03Events and conferences / Conferences

On stage, and in the job.

What applied AI conferences talk about, set against what AI job ads ask for. Talks are read from official schedules and archives with the same rules as the ads.

1,860 conference talks1,937 AI job ads
on stage against in the jobAgents and tool use

59% of talks, 63% of AI job ads. On stage it gets 2.6× the attention it gets in job ads.

Open the skill →

Talks: 1,860 sessions at 8 conferences since June 2025. Ads: AI job ads since 1 October 2026. A talk or an ad counts for every skill it names.

The gap

Where stage and job disagree.

More on stage than in the job

What conferences are excited about before employers ask for it, or what is easier to talk about than to hire for.

More in the job than on stage

What employers ask for that conferences rarely cover: the quiet, everyday work.

Share of attention: each skill's part of all the skills named in talks · in ads, and the difference in points. Long ads and short talks compare fairly this way.

Tools

The tools named most on stage.

  • MCP9.7% of talks · 9.6% of ads
  • PyTorch8.8% of talks · 16.4% of ads
  • RAG4.5% of talks · 12.5% of ads
  • GitHub Copilot4% of talks · 1.9% of ads
  • Microsoft Foundry3.2% of talks · 0% of ads
  • Python3.1% of talks · 62.8% of ads
  • Claude Code2.4% of talks · 7.2% of ads
  • Cursor1.6% of talks · 8.5% of ads
  • OpenClaw1.5% of talks · 0% of ads
  • Hugging Face1.3% of talks · 2.9% of ads
  • SQL1.1% of talks · 13.6% of ads
  • TypeScript0.9% of talks · 17.4% of ads
  • Azure AI Foundry0.9% of talks · 1% of ads
  • Ray0.9% of talks · 1.5% of ads
AI Engineer, year by year

How the topics moved.

The share of each year's recorded AI Engineer talks on each topic, as the conference tags them.

2023202420252026
Coding agents46%36%41%38%
Agent engineering15%11%30%40%
Evals27%27%25%30%
MCP24%22%33%25%
ML fundamentals34%34%25%16%
Enterprise AI22%23%20%18%
Coming up

Official events.

The full events calendar →
Read so far

Every edition, with its source.

Read once a month from each conference's own published schedule or talk archive. Breaks, registration and social events are left out. Only counts, titles, speaker names, organisations and links are kept, never descriptions, biographies or transcripts. The rules are the same as for the job ads, so their error rates apply here too. The method →