A lot of what makes software hard is building things that will work in so many scenarios. Are robust to so many types of users and use cases. Every vibe coded saas product will still have to solve this. But every piece of vibe coded internal software will not.

Ominous vibes outside right now.

two thousand contributions on github in last 6 weeks for me

Who will build the first truly great big website? Every big website has fairly bland experience and design because things like the component and asset libraries need to support every scenario, every language. These things then get used by dozens, sometimes hundreds of different authors. Scale becomes the design constraint of everything. And scale systems are almost definitionally generic ones. It gets really hard to invest in things you won’t use a ton. It just objectively doesn’t make sense operationally or economically. Or at least it didn’t. In the past.

The most user hostile thing you can do.. -create a complicated product -where certain actions must be performed in the ui -constantly move stuff around and rename things so agents are always wrong about how / where to do the thing The least you can do is nothing tbqh.

The gdp will only explode when we all realize we should be 10x more ambitious.

What if a topic cluster was more like this?

Working on the web is gonna so much the next few years. There are so many things to rebuild in more compelling ways. You thought informational content was dead or commoditized? Think again. It’s gonna be incredible.

This is my favorite moment.

Like all questions of this kind, the answer is conditional. If we knew if the woodchuck was at smb or enterprise we might get an answer.

I honestly don’t know why I only work in the cli now. I didn’t decide to do it consciously. I am just here.

There’s a new maker/manager schedule with agents. You alternate between deep work/thinking and babysitting your agents and answering their questions.

AEO is the perfect channel for doing some grifts. Weird measurement. Little first party data. Is ~~somewhat~~ a channel aggregator vs a truly distinct channel. Everyone is buying. You can sell anyone anything right now. Does it work? Whos to say. Just add some new prompts.

As a consequence of the below being true, I would go even further and say that building software will be a normal part of every knowledge workers toolkit and job. Documents, spreadsheets, presentations, software. And actually the first three might end feeling more like software too. Excel I think is a good analogy because I suspect to the extent anything is important, understanding data, how to structure it, schema is all you need, etc will be a useful intuition. And people who understand that well today are typically pretty handy in excel. I also suspect the same folks who are handy in excel today will move their files to databases. And then build interfaces on top of that data. There will just an unimaginable amount of software.

Context is a poorly understood concept, especially in marketing. The job of mgmt is never to pass along context, it is to create context. If you think you can give your workflows sufficient context by giving them access to some existing documents, you are wrong. We hired engineers and journalists for our content function because our baseline assumption was that the information required to do anything did not exist and certainly was not packaged in a way that was useful. So we hired people who were talented at closing gaps in context. People who could find answers where nothing existed. We do need to give agents context. But for many use cases it is wishful thinking to believe it is a systems integrations problem. The problem is that the context does not exist. Creating context is the creative act.

One half the timeline is like vibe coding is a waste of time. The other half is like ya long-term we won’t even be doing code review. What I think will happen is these two timelines ultimately converge and the people that stuck with it and became great vibe builders will wake up one day to discover that they are professional developers.

I have always equated search with the offensive line in football. Doesn’t win games by itself. Not the flashiest discipline. But everyone wants to play behind a great o-line. And good luck sustainably winning games without one.

Recently I’ve been building sites on some of the popular headless cms and then built my own cms. A lot of the “cms is dead” dialogue has been focused on authoring/updating blog posts but I think that’s a tier 3 consideration. What has become magical recently is that non-technical builders can create highly expressive sites without using templates or page builders. You can articulate what you want to claude and get custom schemas and frontends in minutes. If you know what you are doing with coding agents at all, it is easier and more satisfying to build even personal sites this way than something like wordpress. The terminal is a weird and uncomfortable place for most people but that wont be the destination. You’ll just tell an ai what you want and you’ll basically get it. Thats what headless cms feels like right now to me. I love it. And the headless cms has many advantages over no cms or using git/file system. A website of any size requires rbac, structured data, input validation, relationships, queries, filters, yes an editor etc. The big updates for me are that I can get a frontend that does anything I want with a fully editable, user friendly input layer. And it’s trivially easy. That if the site is all structured data I can work across the site in big sweeping ways, particularly on the meta layer. That when I need to do anything at scale I can. At a pace that was just not possible previously. All that is driven from coding agents. That’s all the new exciting stuff to me. But single, precise, page level actions? I still want a cms ui. Not necessarily a traditional cms wysiwyg. But just the structured content layer/interface you get with a headless cms. Maybe I won’t always use that. But hard to imagine a world where a business doesn’t need that. Plus what I expect to actually happen is that the cms will become a critical part of a larger platform. And the thing you’ll really be buying (one day) is not content storage and query, but intelligence/agency. CMS is the center of a company’s acquisition engine. It will become more than a dumb storage layer when the dust settles.

Agree with this. Especially early.

I think if you played the last couple years in web/search/growth/content over a hundred times, you’d get pretty close to the same outcome 100 times. Because the basic forces were too strong. AI visibility arrives and is important. But the first party data and overall measurement quality is quite bad by default so it’s easy to game. It nonetheless erodes the impact and profitability of traditional search. AI also brings with it AI for execution. Namely around content and automation (no more legacy manual config of node on canvas). Which is very convenient if you can’t really measure one channel (ai viz) and the economics of the other (traditional search) are compressing. So you wind up with platforms that do ai visibility, automation and content. Maybe you can just hire a tool. Every team got pressured to figure out ai visibility and to adopt ai and so win win for everyone. Except it mostly didn’t work. Most ai visibility setups are deeply flawed. The net quality of tracking setups across the industry would almost certainly be better if they were all just derived from the distinct search intents associated with topics. Maybe with some persona modifiers. Rather than pretending like they are two totally different things. Combine that with people saying you should only care about “best” queries and chatgpt doing ridiculous fanout structures where it stuffs 5 brands and 2026 into a query means the only thing were measuring is a query where the only way to influence it is with a listicle where you put yourself first. We created a stupid game and then followed its even dumber rules. Content was the thing people thought would be easy to automate but turned out to be the hardest thing. Companies that beefed up their growth teams and cut or didn’t invest in content probably got it wrong. And the thing we should have been building were content tools and skills to make skilled authors better. Or get unique insights out of people without the time or natural inclination to write. Not to automate everything and have an editor clean it up. Editing slop is impossible. I look at all this and wonder how little of value was learned or built over the last year or two in pursuit of ai visibility. Almost all of it is one algo, index or fanout change away from being worthless. So 98% of companies would probably be better at marketing if they were following a 2024 playbook. And probably better positioned long-term for ai visibility too. The goal is to deliver value to the audience. And I simply have not heard a lot about that recently.