People code switch. Every ethnic minority has lived experience of this. When you talk to me in Singapore, I sound very different than when I’m in San Francisco. And the way you talk at work is probably very different from how you talk to your kids and how you talk to your lover.
I’ve been dealing with a particularly virulent form of code switching in the simple form of titling blogposts and talks. When you ask the writer face to face what their thing is about, they can give you a perfectly human explanation. And then when they send you their piece, it’s all “Leveraging GenAI with Robust Data Foundations to Transform Businesses”. I’ve left my body by the time I read the first word.
My hope in writing this piece is to put an end to bad titles.
If I sent you this guide, it is not an insult. You just don’t have good title game. That’s ok. You’re great at what you do, title game isn’t your job. I’m the guy that wrote How To Name Things and Two Words. I sent you this guide because your shitty title is the main thing stopping your readers from getting to the good part. If you didn’t have anything good, I wouldn’t bother with a lost cause.
First…
Towards Leveraging Corporate Speak For Maximum Enterprise Synergy
DO NOT:
- Start with “Leveraging” or “Towards” or “Navigating” or “Empowering”
Yeah just stop. There’s no rescuing this when you start there. Just delete meaningless corporate weasel words. Anything >3 syllables is a red flag.
2026 update. Here's some data:
Make It Quotable
If you do not try to make your title roll off the tongue, it probably won’t.
Literally imagine telling your friend about your talk. Telling a new acquaintance to google your blogpost. Having a fan of yours cite it back to you. How much do they stumble? Yeah. This has all happened to me.
There’s an extremely high correlation between what are considered the best talks and the quotability of titles. Simple Made Easy. Inventing on Principle. Our list of best demos/pitches of all time.
Resist colons. It lacks confidence. Why Use Lot Word: When Few Word Do Trick? You’re not High School Musical: The Musical: The Series.
Put the SEO keyword vomit in the description.
Titles are for humans.
Prescribe A Strong Opinion
Here are some real submissions that I see side by side when I review talk titles for the same conference:
How to Build AI Agents that Actually Work
Building enterprise LLM agents that work
Real Agent Evals: Evals that don't suck
Lessons from building GenAI based applications
Working with Coding Agents
I am not at all saying that these talks cannot do well. But if they do well, they do well despite their titles. What you’re doing here is offering a vague promise, and then it’s up to the viewer to click through and see if you fulfil that promise. Through no fault of your own, most talks do not fulfil these kinds of promises, so by extension, the viewer does not expect you to fulfil yours, whether or not you actually do.
Ok, what to promise instead?
Have a strong opinion.
Opinions are one of the most quotable things.
But more importantly, a good Opinion Piece is load bearing because it is eminently forwardable, and therefore viral and impactful. Your Job To Be Done is to be the definitive, conclusive, piece on that Opinion. Stake out territory in Opinion Space, list out all the arguments for, counter all the arguments against, provide datapoints, and watch that go further than anything you’ve ever done.
How to know what’s a good opinion? Three Strikes rule: If you repeat yourself three times, ship it. Even if you’re just quoting others, its fine, credit them and make it your own.
Mental Models and Frameworks are more complex opinions that allow others to scaffold their thought around you. My Radiating Circles and Measuring Developer Relations and Third Age of JavaScript did this.
Opinions contrary to an established opinion are the hardest to do well and the best when correct. What Clayton Christensen Got Wrong made Ben Thompson’s career. Rise of AI Engineer countered the prevailing Prompt Engineer -and- GPU Rich Startup/ML Engineer narrative at the time, with room to grow.
Curiosity Gap Dynamics
Marketers know the power of the Curiosity Gap. Here is where we toy closest with clickbait. I’ll be honest that I wasn’t even sure if I wanted to give it its own subheading, but its a very important concept that you should at least be aware exists, even if you’re going to ignore it.
Strong enough opinions create their own curiosity gaps; you won’t have to try very hard to make it interesting if you got my attention in 4 words anyway.
Question-titles are the strongest curiosity gap creators. One of the top AIE talks of 2025 asked an important question, and hinted at its credibility: “Does AI Actually Boost Developer Productivity? (100k Devs Study)”
That’s enough to cause curiosity. Most of the comments on that video aren’t even about the talk, they are just using the title as an excuse to get something off their chest, which works just fine for the youtube algo. Therefore, ask questions everyone not only wants the answer to, but they have an answer from their personal experience as well.
Curiosity gaps done poorly can lead to “burying the lede”, which make your talk less Quotable and you also don’t want to do that.
Simplest way to put it is; if the most interesting thing about your content is your title, and what people take away after having read your post matches what they would have predicted after reading your title, then you’ve probably said too much and could cut.
Play Madlibs
You don’t have to invent your title from scratch (nor should you always do what I’m about to suggest). Here’s some templates that work well:
{BENEFIT EVERYONE WANTS} without {DOWNSIDE EVERYBODY EXPECTS}
{PRESCRIPTIVE OPINION}, don’t {THING YOU DO}
e.g. Parse, don’t validate
alternative prescriptive opinion: Never {THING YOU DO}
see also title of this post
{POPULAR THING} IS DEAD
Fundamentals of {THING}
straightforward, people love fundamentals
also How {THING} Really Works
You can also flip it and go for the Advanced {THING} tier content
{CONVEY EXPERIENCE/SCALE} with {THING}
There’s loads more templates that work, please suggest in comments and I can add/review.
You can also browse the masters of microcopy like Steph Smith and Shaan Puri, and if you are worried about clickbait, you can see Veritasium’s take or Simon Willison’s Highlights.
also Andrew Chen senpai:
Buzzwords work. Because of a mix of human psychology and algorithmic recsys. Don’t avoid them, just use them judiciously. This involves the ability to read the room, which many smart people sadly cannot without significant involvement in the broader community you are trying to speak to. I don’t see any way around it other than trying to authentically immerse, or consult a friendly who does.
Not all buzzwords are created equal. Databricks/Berkeley tried to push Compound AI Systems, Gartner is pushing Composite AI and Composable business — this resonates with their audience, but not more broadly (see my spiel on Gartner’s Grift).
Pick the shibboleths that work for the community you want. AI Engineer worked for reasons that are well understood now. GraphRAG is working outside of the company that coined it. These buzzwords travel, go ahead and put them in there.
You can kill a good buzzword with bad execution. “ with , and <ANOTHER POPULAR OPEN SOURCE>” is a formula that is easy to do and almost guaranteed to age poorly because it is self centered, and not doing enough work to answer “What’s In It For Me?” for the audience. Your pitch should not be about you.
Buzzwords rise and fall. Try to catch them on the rise — go harder than everyone else on the rise up and regularly test their effectiveness, such that by the time your main buzzword has tapped out you’ve got 2-3 others cooking.
I write this post out of frustration, but also to capture my own thought process. I’m currently procrastinating while writing a Latent Space post.
The original title was Why every AI Engineer needs an AI Gateway. Good, genuine opinion, load bearing. But not super memorable.
My next iteration was “Stop writing model routers, use an LLM Gateway”. Ok, prescriptive, evocative. But didn’t cover the other elements other than model routing. One could shorten to “Stop writing your own LLM Gateway” but that doesnt quite express what people do.
My next one is “It's not your job to DIY an LLM Gateway”. No buzzword bc “your job” is implied (for a Latent Space audience), a little curiosity gap, LLM buzzword, strong opinion. But still not strong enough.
Final one is “LLM Gateway: The One Decision That Removes 100 AI Engineering Decisions”. Leans on Tim Ferriss’ madlbs.
It’s very common for the “tail to wag the dog to wag the tail” - the title is so important to clickthrough and so impactful to the message, that you spend a bunch of time on just the title itself, because that determines the piece you will write, then write the piece, then you go back and tweak the title based on what happened.
Amateur writers think their content has 95% of the value and then spend 5% on the title. Pros will put the value of titles closer to 50% and then act like it.
Make the Talk fit the Title
A very common practice is to word vomit whatever you have on mind into the talk, and then leave it to the end to come up with a title that kinda sorta fits the talk you already decided to give.
No.
Make the biggest, strongest promise you can with your title, and aggressively cut, reformat, flesh out your talk to do justice to that title.
Think about how you can inform, educate, and entertain, and do several rounds back and forth thinking about how the talk should best serve the title you’ve promised, change the title because that’s not quiiiite something you can say yet, change the talk because you changed the title, and so on.
You Should Break The Rules Once You Understand Human Behavior More Intuitively Than Rules Can Model
See title of this post. No better aura than intentional, conspicuous, successful rulebreaking.
