Positive users agree AI prompt education is a useful starting point for building with tools, while negative users dismiss the discussion as unhelpful or pointless.
Most Activity
This is why we created http://paxel.ycombinator.com
Educating people on how to use the AI tools has become a serious bottleneck
@fulg0re What are some things you would like it to do?
@garrytan literally half the companies I talk to are sitting on tools they don't actually know how to use lol
@garrytan I'm trying! But there's a lot more to do. Open to ideas for things to cover.
@garrytan Some form of "do less of this" or "do more of this" to catch obvious bad patterns and encourage good patterns. The pure "this is what we observed" is kind of fun as gimmick but not much value beyond that.
That's what we're building at Grapes. There are two challenges when it comes to people using AI tools: 1) People are bad at saying what they want, or knowing what's even possible with AI. They simply can't pin it down, and we see it day by day. 2) Models are terrible when that intent is ambiguous.
@garrytan It's also challenging to distill what the drug dealers are selling vs what actually makes a difference and should be used by the normal person in the agentic space.
@garrytan 100%
Okay 1. honest evaluation of your project - prompt it 2. dir inventory check & asking to do a deep audit & checks on what are stub works 3. vision vs implementation checks also keeps you align 4. Short sessions with saving your last thread and context in file helps 5. Save everything in .html it might look good interactive readable and have meta info thats easier for model to recall.
For me it solves most problem.
@garrytan History's greatest entrepreneurs all read; it's a common trait well documented in @FoundersPodcast podcasts. While there is a bottleneck, so is there for tons of other areas of life.
@garrytan I heard last week that there are companies offering legacy orgs Agentic coding training 😂. I strongly opposed it, and proposed a continuous learning framework internally which is currently under consideration. There is no way a costly 1 time training is the right way.
@garrytan Everyone keeps talking about AI replacing jobs.
Meanwhile AI is still waiting for half the workforce to figure out how to upload a PDF, bro.
@garrytan the tools aren't gatekept. the mental model is. and nobody's figured out how to teach that at scale without it turning into a prompt engineering course nobody finishes
Looping is not hard. If you can't figure it out on your own, idk what to tell you. It's pretty self evident and not earth shattering. If you've been working seriously with agents, you already figured it out and you're already doing it.
Money to pay for the requisite tokens...now that's hard.
@garrytan You have to start somewhere; only then can you move forward.
@garrytan We’re here to help… 🤓
@leerob @garrytan *whispers in ear*
"goal modeeeee"
@garrytan But Paxel can't read if there is a loop and has GStack inside it. It ends up not recognising GStack plan-eng under plan mode.
@leerob @garrytan Start with the loops
@garrytan disagree. AI applications have a long way to go to improve their UX and show real benefits for average users (not prosumers like developers). that burden is largely on AI consumer application developers not people.