Agentic, tokens, inference: what they actually mean

"We're moving to an agentic model, but are worried about token burns during inference." If you've nodded along to a sentence like that, this one is for you. Six of the working words of AI, one at a time: the model, training and inference, the prompt, tokens, the harness and the agent, and then that sentence again, translated.

The companion episode, Seven AI words to hold your own with: https://youtu.be/JUg6TTRAgZ0

The Substack article, with the sources:
https://www.aitransition.com.au/p/i-tried-copilot-it-was-rubbish?r=4jl4m9

Chapters
00:00 Intro
00:38 Model
02:08 Training and inference
03:39 Prompt
04:50 Tokens
06:12 Harness
07:40 Agent
09:07 Summary

Sources
- Vaswani et al., Attention Is All You Need (2017): https://arxiv.org/abs/1706.03762
- Shakespeare, Othello (c. 1604)
- METR, Measuring AI Ability to Complete Long Software Tasks (2025): https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/
- Russell & Norvig, Artificial Intelligence: A Modern Approach (1995)