inputs for AI systems

---

Overall Takeaways

• The shift towards context engineering is critical for marketers leveraging AI in 2026.
• Success depends not on overwhelming AI with information, but on precisely curating and supplying the right data and tools for the job at hand.
• Many challenges attributed to AI are actually due to a lack of organizational clarity about processes and goals.
• The "Goldilocks problem"—finding not too little and not too much context—will be a defining martech obstacle in the year ahead.

---

Transcript

[Benjamin Shapiro]

The Martech Podcast is a proud member of the I Hear Everything Podcast Network. Looking to launch or scale your podcast, I Hear Everything delivers podcast production, growth and monetization solutions that transform your words into profit. Ready to give your brand a voice? Then visit iheareverything.com.

[Scott Brinker]

From advertising to software as a service.

[Podcast Guest / Martech Expert]

To data across all of our programs.

[Scott Brinker]

And clients clients, we've seen a 55 to 65% open rate.

[Podcast Guest / Martech Expert]

Getting brands authentically integrated into content performs better than TV advertising.

[Scott Brinker]

Typical lifespan of an article is about 24 to 36 hours. We're reaching out to the right person.

[Podcast Guest / Martech Expert]

With the right message and a clear call to action. Then it's just a matter of timing.

[Benjamin Shapiro]

Welcome to the Martech Podcast, a member of the I Hear Everything Podcast network. In this podcast you'll hear the stories of world class marketers that you technology to drive business results and achieve career success. Here's the host of the Martech Podcast, Benjamin Shapiro.

[Scott Brinker]

Let's go on to our next topic. I want to talk to you a little bit about context. A lot has been said about 2024 was the year of prompt engineering. 2025 was the year of agentic. I think personally, 2026 is the year of context engineering. How important is context and how marketers think about understanding context as we go into 2026?

[Podcast Guest / Martech Expert]

You know, there's probably like a, a parallel here of what AEO is to SEO is a little bit what context engineering is to prompt engineering. Which is to say, you know, most context engineering sort of originated with this idea of prompt engineering of like, okay, if I'm going to ask the AI to do something on my behalf, I need to feed into it as clear information instructions as possible to make sure it doesn't go off and invent its own thing. It gives me what I want. Well, it's sort of just like AEO expanded SEO Context engineering expands on prompt engineering to say like, okay, well it's not just what we want to describe as the instructions to the LLM. It's also that we want to point it at access to data that it might then be able to use in its processing. Because these AI engines are getting agentic, which means they can take actions. We might also want to provide it with access to certain tools that it can use to like, oh, okay, actually you can go out and you can query for this information or you can book this appointment. And so it's that art, that practice of saying, okay, now when I ask the AI to do something, I want to bundle up instructions. I want to bundle up access to the right data for what it might want to need to do. I want to like point it at the tools it can use and executing that. And that's what context engineering is all about.

[Scott Brinker]

I have a. One of my guests this year is a mutual friend of ours, your former coworker Nicholas Holland from Love. Nicholas, the head of AI from HubSpot. And in my conversation with Nicholas, he said that basically the LLMs are so good now, your prompt is kind of irrelevant. You can misspell words and not get everything exactly right, not in the order. And the LLMs will figure out essentially what you mean or what you're trying to accomplish. But without context as a filter, everything gets dumbed down to the norm. How do you think about creating the right context when you're building some sort of an agentic process? What's enough information? What's, what's too much? How do you figure out the right balance of, here's the boundaries of the information an LLM is looking at, and here's what I'm asking you to do.

[Podcast Guest / Martech Expert]

You know, I've so loved the way you frame this question because most of the challenges that people have with AI aren't really challenges with the AI. You know, if you took the AI out and you say, I'm going to have a human who's going to do something for me, not the LLM, we're going to have a human do this, and you're like, what instructions do I need to give them? What data do I need to give them access to? What tools are they going to need to use to execute it? It turns out for a bunch of things that we're asking AI to do, we don't have the answer to that. Well, I'm not actually quite sure what data it's need or where I would get that data, or wait, what tools might it want to use? And so if you really want to get value out of these things, you got to step back from the AI piece of it and say, okay, what's the job to be done? What is the specific task that I want this thing to accomplish? What information, what tools is it going to need to execute and pretend? I mean, you can even fake this with like, hey, Joe, you know, Mary, you know, come on over. Okay, here's what I want you to do. Here's the tools, here's the data you know, can you do? And they might ask questions of like, well, wait a second, this state isn't right. Or, you know, what about, you know, this other data that you didn't give me to. That is all where all the hard work is. You know, we hear the same thing with some of the AI automation things and people like, like, hey, I can't get AI automation to work for my process. Okay, well, step one, can you describe your process? Oh, no, not really, man. They do it different ways all the time. And you know, you're like, okay, until you really know what your process is, it's kind of hard to get AI to go unautomated for you. I don't mean to sound flippant about it, but these, like, these are really the hard challenges that we have got to overcome to tap the capabilities that AI is in a position to deliver.

[Scott Brinker]

Yeah, one of my big struggles as we've started to build out podcast os, the infrastructure we use to produce all of our podcasts is actually not overwhelming the AI with context. Right? Like I want to go and give it a 20 page master strategy document and then an hour long transcript and then the, the guest's entire LinkedIn page bio, including all of their posts for the last six months. It's too much information. It actually, practically speaking, some of the operations we wanted to execute can't happen because it's too much information and we just get everything to time out. And the thing that's actually been surprising to me, the problem isn't how do I give more rich information and let the machines sort through it and figure it out, it's how do I limit the information so it's only what's relevant so then the LLM can digest it. And I think that that's the problem with context engineering. It's not just give the LLM everything. You actually have to find a balance and, you know, not too little, but also not too much. We have a little bit of the Goldilocks problem. It's like Goldilocks, which So well said. Three wolves. Yeah, three bears. Three wolves. Anyway, we've got a problem of finding.

[Podcast Guest / Martech Expert]

Balance 100% I more eloquent than anything I could have said.

[Scott Brinker]

And that wraps up this episode of the Martech podcast. Thanks again to Scott Brinker, the godfather of the Martech industry, for joining us. If you'd like to contact Scott, you could find a link to his LinkedIn profile in our show notes or on martechpod.com you can always visit his website, which is chief martech.com it's Chief M A R T E C. If you haven't subscribed yet and you want a regular stream of marketing and technology knowledge in your podcast feed, hit the subscribe button in your podcast app or on YouTube, and we'll be back in your feed every other week. All right, that's it for today, but until next time, my advice is to just focus on keeping your customers happy.

[Podcast Guest / Martech Expert]

Foreign.

[Benjamin Shapiro]

Thanks for listening to the Martech podcast and I hear everything. Production Looking to launch or scale a podcast like this one for your brand? Then visit iheareverything.com.