Submitter's message ActionItems:
1. Stan will check with Keith to see if he kept and archive of the sources for Adoption TC CNs.
2. Kris will reach out to OASIS to see if our SVN repos still exist.
===============================================
Minutes of the OASIS DITA TC
Tuesday, 17 February 2026
Recorded by Nancy Harrison
link to agenda for this meeting:
https://github.com/oasis-tcs/dita/wiki/Previous-agendas
Attendance:
===========
Robert Anderson, Stan Doherty, Kristen Eberlein, Nancy Harrison, Scott Hudson, Bob Johnson, Eliot Kimber, Zoe Lawson, Christina Rothwell, Frank Wegmann
Business
========
Regrets: Leroy Steinbacher
1. Minutes
10 February 2026
https://groups.oasis-open.org/discussion/dita-tc-minutes-10-february-2026-uploaded (Harrison, 16 February 2026)
Moved by Kris, 2nd Eliot, approved by TC
2. Announcements
[none]
3. Report from Boston DITA Users Group meeting about "Using AI"
- Bob; lot of active participation, Mike Iantosca did about 60%, they're doing a lot of stuff at Avelara. And we got a report from Kris about work at Mayo,
- Kris; still a lot of concern about it, especiallly about accuracy; e.g., there's a lot coming out of AI, the company fired people to have AI do planning, ended up a disaster. Inaccuracy a real concern
- Stan; great session, got recorded, will publish link, Mike did a great job talking about how DITA provides context.
- Kris; when will recording be available?
- Bob; working on it, so soon.
- Kris; and Stan, I'd love to see your notes on it.
- Stan; having discussion on it here is very important.
- Bob; so there are people like Mike and Eliot, the super early adopters, and a lot of orgs are not even thinking about it yet,
- Kris; it's very hot topic at Mayo, though I haven't been super involved. I'll try to find time to summarize what's happening here.
- Eliot; Aldo and I will be giving a talk at Convex on topic summarization; right now we're not doing anything with AI but generating summaries, 0% in anything else, and our process allows for review of summaries before they're published.
- Kris; lets talk about that in detail as part of agenda item #5.
4. Committee notes effort
Committee Notes.
https://github.com/oasis-tcs/dita/wiki/DITA-TC-Committee-Notes Updated description of mission
https://groups.oasis-open.org/discussion/updated-mission-description
- Stan; started out with 'big comprehensive user guide' but that's impractical, so at our meeting, we decided to focus on 3 audiences, new authors, experienced DITA users, and developers. And we also want to use this effort as a way to recruit new TC members. New goals seem more reasonable.
- Zoe; I'll be available to do stuff after March.
- Christina; I haven't had a chance to work on anything yet, but I have some topics I've been thinking of.
- Stan; we want to treat it as opportunistic and doable. We've identified some stuff already in migration guide area in Github.
[Stan then reviewed his mail - see 2nd link above]
- Kris; so, if Leroy is signing up for migration guide, is he signing up to work with Zoe and Eliot? And Robert, you've migrated DITA spec to use 2.0, that should feed into this, shouldn't it?
- Robert; It was easier than I expected; there were a handful of issues, e.g., link becoming linktitle, otherwise pretty straightforward; we'd already gotten rid of obsolete elements and attributes.
- Nancy; that would be great material for the migration work.
- Kris; we're lucky that our spec is much less complicated in organization than a lot of corporate material.
- Robert; and there's a list of what we did.
- Kris; wrt migrating to 2.0, part of our work on the spec was identifying obsolete markup and removing it. but at this point, anyone starting on DITA is already using 2.0. Also in doing this work, be aware that we have plugins to generate CN output, though they're 1.3 based, so we need to update our plugins for generating PDF and HTML. I've already done PDF plugins, and a good bit of HTML.
- Robert; good news, while in Europe, I tried building a CN, and it worked at first glance, so maybe there's not as much to fix as you might think.
- Stan; that's great!
- Kris; we never had writing guidelines for CNs; and the ones for spec writing are out of date. There are definitely some diffs between specs and CNs. I'll try and see if we have other links to writing guidelines for CNs, and see how they differ from ones for specs.
- Stan; when we've written stuff for Adoption TC, we also had some guidelines; one was that they were done in a more conversational style than spec. And does LwD subcommittee have writing guidelines?
- Frank; no, we don't.
- Stan; so we maybe can write guidelines as we go, bottom-up.
- Kris; great. One thing to be aware of; for some of Adoption TC articles that were written in DITA, the source was in SVN, do we have any access to OASIS SVN repos?
- Robert; I couldn't get to them.
- Stan; Keith may have an archive.
***ActionItem: Stan will check with Keith to see if he kept and archive of the sources for Adoption TC CNs.
***ActionItem: Kris will reach out to OASIS to see if our SVN repos still exist.
- Stan; one last thought; this effort is an opportunity to reach out to folks at Convex, would probably be useful.
5. Continued discussion of how DITA TC members are using AI
- Eliot; wrt AI summaries of topics; in the context of our product docs, we're using AI in 2 ways,
1) in authoring, adding Positron to help authors
2) on publishing side, we've implemented automatic summarizing of topics. After we create HTML, we decide if a topic is a candidate for summarizing; out of 40K topics, we summarize around half of them. Process is to send them to an AI bot, then capture the summary; after we generate the summary, we get them into Github, so they can be reviewed. An alternative would be to have Fluid Topics integrate with an LLM and create summaries on-demand. But instead of that, we do it in-house, which gives us more control.
- Bob; what's in the summary that a shortdesc doesn't provide?
- Eliot; our summaries give you more about the topics, shortdesc is a single short sentence, summaries can be 2-3 paras about the topic.
- Nancy; how do you choose the topics? is the selection manual or automated?
- Eliot; we have an algorithm to go thru the topics, e.g., no tasks or super long tables, etc. Also, we capture tokens that get sent to AI that minimize cost, so if a topic has been recently summarized, it isn't re-done. We currently only summarize for English, out of our 6 languages.
- Kris; can you talk about user experience for this? Do users get to see summarize before they get to a topic? how would a user see a summary?
- Eliot; we don't expose a summary in search results; if you go to a topic and it has a summary, you see a big button that says 'summary'.
- Kris; so it's once a user is on the topic page, then they can see it. So they don't see it in a search?
- Eliot; no, not in search. we don't make summaries available in search. The other way we use AI for product docs, we also have on-demand code summaries. The pre-generation of summaries adds about 1.5 - 2 hours to 3-4 hr. build process, so it makes it longer, but not terribly.
- Kris; so you're set up to review these, but not doing it currently?
- Eliot; not currently, it was important to enable review, because we know that LLMs have really big issues with accuracy. And we put effort into the prompt; Abdul, our AI expert, put effort into that. Something we'd like to do is implement AI search within the doc site; I don't know what that exactly means..., but it's the next thing on our list.
- Scott; also, there's some AI in author experience site.
- Kris; Scott, can you give us more info on AI and Positron?
- Scott; our main reason for AI is that our developers are using AI in development, so that's speeded up, so we need AI to speed up writing. We're switching to once a month release cycle. With Positron/Oxygen, Oxygen had a feature where you could point to your own LLM, but we couldn't use that. took time to get an internal LLM that we could point to. We formed a connection with Anthropic, so we now have an internal LLM we can point to. We have about 13 different use cases where we thought we it could help writing, grammar, alt text, all images accessible, etc. Positron also lets you create your own custom action, e.g. associating metadata that goes with your taxonomy, with Positron we created a custom action that searches taxonomy and inserts custom tags for metadata, for autotagging metadata. And Positron provides a preview capability, so you can accept/reject what the AI is doing. About 70% of automated tagging is useful, - otherwise either putting content in wrong element structure or something else, but 70% generally useful. We're now expanding beyond tech writing to our knowledge base content. using Oxygen's WebAuthor product, which also enables Positron, so we're getting good quality, and we can provide context to engine ahead of time, so we've distilled our style guide that gives prompts. Now, we're thinking how to enhance things so we're not running into other issues.
- Bob; wrt adding style guide to prompts, when Keith used Claude for migrating, it returned some non-valid markup.
- Scott; we had to work with fine tuning prompts, esp if you're doing your own custom ones. Oxygen has a bunch of customizations to help.
- Kris; when you talk about uploading architecture, do you mean grammar files?
- Bob; yes
- Kris; in my experience asking ChatGPT to return DITA, I've seen 'creative' combinations of DITA, DocBook, and other assorted grammars...
- Kris; thoughts about next week...?
- Christina; I have a question about Positron, have you run into any issues with creating topics? What templates are used?
- Scott; wrt basic topics, we have all templates in a custom -framework, any new topics use our local templates.
- Kris; next week, I'd like to hear from Christina about what HP is doing.
11:59am ET close
-- Ms. Nancy Harrison
---------------------------------
Nancy Harrison
Principal, Infobridge Solutions
Nancy Harrison (Personal)
---------------------------------