ConVEx 2026: what Pittsburgh taught me about structured content, AI and the people doing the real work

ConVEx 2026 reinforced a simple truth: structured content is no longer optional. It is foundational.

As the community marked 20 years of DITA, this year’s conference felt both like a celebration and a checkpoint. DITA’s journey from its beginnings at IBM to its use across industries has created a practical foundation for something many organizations are now trying to build at speed: AI-ready content.

That matters because AI does not thrive on content volume alone. It needs content that is structured, governed and reusable. It needs information that is human-readable and machine-consumable. It needs content that carries enough context to stand on its own, with relationships, metadata and intent built in from the start.

Without that foundation, many AI initiatives struggle to move beyond experimentation. The technology may be impressive, but the results depend on the quality and structure of the content behind it. When organizations bring content teams into AI pilots early, they give those initiatives a better chance of delivering something useful, trusted and scalable.

Structured content is the rocket fuel for AI, but content teams are the GPS. They understand audience, intent, terminology, governance and usability. They are trained to turn complex information into something useful for the right audience. Increasingly, that audience includes AI systems as well as people.

The conference experience

Pittsburgh in April was a strong setting for a community that still knows how to have real conversations. CIDM created the conditions for practical exchange, not just polished presentations. The hallway conversations, lunches and informal debates were as valuable as the sessions themselves.

What stood out this year was the shift in the quality of the conversation. We are no longer asking whether AI matters to technical communication. The questions have moved on:

What should be automated, and what must remain human-led?
How do we govern AI outputs in regulated environments?
What does a content professional own in an AI-driven workflow?
How do we prepare existing content so AI can use it safely and effectively?

These are the right questions. They also put technical communicators, information architects and content strategists at the center of the conversation. This year’s event showed that the people who understand structure, reuse, metadata and governance are not on the edge of AI transformation. They are essential to it.

The broader lesson was also reassuring. No matter which framework or toolset an organization uses, the principles are consistent. Content must be findable, reusable, traceable and trustworthy. AI does not change that requirement. It makes it more visible.

RWS LEGO Star Wars challenge: congratulations to the winners

There was also time for lighter moments. All work and no play, as they say, and so RWS decided to run a small raffle draw.

The RWS LEGO Star Wars challenge gave us a fun way to celebrate modularity in what has long been used as a metaphor for structured content: LEGO bricks.

While the entrants all expressed delight at the possibility of taking something fun home for their kids, we all know who is most likely to end up playing with them.

Congratulations to:

Michael Tsurikov, Cirrus Logic

Chris Burright, Marvell

Gary Ragland, Tweddle Group

Most entrants said the sets were for their children, we all know who’s going to end up playing with them.

What I learned from the sessions I attended

Several sessions brought the same message into focus from different directions: AI readiness is content readiness. Sarah O’Keefe’s “AI and Content: Avoiding Disaster” set the tone well.

Her warning was direct. AI has real potential, but rushing into AI-first workflows without governance creates predictable risk. If the inputs are poor, AI does not magically repair them. It scales the problem. It can also leave humans with the worst possible job: cleaning up endless low-quality output after the fact.

The point was not to reject AI. The point was to manage it properly. AI-first should not mean chaos-first. Organizations still need versioning, update strategies, source control, risk models and governance. That is especially true in regulated environments, where the consequence of inaccurate content is not inconvenience. It can be safety, compliance, revenue or trust.

What stood out this year was the shift in the quality of the conversation. We are no longer asking whether AI matters to technical communication. The questions have moved on:

What should be automated, and what must remain human-led?
How do we govern AI outputs in regulated environments?
What does a content professional own in an AI-driven workflow?
How do we prepare existing content so AI can use it safely and effectively?

These are the right questions. They also put technical communicators, information architects and content strategists at the center of the conversation. This year’s event showed that the people who understand structure, reuse, metadata and governance are not on the edge of AI transformation. They are essential to it.

Sarah’s comments on PDFs also landed strongly. Many organizations still treat PDF collections as a shortcut to AI readiness. But a PDF often encodes appearance, not meaning. A human can infer structure from layout. A machine needs that structure to be explicit.

Semantic content matters because AI needs more than words. It needs signals. These signals can only be embedded by adding metadata to granular components used to build the output.

Michael Priestley’s session on designing content for AI took that argument into the practical details of content design. One of the sharpest points was that content must be able to stand on its own when separated from the delivery experience around it.

A topic may make sense to a human when it appears inside a documentation portal, surrounded by breadcrumbs, navigation, product context and related links. But if the words on the page do not carry enough context, an AI system may not know what the content is really about.

That has real authoring implications. Headings need to be meaningful. Short descriptions need to explain the user goal. API and CLI reference content needs natural language context, not just syntax. Examples need to reflect real customer use cases, not only the happy path.

Troubleshooting content becomes even more important because users are increasingly asking AI assistants very specific, scenario-based questions.

The before-and-after example around a billing command reference stayed with me. The original description simply said that the command showed enterprise usage reports. The improved version explained when to use it, which business problem it addressed and how it supported use cases such as departmental reporting and chargeback. Same command. Very different value.

Courtney Mauney and Jenifer Schlotfeldt from IBM presented “Transforming Your Organization to AI-First Content Design,” building on the same theme from an IBM Cloud perspective. What stood out was the practical way the team used existing content standards, style guidance, accessibility guidance and templates as the foundation for AI-assisted authoring.

The goal was not to ask AI to invent documentation standards. It was to give contributors a guided way to create better content faster, using standards that already existed.

This is a useful lesson for any organization. If your standards only live in people’s heads, AI cannot reliably apply them. If they are documented, structured and available in a form AI can use, they become part of the workflow.

The team’s use of prompt libraries and guided authoring support showed how AI can lower the barrier for contributors while still protecting quality.

The IBM session also made an important distinction about where AI is useful and where it is unnecessary. AI can help explain dense technical material, convert SME drafts into structured formats, improve legacy reference content and create a stronger first draft.

But it is not the right tool for everything. Simple search-and-replace tasks, large unbounded file operations and vague “fix this” prompts still need judgment. AI works best when the task is clear, the source material is reliable and the review process is defined.

Eliot Kimber’s ServiceNow session on incremental DITA publishing was a reminder that structured content at scale is a very different challenge from structured content in theory.

ServiceNow manages tens of thousands of topics, many publications, multiple versions and multiple languages. At that scale, keys, reuse, metadata propagation, filtering, flagging, generated links and cross-deliverable references are not edge cases. They are everyday operating conditions.

His framing of the publishing problem as cache management was useful. If a single topic changes, the question is not just whether that topic has changed. The question is how that change affects rendering, relationships, context and deliverables across the publication set.

The most interesting part was the move toward dynamic resolution at delivery time. Instead of treating everything as a static build problem, the approach showed how topics, links, filtering, metadata and taxonomy views could be resolved dynamically in an internal preview environment.

The detail that stayed with me was not the technology alone. It was the authoring impact. Faster feedback changes how teams work. If authors can see changes quickly, they can make better decisions sooner.

Michael Iantosca’s session on agentive workflows added another important layer: precision and determinism. His “precision paradox” is worth remembering. When an AI system is only moderately accurate, people tolerate errors because expectations are low.

As the system becomes more capable, tolerance for error drops sharply. In high-risk domains, 80 or 90 percent accuracy is not enough.

That is why deterministic boundaries matter. Agentic workflows can be powerful, but an error in one agent can cascade through the rest of the workflow.

The more autonomous the workflow, the more important it becomes to define the rules, the context, the checks and the points of human accountability. That is not just an engineering problem. It is a content governance problem.

Lance Cummings’ session on evidence-based prompt design

Lance Cummings’ session on evidence-based prompt design added a final piece. Prompting is not dead. It is becoming part of a broader discipline of context engineering.

Prompts are not just clever instructions typed into a chat box. In production systems, they become structured components embedded in workflows, skills, agents and authoring environments.

That should feel familiar to anyone working with DITA or structured content. Prompts need purpose, context, structure, reuse and validation. They need to be tested against outcomes. They need to evolve as models change.

In that sense, prompt design is another place where content professionals already have the mental model. We know how to structure information so it can be reused, governed and improved over time.

Across all these sessions, the conclusion was clear. AI does not reduce the need for content discipline. It raises the stakes.

Wednesday: my presentation marathon

Wednesday was my own ConVEx marathon: three consecutive sessions, one narrative thread and very little time to breathe. It was ambitious, but it was intentional.

The day started with “Content Automation 101: What to Automate and What Not To.” The message was simple. Automation creates value when it is applied to the right tasks, with the right governance. It is useful for high-volume, repeatable, rules-based work. It is risky when applied blindly to judgment-heavy decisions, sensitive approvals or final sign-off.

The discussion in the room confirmed what I see with clients all the time. The temptation to automate everything at once is real. So is the risk of losing quality, traceability and trust.

Good automation starts with workflow understanding. It asks where repetition creates friction, where rules are stable, where data is reliable and where human expertise must remain in control.

That session set up the rest of the day. It positioned automation as a business design decision, not just a technology decision. What you automate says a lot about what you understand.

The next session, “Tridion Docs. Agentic AI, MCP and System Interconnectivity with DXD,” shifted from strategy to application. Nigel Lock joined me for a live Test Kitchen session showing how AI-assisted and agentic workflows can work when they are built on a structured, governed content foundation.

The demo scenario focused on a fictional MedTech company, Veridian Diagnostics, responding to a regulatory change across multiple product families and affected content components.

The workflow showed how a Tridion AI agent could detect a regulatory change, run a gap analysis against existing content, identify critical gaps, generate suggested remediation for human review and surface status through a single dashboard.

The important point was not simply that AI could move work faster. The important point was that human gates were built into the workflow by design.

The PRRC reviewed the gap analysis, confirmed scope and approved consequential steps. Validation and release required explicit sign-off. That is the model regulated industries need: AI acceleration with human accountability.

This is where the broader conference themes connected directly to the Test Kitchen. Michael Iantosca’s comments about deterministic boundaries, Sarah O’Keefe’s warning about governance and Michael Priestley’s focus on context all pointed in the same direction.

Agentic workflows only become useful when they are grounded in structured content, explicit process and accountable review.

A genuine thank you to Nigel Lock and Wiegert Tierie from the Tridion Docs team for the preparation, product expertise and calm under pressure that helped make the demo work.

Live technical demos always carry a little theatre. The best ones make complexity look manageable.

See more on how Tridion Docs can help your organization:

https://www.rws.com/content-management/tridion/docs-demo-videos/

Listening to content: the AI panel

The final session of the day was the one I was most looking forward to and most excited about: “Listen First: Strategies, Structure, Practice for AI-Ready Content + AI Readiness Playbook.”

The 90-minute panel brought together four people I respect enormously: Marianne Calilhanna from DCL, Jack Molisani from ProSpring Staffing, Regina Lynn Preciado from Content Rules and Sarah O’Keefe from Scriptorium.

Thank you to all four panelists for agreeing to participate and for bringing exactly what I asked for: one sharp story, one tension point and one practical action. They delivered.

Marianne grounded the conversation in the realities of content transformation. Her point about legacy content and PDFs was especially important. PDFs may encode appearance, but AI needs meaning. Humans can infer structure from layout. Machines need that structure made explicit.

That is why content conversion is not a cosmetic exercise. It is the work of making knowledge usable.

Jack brought the career and organizational dimension. His message was direct: content teams cannot wait to be invited into AI initiatives. They need to document their value, claim their role and show how their expertise affects business outcomes.

That is not self-promotion. It is responsible leadership.

Regina gave the room the line that stayed with many of us: listen to your content. It will tell you what it needs.

That is a powerful reframing of the content professional’s role. Writers are not simply producing more content. They are interpreting, governing and improving the knowledge layer AI depends on.

Sarah made the strategic stakes clear. When AI initiatives are treated only as technology projects, organizations risk building on the wrong source material. Content teams are not adjacent to the problem. They are central to the solution.

The discussion worked because it was honest. The panelists challenged assumptions, disagreed where it was useful and kept the conversation practical. Nobody defaulted to the conference-brochure version of their answers.

That is exactly what I hoped for.

The takeaway

I left Pittsburgh more convinced than ever of this: AI will not fix your content. But structured content will make AI work.

The organizations that succeed with AI-enabled content will be the ones that treat content as a governed, reusable and strategic asset. They will bring content teams into AI initiatives early. They will invest in metadata, structure, workflows and accountability. And they will understand that automation is most valuable when it strengthens human expertise, not when it tries to bypass it.

That is the real opportunity for technical communicators, information architects and content strategists. Their role is not disappearing. It is expanding. The work is moving closer to governance, orchestration, validation and business impact.

The skills this community has built over decades are becoming more visible because AI needs them to succeed.

A final thank you to the Center for Information-Development Management (CIDM) for organizing another excellent event and bringing this community together. Thanks also to the Tridion Docs team for the demos, conversations and expertise throughout the week.

ConVEx 2026: what Pittsburgh taught me about structured content, AI and the people doing the real work