AI Before the LLM Boom: Notes from 2019

In 2019, as a student, I went to two industry events and wrote them up: an AI conference run by a large technology distributor with sponsors from the big cloud and chip companies, and Big Data LDN, a two-day expo with hundreds of booths and parallel talks. My posts were mostly notes and enthusiasm (and, in the second one, a frank appreciation of the free socks).

Reading them again in 2026 is a useful exercise in how predictions age.

What the 2019 talks got right

• “Data is the new oil”, with a caveat. The line I wrote down was that data is no good on its own: refined data is the value. That’s even more true now. Models are increasingly a commodity; clean, well-structured, trustworthy data and the context you give a model are the differentiator.
• Inference at the edge. The idea of training centrally and pushing a ready-made inference engine to devices sounded niche. Today small models run on phones and laptops routinely.
• Analytics for non-specialists. One talk pitched search as the interface to data: ask in plain language, get insight. In 2019 that meant fuzzy search over tables. Now it means asking a language model to write the query.
• “Challenge complexity, don’t embrace it.” I copied that quote down from a talk about adoption: a tool that isn’t adopted is useless. It’s the same principle I apply to platforms today (Golden Paths).

What the talks didn’t see coming

• The interface became language. Almost everything I noted was about dashboards, drag-and-drop models and predictive analytics. Nobody I heard was talking about general-purpose language models as the main way people would interact with AI.
• Software development itself changed. AI was framed as something developers would build for the business. Within a few years it was also something that helped developers build, and now agents are users of the platforms we design (What Makes an AI Agent).
• The investment figures I copied down were vendor projections. I’d treat those with far more scepticism now.

Where it fits in my learning

Those events sat between my placement and my final year, where I specialised in AI and machine learning and did my dissertation on topic modelling and network analysis of online communities (Pathways into Cybercrime). The fundamentals from that time, such as how models learn, how to evaluate them and why data quality dominates, are written up in Machine Learning and Natural Language Processing, and they’ve aged far better than the conference slides.

Lesson for reading conference notes

Write down the principles, not the predictions. “Refined data is the value” and “adoption beats sophistication” were worth keeping. The market sizes and product names weren’t.

---

Originally written in 2019 → Arrow - Artificial Intelligence Experience and Big Data LDN 2019