Sign in

Isaac Flath

@isaac-flath.bsky.social
82 followers 16 following 126 posts

isaacflath.com

PostsRepliesMedia
Isaac Flath @isaac-flath.bsky.social · 08/03/2026
I visit my own website because it makes me happy. AI wrote every line of code, but I directed everything. If a thoughtless prompt can make what you make, why would I come to you? I can prompt AI myself This post shows the core skill for using AI: taste isaacflath.com/writing/how-...
isaacflath.com
How I Built My Website | Isaac Flath
Read "How I Built My Website" - Technical writing by Isaac Flath on software development, AI, and machine learning.
000
Isaac Flath @isaac-flath.bsky.social · 25/02/2026
In the U.S., new glasses take a few days. In Japan, they took 45 minutes and $70. Transit, check-in, doors, customs, toilets all just felt nicer. I wrote about it. isaacflath.com/writing/japa...
isaacflath.com
The Future is Already Here (and it's in Japan) | Isaac Flath
Read "The Future is Already Here (and it's in Japan)" - Technical writing by Isaac Flath on software development, AI, and machine learning.
000
Isaac Flath @isaac-flath.bsky.social · 26/01/2026
I spent months building MonsterUI. Now, when people ask if they should use it, I hesitate. MonsterUI redefined what I love to build, yet I couldn't explain why I stopped using it. I felt uncomfortable admitting that I failed Here's what happened and what I learned isaacflath.com/writing/lear...
isaacflath.com
Learning from Failure: MonsterUI | Isaac Flath
Read "Learning from Failure: MonsterUI" - Technical writing by Isaac Flath on software development, AI, and machine learning.
000
Isaac Flath @isaac-flath.bsky.social · 23/01/2026
Yes, though this is a saying because the last mile/last 10% taking lots of time is not new with vibe coding. So getting that first 90% in an hour is a massive win!
000
Isaac Flath @isaac-flath.bsky.social · 23/01/2026
AI generates code faster than humans can read. When the machine outpaces the reviewer, the team loses understanding. We need to keep humans in control. Jake Levirne of SpecStory, shared how they adapt the review process to the task's risk for this elite-ai-assisted-coding.dev/p/legible-ai...
000
Isaac Flath @isaac-flath.bsky.social · 14/01/2026
I used an AI agent to build a Discord bot. I wanted it to save images from a channel to S3. The agent wrote the code, explained deployment, and debugged it when it went silent. It's a small tool I use daily. Link: isaacflath.com/writing/disc...
000
Isaac Flath @isaac-flath.bsky.social · 30/12/2025
I often ask: Why this way? What were the trade-offs? Was X considered? Why not Y? Ex: "Why is the react editor in the main python repo and not its own module?" AI logs and other context is a key part of answering that. Here's what I do 👇 elite-ai-assisted-coding.dev/p/how-i-use...
elite-ai-assisted-coding.dev
How I Use My AI Session History
Step-by-step example of the tool I use, how I use the UI, and how I use it agentically.
010
Isaac Flath @isaac-flath.bsky.social · 22/12/2025
Link to the course is here: bit.ly/ai-coding-c...
bit.ly
Elite AI Assisted Coding by Eleanor Berger and Isaac Flath on Maven
Make it code like you do. Turn generic AI assistants into coding partners that actually get your style and have the right context.
000
Isaac Flath @isaac-flath.bsky.social · 22/12/2025
I quit my job ~6 months ago to focus on learning and having a bigger impact. The AI coding course with Eleanor Berger is one (of several) projects that is better than I imagined on both fronts Cohort 1 was a huge success, and Cohort 2 in January is gonna be even better 😄
110
Isaac Flath @isaac-flath.bsky.social · 12/12/2025
My favorite thing about reducing token usage for coding agents with better search with @mixedbreadai
000
Isaac Flath @isaac-flath.bsky.social · 11/12/2025
Multi-vector search means semantic search is back for coding agents...and it was always clear it would
000
Isaac Flath @isaac-flath.bsky.social · 11/12/2025
60% of tokens are spent searching and exploring the codebase. Agents need better search.
000
Isaac Flath @isaac-flath.bsky.social · 10/12/2025
mgrep helps explore complex documents and pdfs
000
Isaac Flath @isaac-flath.bsky.social · 10/12/2025
For a detailed write-up, and the full recording of this talk, go here! elite-ai-assisted-coding.dev/p/mgrep-wit...
elite-ai-assisted-coding.dev
mgrep with Founding Engineer Rui Huang
The Problem with grep for AI Agents
000
Isaac Flath @isaac-flath.bsky.social · 10/12/2025
Unlike traditional RAG that vectorizes large chunks of text, @mixedbreadai 's engine represents every single word as its own vector. This provides much more granular and accurate results. Join @aaxsh18 tomorrow for research details on how it works: maven.com/p/0c0eed/mo...
maven.com
Modern Multi-Vector Code Search
Mixedbread showed in their launch how much faster/better semantic search could make Claude code. Cursor also just announced semantic embedding support, and other agents are soon to follow. I got better quality with 1/2 tokens and almost 2x faster with mixedbread search. Learn what changed from the leading researchers in the space.
100
Isaac Flath @isaac-flath.bsky.social · 10/12/2025
mgrep is also multimodal. It can natively index and search images, diagrams, and PDFs in your repository. An agent can find relevant information in visual assets that are completely invisible to text-only tools Very useful for legal, e-commerce and many other domains And cats
100
Isaac Flath @isaac-flath.bsky.social · 10/12/2025
The results from their internal tests with Claude are significant. Using mgrep led to: 🤌 53% fewer tokens used 🚀 48% faster response 💯 3.2x better quality By getting the right context immediately, agents stay on track. I saw similar results. elite-ai-assisted-coding.dev/p/boosting-...
elite-ai-assisted-coding.dev
Boosting Claude: Faster, Clearer Code Analysis with MGrep
I ran an experiment to see how a powerful search tool could improve an LLM’s ability to understand a codebase.
100
Isaac Flath @isaac-flath.bsky.social · 10/12/2025
Agents use mgrep for broad, semantic exploration and grep for precise symbol lookups Instead of grep commands guessing at keywords, an agent makes a semantic query mgrep "how is auth implemented?" It then uses grep for precise function/class name searches. No guessing 😁
100
Isaac Flath @isaac-flath.bsky.social · 10/12/2025
mgrep is a command-line tool that brings semantic search to your codebase, letting agents search by intent, not just keywords. It's much faster than grep alone, and works much better than traditional semantic search.
100
Isaac Flath @isaac-flath.bsky.social · 10/12/2025
AI coding agents burn tokens guessing keywords for grep and flood the context window with noise There's a better way. I hosted a talk by @ruithebaker, a founding engineer at @mixedbreadai, about their solution. mgrep 🧵 elite-ai-assisted-coding.dev/p/mgrep-wit...
elite-ai-assisted-coding.dev
mgrep with Founding Engineer Rui Huang
The Problem with grep for AI Agents
110
Isaac Flath @isaac-flath.bsky.social · 01/12/2025
For all the code, charts, and a deeper dive into the mechanics, check out the full blog post isaacflath.com/blog/2025-1...
isaacflath.com
Quantization Fundamentals for Multi-Vector Retrieval - Blog
An thorough and complete introduction to Quantization for Multi-Vector Search Architectures
000
Isaac Flath @isaac-flath.bsky.social · 01/12/2025
This two-stage compression (PQ + residual quant) means you get token-level understanding in fraction of the space The Mixedbread team has 2 free eng and research talks soon for more info on research and how to use Talk 1: maven.com/p/9c51af/th... Talk 2: maven.com/p/0c0eed/mo...
maven.com
Modern Multi-Vector Code Search
Mixedbread showed in their launch how much faster/better semantic search could make Claude code. Cursor also just announced semantic embedding support, and other agents are soon to follow. I got better quality with 1/2 tokens and almost 2x faster with mixedbread search. Learn what changed from the leading researchers in the space.
100
Isaac Flath @isaac-flath.bsky.social · 01/12/2025
Step 5: Extreme Quantization ColBERT goes one step further. After PQ, it calculates the "residual" error (the small difference between the original and the approximation). Then, it quantizes that error, often down to just 1 or 2 bits per value!
100
Isaac Flath @isaac-flath.bsky.social · 01/12/2025
Step 3: Store which cluster/centroid each piece belongs to Step 4: reconstruct by looking up centroids and combining
100
Isaac Flath @isaac-flath.bsky.social · 01/12/2025
Step 2: Cluster each collection of sub-vectors separately to find the centroids
100
Isaac Flath @isaac-flath.bsky.social · 01/12/2025
Step 1: Split each embedding in half (make sub-vectors).
100
Isaac Flath @isaac-flath.bsky.social · 01/12/2025
However, AI embeddings aren't single numbers; they're vectors (long lists of numbers). This is where Product Quantization (PQ), comes in. It's specifically designed to compress these vectors. It "refactors" similar embeddings to reduce duplication by using k-means clustering. Let's break it down.
100
Isaac Flath @isaac-flath.bsky.social · 01/12/2025
In the simplest form, quantization is a bit like rounding numbers. You give up precision to save space. With scalar quantization, instead of storing a full 64-bit number, you can store an 8-bit code representing its approximate value (8x compression ratio).
100
Isaac Flath @isaac-flath.bsky.social · 01/12/2025
mgrep (by MixedBread) is how I've been using this. It works with any coding agent as a CLI tool. More embeddings (token vs chunk) means more info, which makes sense. But how can that scale? It comes down to the quantization isaacflath.com/blog/2025-1... Here's the core ideas:
100
Isaac Flath @isaac-flath.bsky.social · 01/12/2025
Q: Why is semantic search for code coming back? (Cursor, mgrep, etc) A: Multi-vector architecture Q: Why do I care? MUCH less token usage + better responses if used Q: What makes it work? A: Token (not chunk) level embeddings with extreme quantization Here's what I learned 🧵
100
Isaac Flath @isaac-flath.bsky.social · 26/11/2025
Here's details on some of my experiments on my own commercial projects that inspired the collaboration elite-ai-assisted-coding.dev/p/boosting-...
elite-ai-assisted-coding.dev
Boosting Claude: Faster, Clearer Code Analysis with MGrep
I ran an experiment to see how a powerful search tool could improve an LLM’s ability to understand a codebase.
000
Isaac Flath @isaac-flath.bsky.social · 26/11/2025
This is the second part of a series on semantic vector search, so join that one too if you want all the details maven.com/p/9c51af/th...
maven.com
The 1/2 Token Codebase Search.
RAG famously (and incorrectly) written off as dead by many when agentic search became more popular for agentic coding. Recent advances in semantic search have dramatically outperformed agentic search. This engineering talk gives you the intuition and information to be an expert user.
100
Isaac Flath @isaac-flath.bsky.social · 26/11/2025
I got better quality, half the tokens, and twice the speed with MixedBread's mgrep The "trick" is a multi-vector architecture. Why now? Because the arch and quantization details are hard Join Aamir Shakir (MixedBread Founder) for a research talk on how it works 👇 maven.com/p/0c0eed/mo...
maven.com
Modern Multi-Vector Code Search
Mixedbread showed in their launch how much faster/better semantic search could make Claude code. Cursor also just announced semantic embedding support, and other agents are soon to follow. I got better quality with 1/2 tokens and almost 2x faster with mixedbread search. Learn what changed from the leading researchers in the space.
100
Isaac Flath @isaac-flath.bsky.social · 25/11/2025
RAG isn't dead, it evolved I compared the new semantic search (RAG) against default agentic search (also RAG 😄) for coding agents. Semantic search won Rui Huang ( MixedBread ) will give an engineering talk on why, intuition, and setup Sign up 👇 maven.com/p/9c51af/th...
maven.com
The 1/2 Token Codebase Search.
RAG famously (and incorrectly) written off as dead by many when agentic search became more popular for agentic coding. Recent advances in semantic search have dramatically outperformed agentic search. This engineering talk gives you the intuition and information to be an expert user.
010
Isaac Flath @isaac-flath.bsky.social · 16/11/2025
I went from "this could be better" to a proof-of-concept, prototyping 5 approaches in an hour with AI. It's a prototype that'd need dev work to finish, but I have a clear idea what LoE and end result would be so I can decide whether it's worth it. elite-ai-assisted-coding.dev/p/prototypi...
elite-ai-assisted-coding.dev
Prototyping with AI
Air + Airpine JS prototyping with an AI Agent
000
Isaac Flath @isaac-flath.bsky.social · 16/11/2025
The result was a nice improvement in DX, though maybe not as nice as jinja tooling that exists. AI acted as a more interactive and dynamic whiteboard during this process. For a simple counter, the code ended up like this:
100
Isaac Flath @isaac-flath.bsky.social · 16/11/2025
I settled on an "ORM-like" API as the target. The goal was to write clean, chainable, and auto-completable Python code that felt more natural. This felt most promising so I kept going with it
100
Isaac Flath @isaac-flath.bsky.social · 16/11/2025
I started by brainstorming with AI. It research the air codebase and we brainstormed several API designs (thanks AmpCode librarian 🫡) For each idea (static methods, context managers, etc), I had it generate a minimal, runnable app to get a feel for the ergonomics.
110
Isaac Flath @isaac-flath.bsky.social · 16/11/2025
I figured I'd experiment/prototype a better way with AI. Here's the~ 60-minute journey from that vague idea to a functional (ish) prototype. AI was helpful as a thinking tool to make *prototypes* and explore quickly. elite-ai-assisted-coding.dev/p/prototypi...
elite-ai-assisted-coding.dev
Prototyping with AI
Air + Airpine JS prototyping with an AI Agent
100
Isaac Flath @isaac-flath.bsky.social · 16/11/2025
Using Alpine.js in Python frameworks like Air, Fasthtml, or Dash often means you end up writing code in strings Writing code inside a string is a terrible dev experience and hard to maintain. But I also kinda hate jinja TBH
100
Isaac Flath @isaac-flath.bsky.social · 08/11/2025
This week I was fixing some basic and classic async problems: a "processing" pill that never updated to "complete." Blech. Read the details from my new newsletter 👇 isaacflath.com/newsletter/2
isaacflath.com
Async Background Tasks - Newsletter
Hey this is Isaac, This week I was fixing some basic and classic async problems: a
000
Isaac Flath @isaac-flath.bsky.social · 07/11/2025
Faced a classic async UI bug: a "processing" pill didn't update I used AI to brainstorm to get 4 patterns (I chose self-polling pill) AI is good at giving common patterns fast so I can spend more time on decisions and considering less obvious options elite-ai-assisted-coding.dev/p/using-an-...
elite-ai-assisted-coding.dev
Using an AI Assistant to Brainstorm
I ran into a classic background task UI problem in my app this week.
000
Isaac Flath @isaac-flath.bsky.social · 31/10/2025
AI demos often get stuck in "Proof of Concept Purgatory" by jumping to agents too early. Hugo shares a framework for building AI products, by progressing through the "agentic spectrum" iteratively , starting with augmented LLMs and workflows elite-ai-assisted-coding.dev/p/stop-buil...
elite-ai-assisted-coding.dev
Stop Building AI Agents... And Start Building Smarter AI Workflows
I was excited to host Hugo in our course for a talk.
000
Isaac Flath @isaac-flath.bsky.social · 25/10/2025
AI Assisted Coding Cohort 1 just ended and feedback is coming in. Really happy to hear V1 was beneficial to so many people ❤️ Feedback also means huge improvements coming for cohort 2 😁 I still have quite a lot to add, but I'm adding reviews here! isaacflath.com/course/elit...
isaacflath.com
Elite AI Assisted Coding - Transform Your Development Workflow
Learn to build better software and learn more with LLMs and AI agents.
000
Isaac Flath @isaac-flath.bsky.social · 18/10/2025
Learn from Open Hands, a powerful open-source async agent platform, about how to build your own agents and how their agent SDK helps. maven.com/p/0fe39a/op...
maven.com
OpenHands: AI software development agent platform
OpenHands (and All Hands, the company behind it) set to push the envelope with an agentic software development platform, all while remaining fully committed to their open-source origins.
000
Isaac Flath @isaac-flath.bsky.social · 14/10/2025
Curriculum details and discount link to the RAG course 👇 bit.ly/isaac-rag
boot.dev
Learn Retrieval Augmented Generation [Full Course] | Boot.dev
Learn all about modern search engine techniques like keyword, vector, semantic and LLM-enhanced search. In this course you'll implement different search techniques from scratch in Python everything from simple keyword search up to a fully functional Retrieval Augmented Generation (RAG) pipeline using the Gemini API.
000
Isaac Flath @isaac-flath.bsky.social · 14/10/2025
RAG is how you use search to gives users what they need. My new RAG course just launched, I hope to see you there!
100
Isaac Flath @isaac-flath.bsky.social · 12/10/2025
Or if you just want to jump in and start learning, go here: bit.ly/isaac-rag
boot.dev
Learn Retrieval Augmented Generation [Full Course] | Boot.dev
Learn all about modern search engine techniques like keyword, vector, semantic and LLM-enhanced search. In this course you'll implement different search techniques from scratch in Python everything from simple keyword search up to a fully functional Retrieval Augmented Generation (RAG) pipeline using the Gemini API.
000
Isaac Flath @isaac-flath.bsky.social · 12/10/2025
I love hearing feedback like this about the new RAG course ❤️ To learn more about the course go here: elite-ai-assisted-coding.dev/p/learn-rag...
100
Isaac Flath @isaac-flath.bsky.social · 09/10/2025
80% of users won’t look past the top 5 results. If those results are bad, they leave. I just launched a RAG course, here’s a post on what you'll learn and why the concepts matter in modern AI systems elite-ai-assisted-coding.dev/p/learn-rag...
elite-ai-assisted-coding.dev
Learn RAG: Build Search That Works
Search isn’t magic. It’s engineering. 80% of users won’t look past the top 5 results. If those results are bad, they leave.
000