LTUK 2026 the AI is coming. Ok wow me please.

I had my JoshBot9000 scan all the vendors at the conference (17 had JoshBot blockers, probably not specifically for my JoshBot but who knows) for the most common AI buzzwords, then had it dig into their landing pages to see what flavor of Buzzword Bingo we were going to play this year.

So here it is, 8 terms grouped by where they live in your stack, what they mean in vendor-speak versus what what could actually move the needle (from a getting stuff done in L&D and how potentially it can impact yoru work flow), and some questions to try and dig into with, as in these are the questions im going to be going to the stand with. *Disclaimer: There are vendors on the floor who have amazing solutions and only partially fall victim to the buzzword bingo game rampant in our industry. Also please note that this is my lens on what I want to see and the promise of what can be designed. I have seen tools that can do a version of everything that is described below.

The Experience Layer. What the learner sees and feels. Data, content, and skills are working together. Most vendors use these terms interchangeably; your stack needs each to mean something specific. Here is what im looking for and asking about.

• Intelligent (49 Vendors). Vendor Play: a positive adjective applied to anything algorithmic. What would be really impactful for L&D: a specific decision the system makes from data the user didn't provide. The digital body language behind the click, the dwell time, the abandoned module. Ask: show me the decision, the input data behind it, and the outcome, what makes it intelligent. Think Intelligent tutoring systems.
• Personalized (65 Vendors). Vendor Play: sorting content by job title or department (or other filters). What would be really impactful for L&D: the full personalization loop. Assess, customize, space, adapt, integrate, reinforce. Filtering by role doesn't even hit the first step. Ask: how do two learners in the same cohort diverge by week two, and what signal drove the divergence? Show me the graphs for two learners and show me the personalizations applied to each.
• Skills (58 Vendors). Vendor Play: tagging content with skill labels. What would be really impactful for L&D: operational capability units tied to outcomes, with a defined system of record that feeds your job architecture and your performance system. Treat them like a supply chain. Forecast, decide build vs. buy vs. Bot, measure quality. (Don't get stuck debating skills vs. competencies vs. abilities vs capabilities. Pick the operational definition that works in your org and move.) Ask: how does your skills definition reconcile with our XYZABC framework, and which wins when they conflict? Show me the golden path when your solution works, and what do I get out of it

The Capability Layer. What the AI claims to do. The gap between Vendor Play and your stack widens here.

• AI Coach (21 Vendors). Vendor Play: a chat interface trained on your content library (SharePoint, gDocs, confluent, etc.). What would be really impactful for L&D: adaptive guidance grounded in pedagogy (cognitive apprenticeship, scaffolded practice, sprinkle in some zones of proximal development, and retrieval practice/spaced repetition etc.), running the personalization loop, measured by behavior change on the job. Ask: who designed the pedagogy, what's the evidence base, and how do you measure behavior change versus engagement in the app? Can we map to business-focused KPIs?
• Agent / Agentic (42 Vendors). Vendor Play: automation framed as decision-making. What would be really impactful for L&D: judgment calls within bounded autonomy when the situation wasn't pre-scripted, with defined scope, escalation, and audit trail. An agent orchestrates within your ecosystem. A script just plays its part. Ask: When your agent encounters something outside its training, what does it do? Show me the human-in-the-loop happy path (ideally the human does know what they are doin.

The Integration Layer. Where the AI lives technically. Ecosystem thinking territory. The power is in integration, not isolation.

• AI-Native (16 Vendors). Vendor Play: a product with AI features in the marketing copy. What would be really impactful for L&D: architecture designed around AI from inception. The kind built by Learning Experience Architects deeply versed in Instructional Systems design, not retrofitted by feature teams. Workflows that exist only because the AI is at the core. Ask: when did your product first ship, when did you add AI, and walk me through a workflow that exists only because the AI is at the core.
• LLM (16 Vendors). Vendor Play: "powered by GPT" in the footer. What would be really impactful for L&D: a specific model doing specific work with quantifiable cost, latency, and reliability. LLM cost is a line item that grows with usage. Model it the way you'd model a skills build vs. buy: cost to run, upside, cost of inaction, cost to switch. Ask: Cost per active learner per month at current usage, the curve at 3x, whose infrastructure runs it, and where that infrastructure sits (data residency matters for regulated industries).
• MCP (Model Context Protocol - 12 Vendors). Vendor Play: Position as future-ready to enterprise buyers who care about agent to infrastructure ecosystems. Most vendors are still figuring out how to talk about this on marketing pages, which is why you'll barely see it on the LTUK floor. L&D (it can be used for Agent to Agent but the better play is Google's A2A) What would be really impactful for L&D: an open protocol that determines whether your learning data is addressable from Copilot, ChatGPT, and Claude, or sits in a silo while every other system in the company gets agent-connected. In ecosystem orchestra terms, MCP is whether your sheet music can be read by every conductor in the building (Think of it as the USB-C port that connects to other platforms). Ask: Show me the MCP server endpoint documentation. What operations does it expose, what's the access model, how do you address security, and when did you ship? How is token economics factored in

Ok gang, this is already running too long as usual. Im going to be asking alot of these questions. PM me with anything cool you find out or if someone is doing something awesome, post it below and share the KB article :) Lets not name names unless it is very much a bait and switch (use redit to do that rather than LI. Lets stay positive if we can).

See you at LTUK. (I'll be the one asking for MCP server docs at every booth that mentions agents. Yes, the slightly annoying guy.)