Technical fluency used to be a specialist concern. AI is making it everyone’s problem, and the marketers who develop it alongside taste and context will own the next decade.

AI is producing more and more of the work — drafts, audits, analyses, creative variations, research summaries — and across organizations, that work is being directed and approved by people who can’t fully evaluate what they’re getting back. This isn’t to say they aren’t highly skilled marketers with impeccable taste. In fact, the output usually looks impressive enough that many, even the most skilled, don’t push back in the moment. The question of whether it was actually any good gets answered three months later in pipeline data and performance reports, by which point the team has moved on to the next deck.

This isn’t a generational problem, and it isn’t unique to any particular kind of marketing organization. The same dynamic shows up in five-person PR shops and 500-person in-house teams. It shows up at B2B SaaS companies that are supposedly tech-forward and at consumer brands you’d assume have it figured out. What we’re looking at is a literacy problem, and it’s about to start defining the next decade of marketing careers because the gap between marketers who understand what AI is actually doing and marketers who delegate to a chat window is widening fast.

Marketing has always evolved with technology. AI changes who has to evolve.

Communications has reinvented its core toolkit several times over the last few decades, and PR is a useful discipline to view through the historical lens because the changes are so vivid.

Media outreach used to mean phone trees, faxed pitch sheets and Rolodexes of beat reporters managed on index cards. That gave way to email blasts and the early years of mass media databases, which gave way in turn to the first digitally-driven at scale outreach through platforms like Muck Rack and Cision, and now lives increasingly inside social DMs and direct relationships built and maintained on LinkedIn, X and Substack. Media monitoring followed a similar arc. Mailroom clipping services used to physically deliver newspaper cuttings to clients in interoffice envelopes. Wire-service alerts replaced the envelopes. Digital monitoring platforms replaced the wire alerts. Today, AI-summarized media intelligence with real-time anomaly detection is starting to replace the dashboards we just finished training the team on. Even the press kit itself has been rebuilt three times over — from paper folders sitting on a receptionist’s desk to static online press pages to fully integrated digital newsrooms with multimedia, executive bios, real-time stock data and embedded analytics.

Paid media has been on its own version of that journey. Manual insertion orders and faxed media plans gave way to ad networks. Ad networks gave way to demand-side platforms with real-time bidding and audience targeting. RTB gave way to programmatic with algorithmic bid management. And now Performance Max and similar AI-driven systems are doing creative optimization on the fly — generating, testing and reallocating spend across variants without a human in the loop for most of the day.

The analytics side has moved just as fast. Quarterly reports built in Excel pivots gave way to BI dashboards in Tableau and Looker. Those dashboards got augmented by product-analytics platforms with cohort analysis and behavioral segmentation. And the current frontier is predictive and prescriptive layers stacked on top of all of it, with natural-language querying that lets a non-technical user ask questions a SQL analyst would have spent a day building two years ago.

Each of those shifts was largely contained, though, and that’s the relevant difference. The PR person learned the new tools. The ad ops person learned the new platforms. The data analyst learned the new dashboards. The rest of the team didn’t have to develop the same fluency because the technology lived in their colleagues’ workflows, not their own.

AI doesn’t work that way. It touches content, paid media, creative, analytics, PR, internal ops, customer research and probably three other new functions by the time you finish reading this paragraph. That’s the part most marketing teams haven’t fully reckoned with yet — the literacy gap isn’t a specialist’s problem anymore. It belongs to the whole team, and pretending otherwise is how organizations get out-executed by competitors that are six months ahead on the same starting line.

The average marketer is using AI through a chat window.

For a lot of marketing teams, the daily AI workflow lives entirely inside a chat interface. ChatGPT in a browser tab, Claude in a web app, Gemini sidebar in the inbox. For casual use, that’s perfectly reasonable. For the kind of work that’s supposed to drive measurable business outcomes, it’s leaving most of the capability on the table.

There’s a meaningful gap between what a marketer can do with structured AI workflows and what they can do typing one message at a time into a chat box, and the gap is widening rather than narrowing. If you’ve only worked with AI through a chat interface, the gap can be hard to see — there’s nothing to compare it to.

A couple of examples make the gap concrete.

The Competitor Audit

Take a competitive intelligence project — say, an audit of how five competitors are positioning around a particular product category across their websites, ad creative, press coverage and earnings call transcripts. In a chat session, you upload materials in batches, ask for analysis, watch your context window fill up, start a fresh chat to handle the next competitor and try to stitch everything together at the end. The model gets foggier as the conversation grows. By the time you’re synthesizing across all five competitors, you’re operating on partial memory and copy-pasting between threads. The work gets done, but the quality drifts and you spend hours doing manual reconciliation that the system itself should be doing for you.

Now imagine the same project structured differently. You build a reusable workflow — it could be a Claude Skill, a custom project, a script or some combination of those — that ingests each competitor’s materials independently, extracts the same pattern set across all of them, normalizes the output and produces a consolidated brief at the end. Once the structure exists, you can run it across the next five competitors in an hour, and the next quarter, you can run it again. The intelligence is the same, but the operational execution is fundamentally different. And the structure compounds, because every time you use it you find a refinement, and every refinement benefits the whole team.

The Ad Campaign Launch

A more everyday example — a campaign launches and you need to generate 150 ad variations across three audiences, four messaging angles and several formats. In a chat session, you start strong with copy that you quickly approve as hitting the mark. By the seventh batch, the model is repeating itself, mixing up the audiences and producing copy that sounds vaguely like a variation from earlier in the conversation. By batch ten, you’re spending more time correcting model drift than approving copy. With a structured workflow, every variation gets generated against the same brief in parallel, with quality checks built into the process, and the output lands in a format your team can actually review without unwinding the model’s confusion first.

The thing worth emphasizing here is that none of this requires becoming an engineer. In fact, most of it can happen entirely in-app with the right structures in place. What it requires is recognizing that the chat window is one tool inside a larger system rather than the whole system itself.

The future marketer thinks like a product manager.

Picture the marketer who isn’t operating strictly in-chat.

She doesn’t open a chat box every time she has a task. She thinks about whether what she’s doing is a one-off question or part of a recurring workflow, and if it’s recurring, she invests fifteen minutes in setting up a structure that the whole team can use. Over time, she’s built a working library of project configurations, custom skills and reusable prompt templates that get refined as the team learns what works. When a new use case shows up, she can usually adapt something existing rather than starting from scratch.

When a colleague brings her a workflow that’s been giving them trouble, she can usually diagnose the issue in a few minutes — not because she’s a coder but because she understands how the underlying systems behave. She knows the difference between “the model isn’t smart enough” and “the prompt isn’t clear enough” and “the context window is full,” and she knows which problems need a better instruction versus which need a different architecture entirely.

The way she talks about her work has shifted, too. She doesn’t ask, “can AI do this for me?” She asks, “what’s the right system to do this at scale, and where does my judgment add the most value?” That’s a product manager’s question rather than a copywriter’s, and it assumes she’s the one shaping the workflow rather than feeding inputs into someone else’s.

This isn’t hypothetical. Marketers like this already exist on plenty of teams, and the gap between them and a chat-ceiling operator shows up across the work. They produce more, faster, with less rework. They catch errors that a chat-only marketer wouldn’t notice because they actually understand what the AI is doing under the hood. And their team’s expertise gets captured in shared structures rather than locked in individual chat histories that disappear when someone changes jobs.

The gap is real, it’s measurable and it’s compounding. Most organizations just don’t know how to measure it yet, which means the marketers building this leverage are operating mostly under the radar. That won’t last.

The mindset shift from operator to system orchestrator

How does someone get from the chat ceiling to the system orchestrator mindset?

The shift is more about how you frame your work than about what tools you have access to. Marketing teams need to start operating the way product teams do — thinking about systems, integrations, repeatable workflows and the upstream and downstream effects of decisions. That’s true at a 35-person integrated comms agency, at a five-person PR shop and on an in-house team of two.

This is also a different conversation from the one we were having in 2023, when the X (the platform, not the variable) topic of the day was how every company will be hiring a “prompt engineer.” That was always the wrong framing of what is actually a baseline business literacy. The way baseline business literacies actually develop is gradual and pretty consistent — typing was specialized in the 1960s, email was specialized in the 1990s, and spreadsheet basics, web search and video calls each started in the hands of specialists and became expected of every knowledge worker as the underlying tools spread through the workplace. There are still typists, email marketers and financial analysts who specialize in advanced versions of these skills, but the baseline literacy — enough to handle your own day-to-day work — became part of the job itself. Prompt engineering is now (and should rightly) settling into that same trajectory, becoming a baseline literacy that everyone develops because the work demands it.

Where this work actually compounds is in reusable, structured AI workflows. Some of those workflows are built in code, but plenty of them aren’t. They might be Claude Skills, custom projects with carefully tuned instructions, structured prompt libraries or integrations between AI tools and the systems your team already uses. The medium matters less than the mindset. If the work you’re doing today can’t be repeated by someone else on your team without you walking them through it personally, you don’t have a workflow yet — you have a personal habit dressed up as one.

Translating that into specific competencies, here’s what’s actually worth developing.

  • Understanding how AI systems behave at the level of context, instructions and limitations. You should be able to articulate why the same model produces great work in one context and mediocre work in another, recognize when you’re filling a context window, and spot whether a problem is a prompt problem, a model problem or an architecture problem.
  • The discipline to build reusable structures instead of one-off chats. Every time you finish a piece of work that you might do again, ask whether it could be a skill, a project, a template or a documented workflow. Most of the time the answer is yes, and most of the time it takes less than an afternoon to set up.
  • A working knowledge of what’s possible with API-driven and code-based workflows, even if you never personally write a line of Python. Knowing what’s possible changes what you ask for. A marketer who knows that a script can normalize 300 customer interview transcripts in ten minutes will request that work, while a marketer who doesn’t know it’s possible will keep asking interns to do it manually.

Familiarity with the tools at our disposal today that lets the average marketer codify workflows. Claude Skills and similar systems exist precisely so that institutional knowledge can be captured, replicated and audited rather than living in one person’s head or one chat history. Marketers who get good at building and using these tools will compound their effectiveness over time. Marketers who don’t will keep retyping the same prompts forever and wondering why their output is always lacking a certain je ne sais quoi.

None of this requires becoming an engineer. What it requires is developing enough fluency to make sound decisions, lead AI-powered work credibly and avoid getting stuck at the chat-window ceiling.

Technical fluency gets you in the room. Taste decides what you do there.

Marketing has always evolved. The list of skills we considered essential ten or twenty years ago looks nothing like the list today. PR adapted. Media buying shifted from manually optimized to AI-infused. Creative and content production tooling has been rebuilt end to end.

What’s different about this evolution is that it isn’t specialist-driven. The next era of marketing belongs to people who are technically fluent AND have real marketing context — taste, judgment, the instinct for what actually moves an audience versus what just looks impressive in a deck, which our Editorial Director Annetta Hanna recently covered wonderfully when looking at what happens when everyone can write. 

I keep coming back to one thing when I think about why this combination wins. There are, unfortunately, thousands of recently out-of-work software engineers who, on paper, have every technical skill described in this post. What most of them don’t have is marketing context: the years of pattern recognition, the feel for what works, the taste that has been the hallmark of the discipline forever. Hand them the technical tools and they still won’t know which audience to write for, which message to lead with, or which campaign to kill before it scales.

The future marketing role belongs to people who hold both sides — the technical fluency and the marketing judgment. Technical literacy gets you into the room where the decisions get made. Taste and context determine what you do once you’re there. The most valuable marketer of the next five years won’t be the one who can build the most or the one who can judge the most. It will be the one who can do both, fluidly, in the same workflow, in the same conversation, in the same hour.

That’s the future for those of us willing to embrace it.