Skip to content

Blog

Follow along as I write about my journey into the emerging field of AI Engineering. Large Language Models (LLMs) open a new category of fuzzy functions allowing us to build software solutions that were not possible before. Regardless of how quickly models continue to improve, there are two categories of techniques that will continue to remain relevant:

  1. Evals - The cardinality of input and output domains of these functions f: String -> String is essentially infinite. This makes the task of measuring quality and performance equal parts difficult and important.
  2. Composite Solutions - We will always be able to achieve more reliable results by composing systems from smaller, more purpose built components.

A Tale of Two Particles

Agentic systems have exactly two fundamental particles: text the model reads, and code it can ask to run. One is stochastic, the other deterministic. Almost every confusion about prompts, tools, skills, and agents starts when you lose track of which is which.

The patterns built on top of these two primitives — filtering, skills, progressive disclosure — are all in service of one problem: keeping the stochastic half on the rails as the system grows. This is the problem domain of context engineering. This is the discrete measurement of agent trajectory.

Improve RAG with a simple technique

Are you burning tokens on irrelevant context? Or losing critical details by sending isolated chunks? There's a better way.

Chunk Windowing

Obligatory AI generated image of chunk windowing

The Problem with Chunks

Once you've matched the top n similar chunks to a query, you could just put each chunk into context by itself. This will work ok, but we can do better! You could put whatever document the chunk came from entirely into context. This is better than supplying the chunk in isolation and works well if all the documents are consistent in size and relatively small. But what if there are really large documents in the corpus? You could end up sending a bunch of irrelevant context to the LLM. Maybe this is somewhat ok too, but it comes with additional latency and cost. If you're looking for a middle ground, you could try chunk windowing.

If you aim to replace humans with AI, you're doing it wrong!

Are you betting on replacing your workers' brains (or your own) with AI? You're wrong! There's a better way to approach AI transformation. To explore the ideas behind this claim as it relates to software engineering, I highly recommend Steve Krouse's talk The Role of the Human Brain in Programming (link below). My key takeaway from his talk is that the proper way to think about the human brain in programming isn't that it's replaced by AI, but amplified by it.

AI Software Factory

Obligatory AI generated image of an AI software factory

AI Engineer World's Fair 2025

The AI Engineer World's Fair conference put on by Swyx and Ben Dunphy is the highest signal to noise you'll find in the sea of AI hype we are all drowning in these days. If you're not familiar with Swyx's other publications, also check out Latent Space and AI News.

I'm fortunate to have attended all three of the conferences in San Francisco and the inaugural event in New York City earlier this year in February. Here are some of my favorite talks, and observations from the latest event held last week in San Francisco.

OpenRouter Rankings

Google has 3 models in the top 5 on OpenRouter this week!

Is All AI Powered Software Development Vibe Coding?

There's a lot of strong opinions on both sides when it comes to the newly coined term "Vibe Coding". I prefer to avoid all the drama. I just like to build software. I think it's important not to throw out the baby with the bathwater. So, vibe coding is a thing now. Ok then, what does AI powered software engineering look like? How is it different from vibe coding?

I think it all comes down to autonomy. First, let's take a look back at the last 2 years of my AI powered software development journey.

Vibe Coding

Obligatory AI generated image of vibe coding
  • in tools
  • 3 min read

My Current Go-To Tools and Services

These are my favorite tools/services as of April 2025. Please note: I use a Mac and some of the software below isn't available on other platforms.

Retro Futuristic Tools

Obligatory AI generated image of tools

Cursor IDE

The ways I use Cursor have evolved over the year and a half I've been using it. I absolutely love the latest update introduction of agent mode. Writing code has never been more fun.

Superwhisper

Superwhisper STT (Speech-to-Text) is the best I've used. If you aren't using a tool like this daily, you are missing out.

Local No-Code MCP Agents

Do you keep notes and other configuration files on your local system under a git repo? Do you neglect committing and pushing because it feels like a chore? In this post I introduce one of my favorite new use cases for local, no-code agents using Anthropic's popular MCP standard.