A PM's Guide to AI Agent Architecture

Stop trying to treat these things as more than they are. Stop trying to be clever. These models are the single most complex things ever created by humans; the summation of decades of research, trillions in capex, and the untold countless hours of thousands of people smarter than you and I. You will not meaningfully add to their capabilities with some hacked together reasoning workflows. Work within the confines of what they can actually do; anything else is complete delusion.

I have a hard time determining if you are in support or critiquing the article. I'm 60% confident it is a critique (I jest, a play on the content :) ).

This is a nonsensical opinion by a person who doesn't know what they're talking about, and probably didn't read the article.

These models are tools, and LLM products bundles these tools with other tools, and 90% of UX amounts to bundling these well. The article here gives a great sense of what this takes.

> This is a nonsensical opinion by a person who doesn't know what they're talking about, and probably didn't read the article.

Ok, but can you please make your substantive points without putting others down? Your comment wouold be fine without this bit.

https://news.ycombinator.com/newsguidelines.html

The AI bundling problem is over. The user interface problem is over. You won't need a UI for your apps in a few years, agents are going to drive _EVERYTHING_. If you want a display for some data, the agent will slap together a dashboard on the fly from a composable UI library that's easy to work with, all hot loaded and live-revised based on your needs.

You must be an easy person to market to.

Who maintains that UI library? Or does the AI create it on the fly too? Why even bother with a library at that point? Just do a bespoke implementation.

I use agents to do so much stuff on my computer, MCPs are easy to roll so you can give them whatever powers you want. Being able to just direct agents to do stuff on my computer via voice is amazing. The direct driving still sucks so they're not a general UI yet, and the models need to be a bit more consistent/smarter in general, but it'll be there very soon.

The library will exist to maintain high quality/consistency and reduce load times. Also, it's faster to generate a page with parameterized components than to recreate all the components. It's a win all around from an engineering perspective, and nobody has to maintain them, there could be an artifact registry where people publish their components and you or AI can just select nice ones for the given use case.

why would anyone want more non-determinism than absolutely necessary?

I won't use agents for everything. Why would I expect tasks to use agents for everything? This is like saying everything is on the web. No, there is substantial number of things on the web, but not everything.

Contrary to my other comment, I 100% agree to this.

What do you do with agents?

I use them as an intelligence layer over disk cleanup tools, to manage deployments/cloud configs, I have big repo organization workflows, they can manage my KDE system settings, I use them as editors on documents all over my filesystem (to add comments for revision, not to rewrite, that's not consistent enough), I use them to do deep research on topics and save reports, to look at my google analytics and seo data and suggest changes to my pages. Frankly if I had my druthers I wouldn't use a mouse, the agent would use visual tracking (eye/hand) along with words and body language to just quickly figure out what I want.

Why are people publishing their components when the UI problem is over and no one builds UIs anymore?

The Juicero moment for software

Tell me you don't want to go hands free and have the star trek computer do everything for you. We could be there in ~5 years.

We also could have warp drives next year!

Except that the main blocker on the star trek computer is the hooks we wire into the agent to manage the computer. Current gen models are almost smart enough, though their long context support and ability to use tools are a little shaky in general (I have walked a lot of agents through using tools, correct shell command use needs more RL for sure). None of this is outlandish advances, it's all just the natural progression of the track we're on.

A widget != a UI. I don't need a stripe app, but things like visualizations are still useful. I want to be able to pull up a graph of my sales on stripe over the last 72 hours using a specific type of plot, cross referenced with my promotions in a dashboard side by side with consistent colors so it's easy to scan. The agent will be able to pull high quality plots of the right type that theme according to my preferences and slot into my dashboard neatly, and I won't have to hassle with stripe or my adtech or analytics or any of that except to configure the agent.

You’re either a decent troll, or absolutely delusional.

You’re saying you’ve found a useful assistant for menial tasks. That’s not consistent with the strong claims you were making upthread.

My claim is that the "useful assistant for menial tasks" is the Wright brothers flyer to what we'll have in a few years. If you have voice chat with an agent on your phone that can just do everything you'd need an app for, what's the point of an app? And it's gonna happen, because if your app doesn't let people's agents handle their business and your competitors' does, people are gonna switch if they can. The computer interfaces of the future are going to be made for agents first.

I genuinely do not want this, it sounds like shit

> they can manage my KDE system settings

Why do you even have KDE installed if AI has replaced GUIs?

> My claim is that the "useful assistant for menial tasks" is the Wright brothers flyer to what we'll have in a few years.

I agree with that.

But what you originally wrote was, "The AI bundling problem is over. The user interface problem is over." It would probably make more sense to say "...will be over."

People tend to be sensitive to those kinds of claims because there's a lot of hype around all this at the moment. So when people seem to imply that what we have right now is much more capable than it actually is, there tends to be pushback.