Have you ever sat there trying to come up with use cases for all of these insanely powerful AI agents that are hitting the street every day? Be careful. The most obvious stuff is usually just a nice-to-have. A vitamin.
The pattern I see everywhere: smart leader gets back from a conference, fired up, starts brainstorming "what could we point AI at?" By Friday the whiteboard has 18 ideas. By next quarter three got built. And nobody can quite explain why revenue didn't move.
Because nobody asked the only question that matters:
Was that problem actually killing us — or was it just sitting there?
"The most common error of a smart engineer is to optimize a thing that should not exist." — Elon Musk
Most of the AI being deployed inside enterprises right now is a vitamin. It's fine. Might help. Nobody dies if you stop taking it.
Painkillers get bought. Vitamins get forgotten.
So the question I keep asking before pointing AI at anything:
- Painkiller? Real, expensive, recurring problem. Worth paying a consultant.
- Vitamin? Technically faster. Old way wasn't broken.
And here's the part nobody on LinkedIn wants to say:
We've run businesses on email and spreadsheets for 40 years. Not because we were too dumb to upgrade. Because they work. Cheap. Universal. The switching cost is higher than the savings.
If it ain't broke, don't fix it.
Musk's 5-Step Algorithm (Strict Order)
He uses this at SpaceX and Tesla. The order matters more than the steps.
- Question every requirement.
- Delete any part or process you can — even if you have to put it back later.
- Simplify and optimize.
- Accelerate cycle time.
- Automate. ← LAST.
Step 5 is where almost everyone fails.
AI is a very smart engineer. It will happily optimize your inherited approval workflow, your 1995 lead-routing process, your six-level expense form — and never once ask whether any of it should exist.
You end up with a beautifully efficient version of yesterday's bureaucracy.
So What IS Worth Automating?
The painkillers. Three filters:
- High volume.
- Genuinely painful.
- Underlying process is something you'd defend on a whiteboard.
Everything else — leave alone. Or better: delete the process and don't replace it.
The Bottom Line
The painkiller list is shorter than you think.
Most of the AI being shoved into your org right now is a vitamin somebody is selling to feel productive. Step back. Question every requirement. Delete first. Automate last.
If the process wouldn't survive a whiteboard defense, don't automate it. Kill it.
(And then read the next one — The CTO's Line in the Sand — for what to do with whatever AI is left after the cull.)
What I'm Watching
- Walter Isaacson's Elon Musk — Chapter 67, "The Algorithm" — the only place Musk's 5-step Algorithm is documented from the inside. Question, delete, simplify, accelerate, automate. It's the operating system behind every Tesla and SpaceX cost takeout. Worth the price of the book.
- Klarna walks back its AI customer service — the canonical "we automated the wrong thing" case study. They replaced 700 reps with an AI, declared victory, then quietly started rehiring. The AI did the easy 70% beautifully. The painful 30% was the whole job.
What I'm Building
- Salesnado.com — outbound sales as a painkiller, not a vitamin. Every B2B company with a quota is bleeding cash on cold outreach that doesn't work. We replace the SDR team with an AI agent that does the painful, high-volume part — without dragging along the six tools, three CRMs, four spreadsheets that should have been deleted years ago. The painkiller is the agent. The vitamin is the SaaS stack we left behind.
- Essentialist.io — the same logic at the infrastructure layer. Clean tool surface. Deterministic dedup. No surprises in the data. If your AI deployments feel like vitamins, the platform underneath is usually why.
Got a take on this? Drop a comment on this issue at theaiplaybook.com/issues/painkillers-not-vitamins — I read every one. Push back, share where you've drawn your painkiller line, tell me where this is wrong.
Need something? Just reply.
If you need any of this, I can help:
- A CTO sounding board on a specific AI decision — 1:1 with me. Just reply and I'll send the link.
- A dev team that actually ships — worked with mine 10+ years, free intro.
- Enterprise-level media services that get proven results — free intro.
The rest of my stack
- salesnado.com — your own AI sales agent
- emailnado.com — launch a fleet of agents, tournament-bracket the winning message
- agents.essentialist.io — have an agent? point it here