DevLearn 2025 was full of good conversations. As always, it was great to see folks I’ve known for years and to meet people in person for the first time. After a conference, I find it helpful to review my notes and pull out some key points and highlights. It’s easy to go to a conference and feel all the energy but not take anything back to implement it. If I go through my notes, I’m more likely to remember what I learned and take action on it. These notes below are not comprehensive summaries in any way, but rather snippets and highlights.
In the past, I have used AI to help me summarize my notes. It is faster to generate a blog post that way; I can upload all my PDFs of scanned notes from my Rocketbook and let AI pick the key points. But while it’s faster, it isn’t as effective for learning. Part of the reflection process is that messy personal work of reading my notes and picking out important points. If I outsource that to AI, then I’m not learning as much. One of the themes of many sessions and discussions at DevLearn was about being intentional and strategic about using AI. This is one place where I don’t want to outsource my thinking to a machine.
Josh Cavalier Morning Buzz: AI & Ethics
Morning Buzz sessions are informal discussions on various topics. Josh Cavalier facilitated a conversation on AI and ethics.
A few themes from the discussion:
When you use images and voices of real people, get model releases and provide lots of transparency on how their likenesses and data will be used.As you remove tasks from people’s jobs, help them see the path forward with other tasks or skills.Consider energy and resource use.
As part of the conversation, I pointed out that water use for AI is actually lower than other things like golf courses and agriculture. The environmental impact of AI is probably lower than you think; video conferencing and streaming video probably have more impact. (Thanks Emil Heidkamp for the article.) That’s not to say we should ignore it; we need more renewable energy and more investments in our electrical infrastructure overall in the US. But the actual impact seems to be a lot lower than early estimates.
Keith Keating Keynote: From the Trusted Learning Advisor to Hidden Value
Revan’s Law: “For an organization to survive, its rate of learning must be at least equal to the rate of change in its external environment.”
In other words, when the world is changing quickly, organizations have to change quickly too.
One of the key ideas was that rather than trying to get “a seat at the table” as is often discussed with L&D, we can build our own table by starting discussions and bringing people in to talk and help us understand the business better. We need credibility to be trusted business partners.
Questions to ask yourself about credibility:
What have I done to earn credibility? (Expertise)How do I show up as credible? (Behavioral—what do you do)
Markus Bernhardt: AI Unfiltered
A lot of AI gives the illusion of progress. There’s lots of investment, new tools, and demos. But the business results are pretty stagnant. We work in silos. There’s organizational fatigue and eroding trust.
With AI, it’s more about change management and people issues rather than tech problems. This requires a workplace transformation, not a technological transformation.
Clark Quinn: Extending Learning Across Time
I enjoyed Clark’s session on “The Whys and Hows of Spaced Practice.” Clark and I definitely have a mutual fan club going (he mentioned me several times; I often cite him in my own sessions).
What is learning? Not just a persistent change in long-term memory, but a persistent behavior change.
What do we mean by practice? Reactivation and reflection
Different types of reactivation
ReflectionReconceptualizationRecontextualicationReapplication
How much practice do people need? About 6 times, but multiple factors affect that.
How much spacing should we provide?
Ideally, just before people forget, but that’s hard to predict.As a starting point: 2 days, 3 days, 4 days, then weeklyWhen you provide new content or have another event: reset the spacing clock
Karl Kapp and Jessica Briskin: Action-First Learning
This was a fun session where Karl and Jessica put some of the action-first learning principles in practice with a story, comic hero and villain images, audience participation questions, and directions to take notes and focus attention.
Examples do better at convincing people than facts and research.Without reflection, it’s just an experience.
Mike Taylor & Bianca Baumann Morning Buzz: Nudging Behavior Change
This discussion included lots of questions and examples about getting people to change behavior. One example for safety training included asking employees to put a picture of a loved one near where they work to remind them why they follow safety procedures. The emotional connection is more effective and gives them a clear and personal “why.”
Someone had just-in-time training on SmartAccess scanners for frontline workers. That meant it was right in their work environment; a challenge for deskless employees.
Do you need to learn everything? You can watch a video on YouTube and successfully fix something even if you don’t have any long-term memory or behavior change. You didn’t really learn it, but you did perform the task and do the necessary behavior.
Yulia Barnakova Keynote: Architecting What’s Next
I enjoyed seeing how she used AI-generated images to reinforce her theme of sailing into the storm and maintained some consistent colors like the pink/purple in the sky throughout.
She shared ideas for an individual toolkit to grow your own skills and be ready for the future (the “Future-Ready Fanny Pack”). This includes places to start experimenting and piloting.
AI Content Generation: try at least once a monthAI Coaching Tools: try using voice mode in ChatGPT or another LLMIn the Flow Learning AssistantsVibe-Coded SimulationsAI-Ready Learning Mindset
Like a number of other people, I felt some skepticism about her example of her son using ChatGPT after finding a turtle. Using AI to identify the turtle is great—that’s a perfect use. The risk is low if you get the wrong species name. But asking it what to feed a wild animal? The risk is much higher, and I wouldn’t trust AI for that without verification from other sources.
Tristia Hennessey: GenAI Powered Storytelling
Tristia shared a case study of a multi-year training blended training program and how she used AI to create an episodic scifi webcomic. This included 8 characters, which is a LOT to work with in AI.
This session reiterated to me that I need to spend some time experimenting with Freepik. This was all built before Freepik Spaces, so I think some of this work might be easier now than what Tristia had to do.
Some of the challenges she noted were getting consistency in characters in styles, the homogenized “AI-style” characters, the difficulty of generating women without makeup, diverse and realistic features, and hands.
Lessons Learned:
The custom GPT wasn’t worth it. Keep your prompts in one conversation for continuity in the story.The custom characters were a lot of work, but were worth it.Be careful of too much iterating in prompts.Plan to edit heavily.Keep a library of image prompts for environments and characters.Save credits for retouches (every edit and retouch in Freepik requires credits).Experiment with different tools.Aim for a strong but brief story.
Guild Master Panel: AI and the Future of L&D
This was a massive panel discussion with nearly all of the Guild Masters who were present at DevLearn.
Most people using AI now already know how to do their jobs and therefore can judge if the AI outputs are OK. But what about people who don’t already have the skills? How do they judge the output? How do we train people to judge the AI output if they don’t already know how to do the job? (Julie Dirksen)
Tech-enabled bad design is still bad design. (Clark Quinn)
If AI is filling the role of our interns or entry-level people, we need to help people “jump skill” up multiple levels without the slow experience at each level. How do you learn to be a manager if you didn’t do the work previously? (Diane Elkins)
AI is only as smart as the methodology that drives it. (Conrad Gottfredson)
Multiple people commented on the concerns of losing skills and outsourcing too much thinking to AI.
Would you give an intern the ability to publish to prod? Why give an agent that power?
Validation is a big challenge. You need an expert in the loop, not just a human in the loop.
Sarah Mercier & Wendy Morgan: Build-a-Bot Workshop
This was an extremely popular hands-on session. I sat on the floor in the back with a number of other folks when they ran out of chairs. I hadn’t done any experimenting with creating a custom chatbot with ChatGPT before, and I was surprised at how quickly I could get something basic up and running.
My Morning Buzz: Successful SME Engagements
I was worried I wouldn’t have too many people for a Morning Buzz session on the last day, but we had a good crowd and some great discussion.
Questions included:
How do we convince SMEs to do more meaningful interactions than click-to-reveal and adding videos?How do we talk to SMEs about not including everything?How do you handle compliance training where sometimes there are legal requirements about length, legal language, etc.?How do you work with multiple SMEs and conflicting feedback?If you were starting over with a brand new team, how would you do it? What would you do to make the IDs and SMEs both successful?
Keith Keating Fireside Chat
This was an informal conversation and obviously less polished than Keith Keating’s keynote. He shared more of his personal story and how he had to learn how to learn.
Keith doesn’t say “L&D Professional” but “L&D Practitioner.” We need more practitioners rather than influencers and thought leaders. (The irony for me as one of the official “DevLearn Influencers” is not lost.)
We have a low barrier to entry in our field, but that can create damage when people perform poorly. That affects how people view L&D in the long term.
Building relationships—ask people about their learning journey. Did they have bad experiences? Asking helps you understand their context.
Share your experiences, including your failures.
We are problem solvers. L&D is one way we solve problems. If it’s the only way, we’re in a tiny box. Our limits need to be bigger than L&D. Focus on connections and building relationships.
My Session: From Stock Photos to Stunning
My session was in the very last slot before the final keynote. By this point, everyone had already gotten so much information that they were running out of steam a bit. I usually try to chat and warm up the room for 15 minutes or so before I start, but it seemed like people needed a few minutes to regroup instead. So, we took some time to just enjoy the quiet and let what they were learning sink in before I officially started.
I talked about both tools for generating AI images and tips for prompting to get better image results. People asked lots of good questions, and the feedback overall was that this was a very useful session. Everyone got at least one or two ideas or tools they want to go back and test, which was my goal for the session.
Closing thoughts
Obviously, there were lots of discussions and sessions about AI at DevLearn this year. I think some of the AI tools are solutions in search of problems; they’re cool, but they don’t actually make anything better or solve real issues. But the speakers and conversations were more nuanced. Even the people who are pro-AI showed more of the challenges and talked more about the need to use AI strategically and carefully.
I also had no issues finding sessions that weren’t about AI like Clark Quinn’s spaced practice session. I could have made other choices if I had wanted to hear less about AI. So, if you’re thinking about attending next year but turned off by how much AI is mentioned above, know that you could pick other sessions. (You can’t avoid AI in the Expo Hall, but that’s a separate topic.) Last year at DevLearn, I deliberately chose fewer AI sessions and looked for more variety in topics.
I’m looking forward to attending again next year and hoping to make more connections.
