BBC Newsnight's 2017 visit to Microsoft AI research labs offers a useful historical lens — enterprise AI looked very different before the generative wave.
Filmed in 2017, the segment captures a moment when enterprise AI meant research breakthroughs, custom models, and long development cycles — not the copilots and LLM (large language model) assistants teams deploy in weeks today.
Worth a watch if you want context for how far the field has moved — and what hasn't changed (clear use cases, data governance, human oversight).
At a glance
- The segment shows Microsoft Research's lab culture — long-horizon experiments, not product roadmaps
- Enterprise AI in 2017 meant specialized models; today's SMB path starts with operational pilots
- The human element — researchers, ethics, real-world constraints — remains central even as tools democratized
- Research horizon vs operations horizon: both matter — only the second delivers value this quarter
Watch the segment
https://www.youtube.com/embed/jnOjJMbEODA
Why this still matters in 2026
Research labs explore what's possible. Operations teams need what's useful. The gap between those two shrank dramatically after 2020 — but the discipline of starting small, measuring outcomes, and keeping humans accountable didn't disappear.
| Research lab focus (2017 era) | Practical SMB focus (2026) |
|---|---|
| Novel model architectures | Structured pilots on repetitive work |
| Multi-year experiments | 4–8 week measured pilots |
| Custom training pipelines | Validated outputs with human review |
| Publication and patents | Hours saved and error rates reduced |
When leaders ask "where is AI heading?", I point to two anchors: what labs explore (this BBC piece) and what your team can pilot this quarter without betting the business. The second is where I spend most of my consulting time.
Where you are
You've reached the end of Perspectives. Previous: State of the AI — my 2018 talk. Continue with Progressive AI for SMBs.
Curious what AI could do for your operations this quarter — not in a research lab? Let's talk about one measurable pilot.