The Founder's Guide to Blogging

For one brief, deliriously delicious moment in the 2010s, cupcakes were the center of the dessert universe. Sprinkles installed 24-hour cupcake ATMs in six cities, each one playing a jingle while a mechanical arm delivered a $4.50 treat. Crumbs Bake Shop went public on NASDAQ, with the stock peaking at $13 a share. The world craved one thing above all else: baked circles topped with six tons of powdered sugar.

Then, suddenly, poof.

Crumbs' stock fell from $13 to four cents. All 65 stores shuttered overnight. Sprinkles, the company that invented the cupcake vending machine, closed permanently on New Year's Eve 2025. By making cupcakes ubiquitous, the industry had gorged the public on the product until everyone's taste refined past what most shops could deliver. Only the best cupcake shops have survived.

This is now happening to writing in tech.

AI has made prose cheap. Every seed-stage founder with a pulse and a ChatGPT account can produce a competent product update in 90 seconds. Just give an agent access to your codebase, prompt it with "write an email summarizing recent updates we've made and the value they'll offer users" and away you go. This should be the employment apocalypse for the fools who love to wrestle with the written word.

And yet the companies winning in attention marketplaces are investing more in their written product communication, not less. Why? Because when the floor drops to zero, when the world is filled with shitty cupcakes, the ceiling is the only thing that matters.

What a bad feature update looks like

You already know this post. You have written this post:

"We're excited to announce [Feature]. Here's what it does. [Screenshot or a GIF if you are feeling fancy.] We can't wait for you to try it!"

If you stripped away the logo and brand colors, it would be impossible to tell which company in your market wrote that update. If you can swap your name for a competitor's and the post still reads fine, you have written the cupcake equivalent of a grocery store dessert. It's fine. Serviceable, even. It's just that nobody will care about it. All of those expensive tokens, all that overpriced engineering time will be greeted with a collective shrug as your customers move on to more interesting things.

Here are three things you can do to immediately improve your startup's public-facing writing.

1. The why matters over everything

Your customers do not care about your features. I need you to really sit with that and feel it in your bones. They. Do. Not. Care. Most people today have dozens of software products, all of which are releasing updates on a near-weekly cadence. The drumbeat of "new feature!" notifications has become white noise, a hum they have long since tuned out. To get them to pay attention, you have to connect it to why you built it.

For an example of this, see how Linear launched Linear Agent this past March. They open the post with a thesis: as execution accelerates, the bottleneck in product development shifts toward judgment about what to build and where your time and tokens are best spent. The product (an AI agent) was the consequence of that strategic claim.

You can easily imagine the version ChatGPT would spit out: "We're excited to announce Linear Agent: it is not just a new agent, it's a new way to delve into work."

In defense of that phrasing, it is accurate. It's also boring. There is no strategy, no stakes. Customers are betting on a startup's products to wrench them into the future. If you just list out what you released, there is no forward momentum to it.

The key test is this: if your product update could have been written by anyone with access to your feature spec and a chatbot, the update has failed. It should contain insights or secrets that only you know. The customer conversation that sparked the build, the vision for what this unlocks next.

2. Have an opinion

When 37signals launched HEY, it would have been easy to announce "a new email client with filtering capabilities." Jason Fried went with the more challenging approach of opening with an argument about the broken state of email: "You started getting stuff you didn't want from people you didn't know. You lost control over who could reach you. An avalanche of automated emails cluttered everything up. And Gmail, Outlook, Yahoo, and Apple just let it happen."

Good products are built off of spiky opinions. Why should good product announcements be any different? The feature was called The Screener, which forced every first-time sender into a purgatory where the user was empowered to decide whether to ever hear from them again. DHH later said The Screener was "a pretty controversial feature to mandate internally" but it became "perhaps the highest profile feature since launch."

You can disagree with HEY's take on email. I certainly do. But you cannot confuse it for someone else's product announcement. Most product updates hedge everything. You can feel the softening of the language, the committee of product managers, marketers, and sales staff that qualified every sentence. Great writing is expensive because it has a point of view the author is willing to defend. Cheap writing reads like it was reviewed by four people who each added a caveat.

You'll note there isn't anything about AI tells in this section. LLMs shine in helping you discover what your opinion is, even if the writing they actually suggest is usually mediocre. The only caveat is that they make it incredibly easy to convince yourself into any position. Use AI tools to make sure your argument and opinion are clear. Beware the temptation to use it as a source of validation.

3. Show what you chose not to build

This is my personal favorite, and a perpetually underused tool in the writing tool kit. Write about what you didn't launch. By walking your users through the very hard trade-offs every product represents, you create a story. Customers view your product as something built with real risks and choices. It draws them in.

When Superhuman built their AI features, CEO Rahul Vohra wrote about the decision process in detail. "There was no shortage of ideas," he wrote. Should they automatically summarize long emails? Prewrite drafts for you to review? All great ideas that all ended up being rejected. Vohra's reasoning was that those features are all preemptive. The user sets them up once and they run automatically. Preemptive features cost more to run and must meet a higher quality bar. "Imagine if automatic summaries or prewritten drafts were low quality. You would tune them out, or just turn them off," Vohra wrote. So they started with on-demand features instead.

I distinctly remember getting this email. Despite the daily drum beat of feature releases from my personal software suite, I remember admiring the choices they made, even though I didn't necessarily think they had made the right one.

There is a temptation to be Steve Jobs with feature releases. You descend from the mountaintop, enrobed in the black turtleneck of product geniushood, and tell your customers what they want. Unfortunately, you are not that guy. It is much better to show your customers you are human, the hard choices you struggled with, and where you landed. And again, this type of email is something an AI draft will never produce on its own, because AI was not in the room when you decided to ship one thing instead of another.

How to actually use AI to do this

Everything above can be summarized in one sentence: good product writing contains information that only you know, written in a way that only you could write. The practical question is how to get that information out of your head and into the post, using AI tools, without the result reading like it was written by a chatbot with access to your Jira board.

Mess over spec. If you give Claude your release notes and say "write blog post good," you will get a steaming pile of meh every time. Instead, give it all the background data. Pull in Slack threads, customer support tickets that sparked the idea, anything raw. Then ramble into it for five minutes in voice mode about why this blog post matters in the context of the company. AI can put it all together better than you can.

Use AI to find your argument. Ask: "What is the strongest case for why we built this instead of the three other things we could have built?" Use the answer to find your angle. Then rewrite until it sounds like something you'd say to a smart friend. If you would be comfortable sending it as a board update without rewriting first, you've messed up. (Or your board is probably a little too chill.)

Write ugly first. The best AI-assisted product writing starts with a terrible human draft, full of half-thoughts and run-on sentences, and uses AI to clean it up. This is the opposite of what most people do. Starting ugly and polishing with AI produces writing that sounds like a person. Starting clean and trying to add personality produces writing that sounds like a chatbot in a top hat.

Your company can write better than it currently is. Use this post to get started.