I’ve been talking to a lot of people about what they’re building with AI. Different companies, different roles, different experience levels. And I keep noticing something: everyone builds the same things.
Not the same code. Not the same stack. The same ideas. The same problems. Over and over, independently, without coordinating.
It starts with the meal planner.
The meal planner
I don’t know what it is about meal planning, but it’s the universal first project. Everyone does it. I’ve seen at least five people build some version of “tell me what to eat this week and generate a grocery list.”
Sometimes it’s a meal prep assistant. Sometimes it’s a recipe finder that accounts for what’s already in the fridge. Sometimes it’s a full weekly planner with nutritional targets and shopping list sorted by grocery store aisle. The ambition varies. The impulse is identical.
And honestly? It makes perfect sense. Meal planning is annoying, repetitive, personalized, and low-stakes enough that you don’t mind if the AI gets it a little wrong. It’s the perfect first project. You’re not going to crash a server or lose customer data if Claude suggests too much chicken.
The knowledge base
The second thing everyone builds is a searchable knowledge base. Your company has docs scattered across Confluence, Google Drive, Notion, Sharepoint, Slack threads, and that one Google Doc from 2019 that everyone references but nobody can find.
So everyone’s first “real work” AI project is the same: “What if I could just ask a question and get an answer from all of our docs?”
I’ve seen this from engineering teams, support teams, HR teams, and at least one person who built it for their homeowner’s association bylaws. The surface area varies. The frustration that drives it is universal: our information is everywhere and nowhere.
The Slack plugin
Third on the list: Slack plugins. Summarize this channel. Catch me up on what I missed. Answer questions from our docs without leaving Slack. Translate this thread. Remind me about things people mentioned.
Slack is where work happens, so naturally it’s where everyone wants AI to show up. The specifics differ but the pitch is always some version of “I shouldn’t have to read 200 messages to know what happened today.”
The email triage
“Sort my email.” “Draft responses.” “Tell me which of these actually need my attention.”
Everyone’s inbox is a mess. Everyone thinks AI can fix it. Everyone builds some version of an email triage system. I built one. My MCP servers post was basically about this. It turns out that “read my email and tell me what matters” is a shared human wish that has existed since the invention of email, and now there’s a tool that can plausibly do it.
The todo app
This one’s a little different because developers have been building their own todo apps since before AI existed. It’s a rite of passage. But AI gives it a new dimension: “What if my todo app actually understood my tasks and could prioritize them, break them down, or suggest what to work on next?”
The todo app is the “Hello, World” of every new technology. Databases? Build a todo app. React? Build a todo app. AI agents? Build a todo app, but this time it talks back.
Why we all build the same things
We’re all humans. We eat food, we lose track of information, we drown in messages, we get too much email, and we have more to do than hours in the day. These are universal problems, and the moment a new tool shows up that might solve them, everyone reaches for it at the same time.
There’s something comforting about that. In a world of increasingly personalized AI experiences, we all start from the same place: the stuff that annoys us most in daily life.
But there’s something else going on here that I think is worth sitting with.
The build vs. buy inversion
Every single one of these projects, the meal planner, the knowledge base, the Slack plugin, the email triage, the todo app, already exists as a product you can buy. There are hundreds of meal planning apps. There are enterprise knowledge base products. There are Slack bots and email management tools and task managers for every possible workflow and personality type.
And yet, when people get access to agentic AI tools, they don’t go looking for a better app to buy. They build their own.
Why?
Because the version you build is the version that actually fits. The meal planner you build knows you don’t eat shellfish and that you shop at Costco on Sundays and that your kid won’t eat anything green. The knowledge base you build knows your company’s actual folder structure and terminology and which Google Doc from 2019 is the one that matters. The Slack plugin you build does exactly the three things your team needs and none of the forty things some vendor’s product team decided to ship.
Off-the-shelf products solve the general case. They have to, since they’re built for everyone. But the problems on this list aren’t general. They’re deeply personal or deeply specific to your organization. The gap between “what the product does” and “what I actually need” has always existed. We just accepted it because closing that gap used to cost too much.
What changed
There’s an old engineering anti-pattern called “Not Invented Here,” or NIH syndrome. It’s the instinct to reject external solutions and build your own, even when something perfectly good already exists. For decades, fighting NIH was a sign of engineering maturity. “Don’t reinvent the wheel.” “Buy, don’t build.” “Your competitive advantage is not your internal todo app.” Smart leaders stamped it out. And they were right to, because building custom software was expensive and the result was usually worse than what you could buy.
That’s what makes this moment strange. The instinct that was wrong for thirty years might be turning rational.
For most of the history of software, “why would we build that when we can buy it?” was almost always the right question. Building custom software was expensive. You needed developers, designers, infrastructure, maintenance, time. Buying a SaaS product that got you 80% of the way there was the rational choice, even if the last 20% never quite fit.
That math is shifting. Not for everything. I’m not suggesting anyone build their own CRM or accounting system. But for a certain category of problems, the ones on this list, the personalized, repetitive, workflow-specific ones, the cost of building a custom solution just dropped to nearly zero.
A meal planner tailored to your family’s actual eating habits? An afternoon. A searchable knowledge base that understands your company’s docs? A day, maybe two. A Slack plugin that summarizes exactly the channels you care about in exactly the format you want? A few hours.
When building is that cheap, “why would we buy that when we can build one that actually fits?” starts to be a reasonable question. Not always the right one. But reasonable in a way it never was before.
The commoditization layer
Here’s the part I keep chewing on. If everyone independently builds the same five things, what does that tell us about the product landscape around those problems?
Maybe it means there’s a huge market for AI-powered meal planners and knowledge bases and email triage tools. That’s the obvious read. Go build one, sell it, get rich.
But maybe it means something else. Maybe these categories are getting commoditized from the bottom up. Not by a competitor with a better product, but by the end users themselves, building exactly what they need with tools that didn’t exist two years ago.
If your head of sales can build a custom dashboard in an afternoon, the bar for what a vendor needs to offer just went up dramatically. It’s not enough to solve the general case anymore. You have to be significantly better than what someone could build themselves in a day, better enough to justify the subscription, the onboarding, the features they don’t need, and the missing features they do.
I don’t know where this lands yet. It’s possible that these DIY versions stay scrappy prototypes that people eventually abandon for “real” products. It’s also possible that for a meaningful chunk of use cases, the custom-built version is good enough and stays good enough, and a layer of software that used to be worth paying for just… evaporates.
If you think that’s hypothetical, the stock market doesn’t. When Anthropic announced legal workflow plugins for Claude Cowork, including contract review, NDA triage, and compliance workflows, Thomson Reuters dropped 18%. RELX, the parent company of LexisNexis, had its steepest single-day decline since 1988. A Goldman Sachs basket of software stocks fell 6%. In total, roughly $285 billion in market value disappeared in a day.
Not because the tool replaced lawyers. It explicitly doesn’t give legal advice. But because the market looked at a general-purpose AI tool moving into a specialized workflow and immediately did the math on what that means for every company selling specialized workflow software. If the AI layer can handle contract review well enough, why does a law firm need a six-figure annual license for contract review software?
That’s the meal planner thesis at enterprise scale. The same logic that makes you build your own meal planner instead of downloading an app, because the AI-built version fits better and costs less, is the same logic that made Wall Street reprice an entire sector in an afternoon. The category of problem is different. The dynamic is identical.
The shared starting line
The meal planner isn’t original. It doesn’t need to be. It’s the shared starting line. And the fact that we all start there, independently, instinctively, says something about both what we need as humans and what just became possible as builders.
What happens after the meal planner is where it gets interesting. That’s where people’s actual problems, the ones specific to their work, their team, their industry, start to surface. But getting to those problems usually requires building the meal planner first. You have to learn the tool on something you understand before you can point it at something that matters.
So if you just started using AI and your first instinct was to build a meal planner, congratulations. You and everyone else. Welcome to the club. The good stuff starts after.
Sources: Stock market figures from Bloomberg, CNN Business, and ComplexDiscovery.