@alanonai: Five videos of me telling you why building alone is hard is enough. Here's the thing you can actually do, and it takes about two minutes at the end of the turn. You get your AI to tell you what kind of work it just did. That's it. That's the whole thing. And I know how small that sounds, so let me tell you why it works. The obvious version is that you notice the switches yourself. Decide what you're working on, tell the AI, stay in your lane. It doesn't survive contact with a Tuesday. You sit down to fix a button, and forty minutes later you're deciding which plan a feature belongs on. You couldn't have called that in advance, because you didn't know it was coming. That's what makes these expensive. They're unannounced. So flip it. Your AI is right there in the session. It knows what you just asked and what got decided. It's got a clear view of the crossing, because it isn't the one carrying seven kinds of work in its head at once. Your job drops from noticing to confirming. That's a much smaller ask at the exact moment you've got the least to give. So you tell it, once, at the start: track the decisions I make, sort each one into business, product, design, technical, build, verify, or operations. Use only those seven. Don't invent new ones. Don't interrupt me while I'm working. At the end, give me each decision with which part of the product it applies to and why I made it. And then the line that does more than all the rest of it put together. If I made a decision without seeming to notice it was a decision, say so directly. That's the one that catches the pricing call you made at eleven at night while you were moving a button. Three rules or it falls apart. Keep the list closed at seven, or it'll invent forty labels and you're back where you started. Decisions only, never the execution, or you've got a transcript nobody reads. And get it out of the session, because a summary that lives in a chat dies with the chat. Most days this gives you three or four things. That's the right amount. It should feel almost too small. Article's called "Tell the AI Which Hat It's Wearing." Link's in the comments. #ai #business #vibecoding #coding #programming
Alan on AI
Region: US
Friday 28 August 2026 13:45:58 GMT
Music
Download
Comments
TheCardashian :
For data architecture…. It looooves generating way more data fields than what is necessary. 😭. Bloated columns then essentially adds to context bloat. Downstream it then ends up referring to the wrong field to perform a calculation etc. Just witnessed this with my team, vibecoders were not staying in the loop themselves…. 😭. Had to do some serious refactoring.
2026-09-01 00:12:14
1
Curiosity Reborn :
My issue is I’m fairly sure I’ve infected my ai with my ADHD,
2026-08-30 18:30:11
2
a7md :
Yeah it's context window Like when for example I give it file if I want a general review it covers that just looks onto a few fields but if you really want A good result no matter what your prompt is take a section of it app and give it to ai then other section maybe even individual files It is better because Ai Is becomes bad when you inject everything at once Not about hitting the context limit all the models are relatively large but yhr working context is kinda low tbh
2026-08-28 15:24:06
1
user360774731 :
I have a roadmap in a separate document in Google Drive. AI must refer to the roadmap before building to see whats been done and whats to come. Instructed to work in order.
2026-08-28 15:25:51
1
sunset140 :
i will give warning always when halusenation happens
2026-08-28 15:33:28
1
Alan on AI :
https://planningroom.tech/articles/tell-the-ai-which-hat-it-s-wearing
2026-08-28 13:47:11
0
Christian Moya :
love it
2026-08-31 15:28:36
1
nobody :
I never or have I ever seen someone do really good work when focused. Most people need things they can move between to prevent burnout on a topic. Everyone is trained through a multi subject system called school yet business tries to put someone’s head down one track. Well I guess it depends on the person
2026-08-28 18:08:29
1
To see more videos from user @alanonai, please go to the Tikwm
homepage.