Process and power at twenty: Write yourself into the AI shift

On June 6, 2006, I defended a PhD dissertation in Rhetoric and Composition called Process and Power: Building Strategies for Technical Communicators to Effect Organizational Change. Twenty years later, I keep coming back to the argument at its center, because the industry keeps proving it.

The claim was this. When a company changes how it builds products, it disrupts who holds power on the team. My case at the time was the move from waterfall to iterative software development processes like agile. That shift didn't just change schedules and ceremonies. It rearranged the relationships between engineers, managers, and the people responsible for language and information. In that disruption, I argued, a gap opens. The people who work with words and meaning can either get written out of the new process, or write themselves into it.

I leaned on Foucault for the theory: power isn't a thing you hold, it's a relation that flows and shifts. I leaned on Brenton Faber for the mechanism: change is a discursive project, built and rebuilt through what a team decides can be said and done. And I leaned on both Faber and Johndan Johnson-Eilola for the stakes: professions built on product knowledge were giving way to professions built on process knowledge. The writers who understood the process, not just the deliverable, were the ones who could move from the edge of a project to its center.

I studied this at IBM and at Braun Corporation. I watched writers do exactly what the theory predicted. One writer at IBM volunteered to drive the scenarios for a team trying a new design method, and her manager came to see her as a leader. Not because her title changed. Because she stepped into the gap the process change had opened.

I didn't know it then, but I was describing the shape of my own next twenty years.

The agile shift I studied did open a gap, and writing fields moved through it. Technical writing and information development remain, but UX writing, content strategy, and content design — a discipline that sits inside the design process instead of documenting after it — appeared. That move from documentation to design was, for many of us, a real gain in status and scope. It wasn't the only path out of the gap, and not every writer became a content designer. But the direction in my case was unmistakable: upward, and closer to the center of the work.

That path is my own. I went from leading information development and UX writing at IBM, to building content design as a discipline at McAfee, to leading it at SAP. Each step happened in the seam of a process change. The dissertation wasn't a prediction. It was a map I've been walking ever since.

The biggest shift yet

Now the biggest process shift of my career is here, and it's being driven by the emergence of AI.

Software is starting to generate its own language. Interfaces are assembled on the fly. Answers come from models that few people on a team fully understand. This is the same disruption I wrote about in 2006, at a far larger scale, and it puts the same choice in front of everyone who works with content and meaning. We can be written out as the machine takes over the words, or we can write ourselves into how these systems get built.

But the scale is different in a way that matters. The waterfall-to-agile shift changed when we worked and who we worked with. The AI shift changes what the product is made of. The product now produces language on its own, in real time, in front of the user. That raises the stakes on every decision about words and meaning, and it scatters those decisions across the whole system instead of parking them at the end of it.

Here's why I think this shift offers writing professionals even more than the last one did.

Where meaning lives in an AI system

When content was the user manual at the end, our value was easy to underestimate. You could ship the product and add the words later. In an AI system you can't, because the words are the product, and they're generated from decisions made long before anyone sees a screen.

Patricia Sullivan, my dissertation advisor at Purdue, and James Porter argued in Opening Spaces (1997) that changes in writing technologies change the way we write. AI is the most consequential writing technology yet, and it changes not just how writing happens but where it lives in the system.

It helps to see where meaning actually lives. There are roughly five places, and content work runs through all of them:

What the system knows and alignment around the words it uses. Every model answer rests on a vocabulary and a set of definitions. If an organization uses three words for the same thing, or never decided what a term means, the model will pick one at random or invent its own. Trust breaks on the first answer.

What it pulls in before it answers. A model is only as good as what it retrieves, and whether the right information is even findable depends on how it's structured and labeled. Get that wrong and the model fills the gap by guessing.

How it's told to behave and talk. The instructions behind the scenes and the voice on the surface decide whether output is helpful or merely fluent — whether it confirms before it acts, knows what it shouldn't answer, and answers the real question rather than the literal one.

How it assembles what you see. More and more interfaces and answers are stitched together from parts at runtime. Without a content standard holding them together, you get something that might be correct in pieces but is incoherent as a whole.

What it finally shows a person. The surface is still designed. It isn't a finishing step you sprinkle on at the end. It's where everything above either adds up or falls apart.

Every one of those is a content problem wearing an engineering costume. The work is more varied than it has ever been, and it touches more of the product than it ever has. We are the glue, more than ever.

What the machine can't do

And the thing the machine can't do on its own is the thing our field has always done: be the advocate for the user and make the meaning hold together. A model can produce fluent text all day. It cannot decide, on its own, whether that text is true to the user, consistent across a hundred screens, and worded so a person can trust it. That is rhetorical work. That is judgment.

This is the part that doesn't get automated, because it isn't a writing task. It's a decision about what's true, what matters, and what a particular person needs to hear in a particular moment. The better models get at producing words, the more valuable the judgment about those words becomes. Fluency is now cheap. Meaning is not.

The four moves, translated for this shift

The four strategies I ended the dissertation with still hold, almost without translation. Here's what each one looks like right now and where to start as a writing professional today.

Document your role. In 2006 this meant getting written into the process documents before the new process hardened around teams that forgot to include you. Today, AI workflows are being defined in real time, and almost none of them name an owner for content and meaning. Find where AI features are being spec'd in your organization and get a content role into the definition of done — the system's terminology, its voice, the words users actually see, and test/review steps for what the model produces. If there's an AI working group, ask to be in it before the patterns set. A documented role now is also what protects you at the next shift.

Show your value. AI makes content quality measurable and visible in a way it never was. Tie your work to the things leaders already lose sleep over: accuracy, trust, consistency, hallucination, brand and legal risk. Frame meaning work as risk reduction and trust-building, not polish. The most persuasive thing you can bring to a skeptical room is one concrete example — a place where a vague or inconsistent term produced a wrong or untrustworthy answer, and what it cost. That's your business case, and it writes itself once you go looking.

Join the team building it. The people deciding how the model behaves, what it retrieves, and how it talks are often engineers with no content lens — not because they don't care, but because no one told them it was a discipline. Right now, unknown unknowns are an issue everywhere. Get into those rooms. Volunteer for the high-value work others find tedious: defining the controlled vocabulary, writing the voice and conversation guidelines, setting the criteria for what "good" output even means. Twenty years ago, I watched a writer volunteer to drive the scenarios on a team that was transitioning to agile and become a recognized leader. The same move is available now, and it's up to you to convince those rooms that having you there is a help, not an impediment to their work.

Drive the change, don't just survive it. Iterative methods came with their own artifacts — scenarios, story cards, personas — and the writers who used them did better work. The AI shift has its own: prompts, guidelines, evaluations, guardrails. Use them. And don't write a style guide as a document nobody reads. Build the standard into the system as constraints the model is actually held to, so a rule set once is enforced everywhere the system generates. Your craft becomes the rules the system writes against. Craft is the upstream of the system.

The fear is back — and so is the answer

There's an irony I find almost too neat. In 2006, the loudest worry in my field was that reusable content stripped of its rhetorical context would make writing look like assembly work and invite companies to outsource it. The same fear is back with AI, at a much larger scale. The answer is the same as it was then: context is the value.

I don't have to make that point alone anymore. Enterprise software companies are saying it out loud. SAP's published AI architecture calls context "the moat" — the durable advantage once the models themselves become a commodity. That's the bet I made in my dissertation twenty years ago, now sitting in a leading software company's strategy. The teams that win will be the ones who can author and govern meaning, not only generate text.

None of this is guaranteed. A gap can be walked through from either side. If the people who work with language don't step into these decisions, someone else will make them by default, and we'll spend the next decade cleaning up output instead of shaping it. That was true of the agile shift too, and the field that exists today is proof that enough of us chose to act. The AI shift is the same test, with more on the line and more to gain.

Twenty years on

So, here's where I've landed twenty years later. The argument aged well. It got more urgent, and more hopeful as writing (and now design) professionals. Every process shift opens a gap. AI has opened the widest one I've seen. The last shift moved content earlier in the process. This one makes content and meaning part of what the product itself is. For the people who work with language, meaning, and the user's experience of both, this is not the moment to watch from the sidelines.

To my word people, if you take one thing from this, make it a single move this week: find the place in your organization where AI decisions are being made without a content voice in the room, and put yourself in it. That's how the gap is walked through.

I made that case to a dissertation committee in 2006. I'm making it to design and content leaders now. Content is not the polish at the end. It is not some words you bring someone in to clean up. It is not the thing you add when the real work is done. It is the meaning the product runs on. The leaders who understand that will define this shift. The ones who don't will be defined by it.

Sources