From Chat to Coworker: What the Latest AI Shifts Mean for Learning Professionals

Insights from a recent Digital Learning Institute (DLI) webinar with Joe Houghton

AI is arguably the fastest-moving technology most of us have seen in our working lives. Even staying on top of the main tools, such as ChatGPT, Copilot, Claude, Perplexity and Gemini, can feel like a full-time job.

For learning professionals, keeping up matters. In many organizations, it is still unclear who owns AI upskilling. When IT leads it, the focus tends to fall on tools. Learning professionals are well placed to take ownership instead, but that means keeping our own AI skills current.

Drawing on insights from AI educator Joe Houghton, this blog looks at what has changed in AI over the last six months and what it means for educators, learning designers and L&D teams.

Until recently, most AI use happened inside a chat window. You wrote a prompt, refined it through conversation and got more information than a search engine could give you. But the work stayed in the chat.

The biggest shift of the last six months is that AI can now work in your folders and files. It can create, edit and delete documents, and connect to applications such as Notion, Canva, Gamma and Microsoft Office through tools like MCP (Model Context Protocol) servers. AI tools can even take control of your browser and click through websites on your behalf.

The result is AI that does real work, not just answers questions. Asked to prepare a Notion page for an upcoming session, for example, Claude can find previous pages in Notion, pick up their format and structure, and build a new draft without being told every detail.

In effect, AI is becoming a hub for all your other tools. Rather than opening each application yourself, you brief the AI and it works across them. That is a big change in how we behave with technology.

A useful way to think about AI adoption is as a six-stage maturity scale. Individuals and organizations are all on a journey, and it is fine to be at any stage.

1. Basic prompting: asking AI single questions.

2. Multi-shot conversations: going deeper through iteration and follow-up.

3. Connecting tools: linking apps such as Notion, Canva or Office to your AI.

4. Working with your files: letting AI read, create and edit documents in your folders.

5. Building applications: creating tools, apps and finished outputs rather than chat replies.

6. Scaling across the organization: sharing agents and skills across teams of 50, 100 or 1,000 people.

Scaling is the big challenge for many organizations right now. Building skills and agents in your own AI space is relatively easy. But when everyone builds their own, you can end up with 50 research agents instead of one good one that everyone can use and manage.

Should you give AI access to all your organization's data? Many teams are wrestling with this question. Some have decided not to connect AI to every SharePoint and Microsoft 365 folder because of data security concerns.

There are pros and cons. The more AI knows about you, your history, your templates and your courses, the less you have to tell it and the better it gets at helping you. Giving AI access to good data from your own sources also brings hallucinations down, because it does not need to make things up.

Against that are real concerns about GDPR and data security. A practical middle ground is to take baby steps. Rather than opening up a whole server, give AI access to a specific folder and curate what goes into it. That way, you separate what you want AI to see from what you do not.

All the major AI tools have now adopted skills. A skill is a set of instructions that gives AI extra powers, such as your brand colours and logo, or the font and heading styles you want for every document. You tell it once, save it as a skill, and the AI applies it whenever it is relevant.

Skills are quick to create. You do not need to write pages of instructions. In Claude, for example, you can build a branding skill by uploading a logo and pointing it at a website, and it will work out the colour palette itself. If it gets something wrong, you simply tell it to change it.

This is where the real value lies. "If you don't train the assistant, you've just got a generic, reasonably bright assistant," says Joe Houghton, and what it produces looks like everyone else's. Putting time into skills, context and connections to other tools lets AI do far more.

Skills are also becoming portable. They are just small text files, and skills created in Claude can now be used in ChatGPT. ChatGPT is retiring custom GPTs in December and moving to skills instead.

Who owns the skills you create? If you build them on a subscription paid for by your employer, in work time, the organization probably owns the IP, as it does for other material you create at work. In effect, you are packaging your own expertise into a reusable asset.

Across all five major platforms, three shifts stand out: learning modes are now everywhere, AI is moving from chat into agents, and models have made big jumps in capability. Here are some of the developments most relevant to teaching and learning.

• ChatGPT: study mode, quizzes and flashcards, ChatGPT Edu and ChatGPT for Teachers, plus add-ins for Word, Excel and PowerPoint. It is particularly strong for images and infographics. Some features are not yet available in Europe because of the EU AI Act, and availability also depends on your plan.

• Gemini Notebook (formerly NotebookLM): new interactive reports, improved slide decks and video overviews, data tables extracted from your sources, video overviews in 80 languages and more quiz formats.

• Microsoft Copilot: notebooks, study guides and a new Study and Learn agent. Microsoft deliberately stays a few months behind the cutting edge, but Copilot has improved a lot. To get the most from it, turn on Work IQ, use Cowork where it is available and choose the latest model rather than leaving it on auto.

• Perplexity: fast, reliable answers with sources. It now searches Wiley journals and books automatically, and an Education Pro plan adds computer and learn modes.

• Claude: arguably the most capable tool overall right now, apart from image generation. It works as a true coworker, running multiple tasks, building apps and creating shareable interactive artifacts. Add-ins for Microsoft Office put it in the sidebar of Word and Excel.

Many tools now offer free or reduced-price plans for educators and students, so it is worth checking what you are eligible for.

The next stage of maturity is building applications. Rather than chatting with AI, you give it an outcome, such as an app or a finished document, and it goes off, does the research and builds it.

Three examples from Joe Houghton's own work show what is now possible:

• An automated study workflow. For a master's program, Claude collects new course material from the LMS each week, saves it into folders, loads it into the right Gemini Notebook and creates a 20 to 25 minute audio overview for each session, ready to play on the commute.

• An interactive trip planner. A single request produced a shareable, interactive website for a photography road trip around Norway, with a map, a day-by-day itinerary and sunrise and sunset times. It works on phone and computer and can be updated simply by asking.

• An exposure triangle app. In about three hours, a web app was built to teach beginner photographers how shutter speed, aperture and ISO interact, using AI-generated photos that change as learners adjust the settings. It is already being used with camera clubs.

For educators, this opens up a lot. Interactive learning tools like these can now be built by the people who teach with them. As Joe puts it, "You just have to know the questions to ask."

AI is now very powerful and very convincing, but that does not remove the need to check its work. This matters most in a work setting, and especially for accredited teaching.

A few principles help:

• Always verify. Check outputs for accuracy and make sure they are actually what you want.

• Give feedback inside the AI. Asking AI to make your corrections, rather than editing the document yourself, helps it learn what you want next time.

• Know where your data goes. IT teams should be aware that some models, including Claude, sit outside the EU data boundary. If you handle sensitive information, this matters.

• Plan for cost. Most tools now limit usage through tokens. Free plans stop when you run out, while paid plans range from around $5 or $20 a month up to new $100 and $500 tiers.

• Remember sustainability. AI runs on data centre servers, and there is a sustainability issue to consider.

Local AI is also on the horizon. Specialist AI PCs that run models on your desk currently cost between €3,000 and €5,000, and use no tokens. Within two to three years, they may cost closer to €2,000.

• Take ownership of AI upskilling. L&D is well placed to lead it, but only if we keep our own skills up to date.

• Move beyond the chat window. The real gains now come from AI that works with your files and tools.

• Invest in training your AI. Skills, context and connections turn a generic assistant into a useful one.

• Start small with data. Curated folders let you benefit from context without exposing everything.

• Build, don't just ask. Educators can now create interactive learning tools without a developer.

• Stay critical. Verify outputs, understand data and IP implications, and weigh cost and sustainability.

AI will keep moving at speed, and keeping up to date is part of the job. For learning professionals, the opportunity is to lead AI upskilling rather than leave it to others.

To develop your own AI capability, explore DLI's Professional Certificate in Applied AI Literacy.