What agent do you need to build?
Take one step back with me
There is plenty of advice online on HOW to build your agents..
But there is ZERO advice on WHAT agent do you need to build in the first place.
And the problem is, you or your company will waste too much time and money, building a super technically-advanced-agentic-workflow solution… for the wrong business problem 😭
This is the blind spot that I want to address today with a real world business problem I am working on at the moment.
Do you have 4 minutes?
When do agents enter into the game?
On October 6th Marius and I will start building a semi-automatic trading platform using agentic software.
The idea is simple: you can learn with us some real world LLMOps by building and deploying production-ready agentic software that moves the needle, in our case, in the financial world. This is not a toy, but a fully working system running on a Kubernetes cluster.
And the thing is, there are million “cool” things we could build using financial data.
But that is not the problem.
It is not about “cool” or “not cool”.
It is about asking yourself WHAT is the most impactful thing we can build to impact our client business metrics, in this case a trading firm.
And the thing is, agentic software that works in the real world (at least nowadays) DOES NOT replace human jobs, but accelerates the humans doing these jobs.
This acceleration happens when you
Break down a human job into tasks, and
And identify the bottleneck task. That is, the one task where an LLM/Agent would accelerate the human the most.
Let me show you with an example.
What’s the task your agent needs to solve?
The 3 most fundamental tasks a human trader performs are:
Finding a potentially good trading idea by monitoring the financial markets, and leveraging previous knowledge.
Validating if the data is actually good by running a backtest.
Executing the idea.
Tasks 2 and 3 are well automatized by current software systems. Task 1 is where most of the human cognitive overload happens.
This is essentially what good human traders do. They “blend” their prior knowledge, with the current market situation, to come up with winning strategies.
If we build a LLM powered app that that replicates this behaviour, we increase the number of trading ideas we can test per day, and impact directly the probability of success (and hence profit) for the company.
BOOM.
This is the task we will build an agent for.
Wanna learn hands-on how to build agentic software that works in the Real World?
On October 6th, I will start the first cohort of my new live course "Let's build an agentic trading platform. Together".
We will work super hard for 8-weeks and live code together 50+ hours, to build from scratch a semi-autonomous production-ready agentic system for trading, that we will deploy and operate on Kubernetes.
Along this 8-week journey, you will learn:
How to map a business problem to an AI system solution with a real world example.
Universal design patterns for building real time agentic systems.
Real time data processing, transport (Kafka), and storage (Real time DBs and VectorDBs).
Structured output generation and tool usage to guide agents to success.
Vanilla, RAG and Agentic RAG to increase model output quality.
Tool server design and implementation.
Containerization with Docker and deployment with Kubernetes.
Continuous Integration and Continuous Deployment with Kubernetes.
Production-ready LLM serving.
Tons of Python and Rust tricks.
It will be hard.
But hey, no pain no gain.
Gift 🎁
As a subscriber to the Real World ML Newsletter you have access to a 45% discount if you enrol in the following 72 hours.
Again, it will never get cheaper than this 😉
See you on the other side,
Pau
Very interesting! You are among the rare people who talk about agent without bullshit 👍