The pillars of a painkiller AI app

Everyone is building apps today. It has never been so easy. Just learn some prompting and that's it, right? Sure — until your (most likely wrapper) AI app becomes a feature inside one of the giants. It has happened. It will keep happening.

There's a name for this: getting Sherlocked.

Your AI app needs to be a painkiller, not a vitamin. Nice-to-haves get Sherlocked. Need-to-haves get paid.

But how do you build a painkiller? I've been building for a while now, and here is what I've learned that separates the apps clients can't live without from the ones they forget by Friday.

Fix the friction nobody defends

The instinct is to automate the impressive work. The real opportunity is to automate the work nobody is proud of doing.

Imagine a young intern starts at a law firm. They're smart, they use AI, and they build a workflow that reads incoming client emails, analyzes the request, and prepares a brief for the lawyer. It looks great on paper. It saves hours.

Then they show it to the lawyer.

First problem: privacy. A law firm cannot let client data flow through a third-party model. Second problem: what happens when the AI hallucinates? Isn't it safer to let the senior secretary, who has been doing this for fifteen years, handle it? Third problem, the one nobody says out loud: the lawyer bills by the hour. Saving them time on the work they bill for is saving them revenue.

The intern goes back to their desk. They watch their colleagues work. They notice everyone wastes ten minutes a day searching for documents because the shared drive is chaos. Files named final_v2_REAL_final.docx, folders nobody owns, no tagging system. People complain about it constantly. Nobody fixes it because nobody is paid to fix it.

The intern builds an automation that organizes files by tags, names, and dates. Now we're talking. No client data leaves the building. No expert is being replaced. No billable hour is being eliminated. Just a frustration everybody had and nobody wanted to own.

That's the shape of a painkiller. Not the workflow people are proud of, but the friction nobody defends.

Context is not what you think

You think context means feeding more data to the AI, right? Dump the document in, get a smart answer out. That's not how it works.

We all know AI can hallucinate. But that's not the real problem. The real problem is more interesting.

You get your blood test back from the clinic. You upload the PDF to ChatGPT or Claude. The model has real data — actual numbers, not abstract symptoms. It gives you a confident answer. Maybe asks a clarifying question or two.

But the AI doesn't know if you smoke. If you drink. If your father had a heart attack at 50. If you've been on a cutting diet for three months or eating takeout for ten years. If you go to the gym or sit on the couch watching Netflix. The data on the page is real, but its meaning depends entirely on who you are. Same numbers, two different people, two different diagnoses.

This is why the doctor you've been seeing for twenty years can tell something's off just by looking at you. They have data the chart doesn't.

Your app needs the same thing. Not your context but your user's context. The history, the preferences, the constraints, the patterns. The boring details that change everything.

There are tools for this: RAG, memory layers, wikis, glossaries. Markdown and vector databases are your friends. But the tools are the easy part. The hard part is deciding what's worth remembering, and asking for it without making your user feel interrogated.

A painkiller app builds a relationship. A vitamin app runs a query.

Sell to businesses, not consumers

This one is harder to swallow, but it's the truth.

Unless you're a former senior LLM engineer at Google who also spent three years at Palantir teaching neural networks to sound human, nobody is going to care about your consumer AI app. Here's why.

Consumers are drowning. They already have N8n, Lovable, Gemini, Claude, ChatGPT, every Lovable clone, and twelve other free tools they signed up for last month and forgot about. Their phone is a graveyard of apps they don't remember installing. Yours will join the pile by Friday.

Now add the supply side. Meta, Microsoft, Amazon, Cloudflare, Google, Oracle — all of them are laying off. If you think you're smart, imagine what a senior engineer with ten years at Meta and a layoff package can build. And can you walk into an investor meeting and pitch your idea? Because plenty of those people can, and many already have. You're not competing for users. You're competing against hundreds of Lamine Yamals for a place on the squad.

Businesses are different. They're not chasing the latest model release. They're stuck on running old software full of their entire history, weighed down by bureaucracy nobody has the appetite to rewrite.

The company that makes toilet paper? That's your ideal customer.

I live in Andalusia. Andalusians don't like apps. They like WhatsApp. Send ten emails to ten businesses and wait a day for an answer. Send a WhatsApp message and you'll have a reply in an hour. That's the gap. That's where the work is.

The law firm from Pillar 1 won't switch off the giant legacy system that owns their entire workflow. But the lawyer will happily pay a monthly fee for something that organizes their files and saves them four hours a week. The small business in Marbella will not learn a new app to automate their customer support, but they would absolutely pay for something that replies on WhatsApp to their multilingual customers without them lifting a finger.

Consumers want the coolest thing. Businesses want the boring problem solved.

Signals, not analytics

Stop staring at your dashboard. Bounce rate, exit pages, keyword breakdowns. That's the autopsy. You want the heartbeat.

Look at how your users actually use the app. What frustrates you in the apps you use every day? I was talking to a friend who runs engineering at a big company, and I asked him what bothered him about an internal tool. His answer wasn't "the model is wrong" or "it's slow." He said he felt overwhelmed by the UI. Sometimes he just wanted to find a button.

This guy ships software for a living. If he's lost in your UI, your average user gave up three screens ago.

If you can talk to your users, talk to them. If you can't, or if like me you'd rather build than schmooze, there are tools for it. Microsoft Clarity. PostHog. They show you where users stop, where they click, where the mouse hovers, where they rage-quit. That's gold. Your bounce rate isn't a number to optimize. It's a symptom of something you haven't looked at yet.

This post is based on my experience building products — one I sold, though it didn't make me rich — and a lot of time observing and talking to small businesses. It's incredible what you can learn from a business owner just by asking "Hey, how was your last week?" and shutting up long enough to listen.

Now go build that lawyer automation tool that's going to make you rich. 😉