A PM's Guide to AI Agent Architecture

https://www.productcurious.com/p/a-pms-guide-to-ai-agent-architecture

by umangsehgal93 • 11 months ago

208 48 11 months ago

barbazoo• 11 months ago

Confidence calibration: When your agent says it's 60% confident, it should be right about 60% of the time. Not 90%, not 30%. Actual 60%.

With current technology (LLM), how can an agent ever be sure about its confidence?

esafak• 11 months ago

I was about to say "Using calibrated models", then I found this interesting paper:

Calibrated Language Models Must Hallucinate

https://arxiv.org/abs/2311.14648

https://www.youtube.com/watch?v=cnoOjE_Xj5g

fumeux_fume• 11 months ago

The author's inner PM comes out here and makes some wild claims. Calibration is something we can do with traditional, classification models, but not with most off-the-shelf LLMs. Even if you devised a way to determine if the LLM's confidence claim matched it's actual performance, you wouldn't be able to calibrate or tune it like you would a more traditional model.

ramesh31• 11 months ago

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.

tomrod3• 11 months ago

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 :) ).

sixo2• 11 months ago

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.

dang• 11 months ago

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

CuriouslyC8• 11 months ago

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.

dfsegoat• 11 months ago

I'm typically pretty critical of PM oriented pieces, but I found this to be a decent overview of how to reason about building these systems from first principles + some of the non-tech pain points + how to address them.

ricardobeat2• 11 months ago

What does the PM title even mean at this point? It's a bit surprising to see a deep dive into technical architecture - though there is massive value in understanding what's involved - as a PM responsibility, this is more in TPM (technical program manager) land which is a different job.

In my book they ideally focus on understanding scope, user needs and how to measure success, while implementation details such as orchestration strategies, evaluation and making sure your system delivers the capabilities you want in general, are engineering responsibilities.

charcircuit• 11 months ago

This post does not do a deep dive into technical architecture.

ricardobeat2• 11 months ago

When it comes to LLMs and agents, this is the technical architecture: orchestration patterns, memory, context handling, MCP, etc.

MangoToupe• 11 months ago

The PM's role is to whip devs until the requirements are met. That seems apt here. Even if the requirements make zero sense

gillesjacobs• 11 months ago

Nice framing for PMs, but technically it is way too rosy. MCP is real but still full of low utility services and security issues, so “skills as plug-ins” is not production ready. A2A protocols were only just announced this year (Google, etc.) and actual inter-agent interoperability is still research grade, with debugging across agents being a nightmare. Orchestration layers (skills, workflows, multi-agent) look clean in diagrams but turn into brittle state machines under load. LLM “confidence scores” are basically uncalibrated logits dressed up as probabilities.

In short: nice industry roadmap, but we are nowhere near robust, trustworthy multi-agent systems yet.

gabriel666smith2• 11 months ago

The idea of giving an LLM with a tool any kind of control over an actual user's account remains (though you put this more elegantly) batshit insane to me.

Even assuming you've correctly auth'd the user contacting you (big assumption!), allowing that user to very literally prompt a 'semi-confident thing with tools' - however many layers of abstraction away the tool is - feels very, very far away from a real-world, sensible implementation right now.

Just shoot the tool prompts over to a human operator, if it's so necessary! Sense-check!

fny• 11 months ago

I really don't understand how people given access to a pile of tools and data sources and unleash them on customers. It's horrible UX in my experience and at times worse than a phone tree.

My view is that you need to transition slowly and carefully to AI first customer support.

1. Know the scope of problems an AI can solve with high probability. Related prompt: "You can ONLY help with the following issues."

2. Escalate to a human immediately if its out of scope: "If you cannot help, escalate to a human immediately by CCing bob@smallbiz.co"

3. Have an "unlocked agent" that your customer service person can use to answer a question and evaluate how well the agent performs in helping. Use this to drive your development roadmap.

4. If the "unlocked agent" becomes good at solving a problem, add that to the in-scope solutions.

Finally, you should probably have some way to test existing conversations when you make changes. (It's on my TODO list)

I've implemented this for a few small businesses, and the process is so seamless that no one has suspected interaction with an AI. For one client, there's not even a visible escalation step: they get pinged on their phone and take over the chat!

risyachka• 11 months ago

> really don't understand how people given access to a pile of tools and data sources and unleash them on customers

It’s pretty simple. When a non-tech person sees faked demos of what it can do - it looks epic and everyone extrapolates results and thinks AI is that good.

small_scombrus• 11 months ago

Doubly so if the person deciding what gets implemented doesn't really get what their staff actually do.

LLMs ability to give convincing sounding answers is like catnip for service desk managers who have never actually been on the desk itself

myhf• 11 months ago

The purpose of customer support is to convince the customer that it is not worth their time to pursue support. A worse experience achieves that goal faster.

Using GenAI is a huge breakthrough in this field, because it is a socially acceptable way to tell someone you don't care about their issue.

politelemon• 11 months ago

You've articulated it better than I could. I think, reading through this author's post, they've misunderstood the objectives.

The purpose has been achieved, in that there is a large drop rate. The product manager has met their goals, cut costs, and might be looking forward to their bonus.

It would be far more expensive to make the LLM behave effectively than it would be to do nothing. Any product manager that sincerely cared about customer support wouldn't be inflicting a personalised callous disregard for service. Instead they'd be focusing on improving documentation, help, and processes. But that's not innately quantifiable in a way that leads to bonuses, and therefore goes unnoticed.

android521• 11 months ago

A lot of the agent tools/frameworks don't dare to have an agent on the site to answer user questions. For those who dares, it sucks. eg. Mastra.ai is supposed to be a framework for building agents but their agent on the website cannot answer any question ( i asked ~20 questions and got 0 satisfactory answers)

gabriel666smith2• 11 months ago

I MVP'd one of these (a basic sequence of LLM customer support 'agents') at my last job, I guess spring 2024. So much has changed since then!

'Routing through increasingly specialised agents' was my approach, and the only thing that would've done the job (in MVP form) at the time. There weren't many models that would fit our (v good) CS & Product teams' dataset of "probable queries from customers" into a single context window.

I never personally got my MVP beyond sitting with it beside the customer support inbox, talking to customers. And AFAIK it never moved beyond that after I left.

Nor should it have been, probably - there are (wild, & mostly ineffable) trade-offs that you make the moment you stop actually talking to users at the very moment they get in touch. I don't remember ever making a trade-off like that where it was worthwhile.

I _do_ remember it as perhaps the most worthwhile time I ever spent doing product-y work.

I say that because: To consider a customer support query type that might be 0.005% of all queries received by the CS team, even my trash MVP had to walk a path down a pretty intricate tree of agents and possible query types.

So - if you believe that 'solving the problems users have with your product' = 'making a better product'. then talking to an LLM that was an advocate for a tiny subset of users, and knew very intimately the details of their issue with your product, that felt really good. It felt like it was a very pure version of what _I_ should be to devs, as any kind of interface between them and our users.

It was very hard to stay a believer in the idea of a 'PM' after seeing that, at least. As a person who preferred to just let people get on with things.

I enjoyed the linked post; it's really interesting to see how far things have come. I'm surprised nobody has built 'talk to your customers at scale', yet - this feels like a far more interesting problem than 'avoid talking to your customers at scale'.

I'm also not surprised, I guess, since it's an incredibly bespoke job to do properly, I imagine, for most products.

majormajor• 11 months ago

I enjoyed the linked post; it's really interesting to see how far things have come. I'm surprised nobody has built 'talk to your customers at scale', yet - this feels like a far more interesting problem than 'avoid talking to your customers at scale'.

This sounds hard to pull off in a very similar way to getting good data through surveys.

I generally don't want to talk to my tools. If I'm motivated to talk to you, it's probably because something went wrong. And even if I talked to you when not annoyed, I'd struggle to articulate more than "it's working good" at any given moment - when what you really want as a product person is to know "it's working good, but I had to internalize this workaround for something for my use case that no I don't even think about but originally I found offputting and almost bounced because of" or whatever.

jbmsf• 11 months ago

As an engineer, I like this framework but can think of approximately zero PMs who could use it to build a product.

pluto_modadic• 11 months ago

reading this as a security engineer trying to get ahead of misguided PMs who buy into the AI hype and don't know 1) it's immature 2) it's not secure & 3) whether their business use case is viable for the R&D we're about to put into it.

I get the feeling there's going to be either 1) a great revert of the features, 2) a bunch of hurried patches, or 3) a bunch of legacy systems operating on MCP v0.00-beta (metaphorically speaking)

:lol_sob:

harryf• 11 months ago

We need to take the focus off cost savings. None of this tech is anywhere near mature enough to replace humans yet.

Far better to focus on enhancing human capabilities with agents.

For example while a human talks to a customer on the phone, AI is fetching useful context about the customer and suggesting talking points to improve the human conversation.

One example of a direct benefit for business using AI this way is reducing onboarding times for new employees

A PM's Guide to AI Agent Architecture

Related Stories