我們演講者如何搞砸了 AI 大會

So, a few weeks ago I was at a conference, sitting in the hallway with a cup of lukewarm coffee and scrolling through the schedule to pick my next talk.

The first AI talk was about orchestrating agents at scale. The next one promised governance for the agentic enterprise, and the one after that was about observability for LLM apps. Then I found orchestration again, this time with "multi-agent" in front of it.

I kept scrolling all the way to the closing keynote, and almost every AI talk had one of those three words in its title or its abstract.

So here's my theory. I'm convinced we can get a speaking slot at any tech conference by adding one of those words to our talk. We pick orchestration, governance, or observability, we put it in the title, and congratulations, we're on stage in June.

Why these words work so well

And I get it, I really do. There are some very human reasons behind it.

First, these words feel safe. AI changes every few weeks, so if we promise something concrete in March, it might look silly by June. "Governance" sounds wise today, and it'll still sound wise next year.

Second, reviewers recognise them. A program committee reads hundreds of proposals in one long afternoon, and when they see one of these words, they think, "Great, this talk belongs to the conversation everyone's having."

Third, they hide the mess. Half the time, what someone means by "observability" is "we finally added a log line that shows which tool our agent called and why." And that's the talk I want to see! It's specific, it's a bit embarrassing, and it's way more interesting. "Observability" is the same story in a suit and tie.

And last, they sound sure. Most of us are figuring out AI as we go, and these words sound calm and certain. That's a lovely feeling when we're pretty unsure on the inside.

The talks I'd happily queue for

Here's what I'd love to see more of, whatever the title says.

I'd love to see an incident with the logs right there on screen. Show us the agent's output, the command it ran, and the Slack thread that followed. One honest screenshot teaches me more than ten slides of principles.

I'd love to see a boring graph. Some team measured review time, bugs, and onboarding before and after an AI tool, and they show us what happened. A small, real number beats a big, imagined one every time.

I'd love to hear someone say, "Last year I believed this, and here's the exact moment I changed my mind." Those talks stay with me for months.

I'd love a live demo that breaks. The speaker debugs it in front of us and thinks out loud, and honestly, that's the most useful twenty minutes at any conference.

And I'd love someone to show us the bill: the tokens, the hours, and the weekend somebody lost. Hardly anyone shows it, and everyone wants to see it.

Two quick checks before you hit submit

If you're writing a proposal right now, here are two little tests that help me a lot.

The first one is the swap test. I take each big word in the title and swap it for a real thing from my codebase or my team. So "Observability for AI Agents" becomes "The Log Line That Showed Us What Our Agent Was Doing." If the new title sounds more interesting, I keep it.

The second one is the Monday test. I ask myself what someone in the audience will do differently on Monday morning. If I can name one action, the talk has a point. If all I can name is a feeling, the abstract goes back into the oven for a bit.

Both take about five minutes, and both make the talk a lot more fun to give.

Let's build a better schedule

So which buzzwords keep popping up on the conference schedules you see? And which talk would you happily queue for? Let's compare notes in the comments. 👇