Loading...
Loading...
Spotify
Back to blog listing

Essential Resources for AI Engineers in 2025

The “AI Engineer”

As I move towards the end of 2025, I’ve been reflecting on where I want to go next in my career. For 5 years I’ve been pretty hyper focused on developer experience, first at Meta and now at Cisco. Whatever I do next will hopefully not only continue that trend, but intensify it.

💡 Career Trajectory: I’ve been deeply interested in devx and productivity since the beginning of my career, and I am always rethinking different approaches.

In the age of AI-powered developer tooling and software, people have been using this term “AI Engineer” to refer to engineers that are building these systems. Since this aligns exactly with the trajectory I’m on, I’ve been diving pretty deeply into the space.

Really I just want to drop a handful of resources that I’ve found useful so far, but then I’d like to discuss some things which are related to this more broadly. I plan to keep this updated as I find more resources, so check back in a few months for more.

Origins of the “AI Engineer” Term

Really quick, who actually started this term “AI Engineer”? I’m honestly not entirely sure, and I’d hate to drop some misinformation, but it seems like it was Swyx (Shawn Want) in 2023, where he discussed the rise of the AI engineer. You can check out his post The Rise of the AI Engineer to get his perspective.

🎧 Podcast Recommendation: He’s also the host of the Latent Space newsletter and podcast, which I highly recommend subscribing to if you’re interested in AI and machine learning.

The Latent Space podcast is a great resource, and not just owing to the world class guests they have on. The hosts really know their stuff, and they are deeply connected to the industry. Other AI-related podcasts tend to waffle despite having great guests, just because the hosts don’t have the same level of expertise.

Why Latent Space works: The episodes are also really digestible, usually being around 60-90 minutes long. This is in juxtaposition to some great episodes of Lex Fridman’s podcast, which can be upwards of 5-6 hours long. Great for really downloading what someone believes in or their broader perspective (assuming you trust them to be transparent and honest), but not ideal for the breadth needed to stay up to date in such a fast moving field.

Latent Space is a must listen, in my opinion. Latent Space strikes a great balance here, at least for my tastes.

Top Resources for AI Engineers in 2025

⚡ Overwhelm Alert: If you want to avoid getting overwhelmed, here’s my TLDR

🚀 Quick Start Guide

  • 📚 Essential Reading: Buy Chip Huyen’s book AI Engineer and read it cover to cover
  • 🎧 Essential Listening: Start listening to the Latent Space podcast and subscribe to their newsletter
  • 📺 Essential Watching: Subscribe to the AI Engineer YouTube channel and watch the videos from past conferences

Core Resources

Latent Space

Essential Books

These are all linked directly from the author’s personal site, so it includes their reference links as well.

AI/ML Publications & Learning Sites

  • Papers with Code - Find the latest research papers and their associated code implementations
  • Arxiv Sanity Preserver - Keep track of the latest papers on arXiv, with a focus on machine learning
  • Distill.pub - Clear explanations of machine learning concepts, often with interactive visualizations
  • The Gradient - Wide range of topics in AI and machine learning, with articles written by experts
  • DeepLearning.AI - Founded by Andrew Ng, offers courses, articles, and resources for learning about AI and deep learning
  • Fast.ai - Free courses and resources for learning about deep learning and AI, with a focus on practical applications
  • Machine Learning Mastery - Tutorials and resources for learning about machine learning, with a focus on practical applications
  • KDnuggets - Wide range of topics in data science, machine learning, and AI

Essential Newsletters

  • AI Weekly - Weekly newsletter that curates the latest news and articles in AI and machine learning
  • Import AI - Newsletter by Jack Clark that covers the latest developments in AI and machine learning
  • The Batch - Weekly newsletter by Andrew Ng that covers the latest developments in AI and machine learning
  • Data Elixir - Weekly newsletter that curates the latest news and articles in data science, machine learning, and AI

Major AI Research Organizations

Leading AI Labs:

  • OpenAI - Research page with latest papers and projects
  • Anthropic - Research page with latest papers and projects
  • DeepMind - Research page with latest papers and projects
  • Meta AI (FAIR) - Research page with latest papers and projects

Big Tech AI Research:

International & Specialized:

AI-First Companies:

  • Hugging Face - Research page with latest papers and projects
  • Cohere - Research page with latest papers and projects
  • Stability AI - Research page with latest papers and projects

Notable Individuals & Their Blogs

AI Engineering & Industry Leaders:

  • Chip Huyen - Author of AI Engineer and Designing Machine Learning Systems
  • Andrej Karpathy - Formerly at Tesla and OpenAI, now working on his own project
  • Swyx - Host of Latent Space and author of The Rise of the AI Engineer
  • Chris Olah - Research scientist and leading figure in AI interpretability and understanding of neural networks

AI Research Pioneers:

  • Geoffrey Hinton - One of the “Godfathers of AI” and a pioneer in deep learning
  • Yoshua Bengio - Another “Godfather of AI” and a pioneer in deep learning
  • Yann LeCun - Chief AI Scientist at Meta and a pioneer in deep learning
  • Ian Goodfellow - Inventor of GANs and author of Deep Learning
  • Fei-Fei Li - Co-director of the Stanford Human-Centered AI Institute and leading researcher in computer vision

AI Education & Practical Applications:

  • Andrew Ng - Co-founder of Coursera and DeepLearning.AI, and a leading figure in AI education
  • Jeremy Howard - Co-founder of Fast.ai and leading figure in AI education
  • Rachel Thomas - Co-founder of Fast.ai and leading advocate for ethical AI
  • Sebastian Thrun - Founder of Udacity and a pioneer in self-driving cars

Technical Authors & Practitioners:

  • Sebastian Raschka - Author of Python Machine Learning and Machine Learning with PyTorch and Scikit-Learn
  • Aurélien Géron - Author of Hands-On Machine Learning with Scikit-Learn, Keras, and TensorFlow
  • François Chollet - Creator of Keras and author of Deep Learning with Python
  • Andriy Burkov - Author of The Hundred-Page Machine Learning Book
  • Luis Serrano - Author of Grokking Deep Learning and Grokking Machine Learning
  • Abhishek Thakur - Kaggle Grandmaster and author of Approaching (Almost) Any Machine Learning Problem

Business & Strategy:

  • Cassie Kozyrkov - Chief Decision Scientist at Google, great resource for understanding the intersection of AI and business
  • Richard Socher - Former Chief Scientist at Salesforce and leading researcher in NLP

Podcasts & Media:

Thought-Provoking Writing:

  • Dan Luu - While he isn’t AI-focused, I find his writing thought-provoking and inspiring
  • The Pragmatic Engineer - Gergely Orosz’s insights on engineering leadership, career growth, and industry trends

Information Overload

🎯 The Core Challenge: When working for a business, one of the biggest challenges is to know what to focus on.

The bigger the company, the more input streams there are for you to pay attention to. It can very quickly become overwhelming just to gather all of the context and synthesize it into a coherent understanding of the business priorities and strategies.

You would think that this should just outright be someone’s job. I often look to my management chain, expecting them to help me really have some solid insight into what matters and why.

What I notice, time and time again throughout my career, is that there’s this huge chasm between the engineers on the ground and the people that are making decisions. Company all hands can be neat, but there tends to be so much fluff and noise that people can often leave more confused than they were before. The more you ask your coworkers for their perspective, the more you realize that everyone has the same problem as well.

The AI Information Firehose

In the current era of AI, I think all software engineers are aware of the hype and buzz around these new tools. Cursor, Windsurf, Claude Code, you name it.

In the world of web engineering, I certainly felt, after a certain point, that there was just too much going on in the industry for me to keep up with the newest thing. People have colloquially referred to this as “JavaScript fatigue,” and I think the same thing is happening in the AI space, but at a much faster pace.

💡 Pattern Recognition: Actually, I think there’s a deep fundamental here that I have noticed is an issue in any long term endeavor. I think you can look to my three examples so far and see a common thread.

Here I’d like to hypothesize something akin to a proportional square law for input streams:

📐 The Input Stream Formula:
Costdecision = (Number of Input Streams)2 / Synthesis Factor

To clarify, an input stream is just some incoming information. An all hands, a company memo, a team meeting, something your coworker said, something your manager needed you to do, Confluence docs, an RFC from that super senior guy, industry news and trends, and so on. Essentially, the more input streams you have, the more difficult it is to make decisions. What ends up happening in organizations is that people just make decisions based on either limited context, gut feel, or following the loudest voice in the org. Often that loudest voice tends to be the person with the most power, not the person with the most insight or wisdom. Meta was the most “organized” company I’ve worked for on this front, and I can tell you it was still complete chaos. Team planning was a massive challenge, and it was so painful to try and synthesize everything into a coherent vision of what actually needed to get done. This often culminated in these planning sessions where we would have to endure an onslaught of brainstorming sessions, discussions, debate, and context switching. And at the end somehow we would still come out with just a massive list of priorities that still needed to be filtered down, prioritized, and distributed among team members. At the height of this, it took us literally an hour just to go through this final spreadsheet of items just reading through them as a group and double checking that it mattered. Imagine having a list of 90 projects to deal with!

AI as a Solution to Information Overload

Ironically, I think AI could actually help with this problem. On one hand, we know that the amount of effort required to make good decisions will increase proportionally with the number of input streams coming in. On the other hand, we know that more context will help us be better informed. Right now we live in a world where people settle for approximations and try to find a sweet spot where things are good enough. In extremely small teams and orgs this problem is much less pronounced internally, but the external context is still there. And this external context is growing at a seemingly endless rate. Compared to everyone that I know personally, I am almost obsessed with trying to stay on top of the latest trends in tech, software engineering,, AI, etc. And yet I still feel like I’m missing out on so much.

Why Traditional RSS Doesn’t Work

🚫 RSS Reality Check: Lately I’ve stopped using RSS. If you just toss all of these resources into RSS, you end up with a firehouse of information that is impossible to keep up with.

Idk who has two hours a day to just sit and comb through a massive RSS output, but it’s just never worked for me. If you want to limit to 2 or 3 sources, you might be better off just keeping up with hacker news or subbing to a few things on substack or whatever.

For my purposes I really want to be up to date with as much as possible, without getting overwhelmed.

My AI-Powered RSS Solution

🤖 The Solution: I’ve built a small wrapper about the RSS process that makes things much more sane for me.

I’d love to share the code, but it’s a hacky mess and it is very particular to my setup and workflow. Luckily, I think the overall principles are simple enough that anyone with purpose can hack together something similar for themselves.

The Foundation: Personal Knowledge Management

📓 My Setup: It all starts with having already formed other habits around productivity. For me, I use logseq to track everything.

If you haven’t heard of logseq, it’s a local first, markdown based, outliner. It’s like Roam Research, but open source and free. I use it for everything. I track my tasks, my notes, my ideas, my projects, everything. I also use Obsidian for some things, but logseq is my main tool.

If you don’t have a system like this, I highly recommend getting one. It’s been useful for me, but it’s also been a bit of a roller coaster, oscillating between being way too meticulous about my planning and being way too lax. The key is to find a balance that works for you.

The Technical Implementation

🔧 The Process

  1. Rules Definition: I have a markdown file called Rules.md with natural language rules about what I want to follow
  2. Feed Discovery: API call to Anthropic Claude to synthesize rules into potential RSS feeds
  3. Feed Validation: Another call to ensure feed URLs actually work
  4. Database Storage: Merge working feeds into local sqlite3 database
  5. Daily Synthesis: Cron jobs pull items and call Claude API for summarization
  6. Output: Directly output into daily logseq journal entry

The schedule: I run this synthesis process once a week on sunday via a Cron Job. Another cron job runs every day at 10 am and 10 pm to pull all the daily RSS items from those feeds, and, most importantly, calls the Anthropic API to have it summarize this list.

Ultimately I get something directly output into my daily logseq journal entry that looks something like this:

Sample Output

  • Daily AI/ML Updates #rss-daily

    • 📋 Daily Digest

      • The tech landscape is seeing significant AI advancements, with a focus on enterprise AI adoption, developer productivity tools, and innovative applications across education and business sectors. OpenAI and other players are pushing the boundaries of real-time AI capabilities and strategic investments in transformative technologies.

    • 🔥 Priority

      • OpenAI Introduces GPT-Realtime A groundbreaking speech-to-speech model with enhanced API features, including image input and SIP phone calling support, signaling a major leap in rea…

    • 📰 Notable

Process Overview

And here’s roughly what a run of the final step looks like. This is what the daily cron job does, more or less:

❯ python3 main.py daily
Initializing RSS Claude Tool...
Running daily pipeline for 2025-08-28...
📥 Updating RSS feeds...
   0 new items from 51 sources
🔄 Deduplicating items for 2025-08-28...
   15 unique items (0 duplicates removed)
🤖 Analyzing 2025-08-28 items with Claude...
    Analyzing: IntelliJ IDEA 2025.1.5 Is Out!...
    Analyzing: Introducing gpt-realtime and Realtime API updates...
    Analyzing: Koog 0.4.0 Is Out: Observable, Predictable, and De...
    Analyzing: ReSharper’s New Out-of-Process Engine Cuts UI Free...
    Analyzing: Supporting nonprofit and community innovation...
    Analyzing: 4 easy ways to personalize your Pixel 10...
    Analyzing: Trump administration’s deal is structured to preve...
    Analyzing: Anthropic users face a new choice – opt out or sha...
    Analyzing: AI or not, Will Smith’s crowd video is fresh cring...
    Analyzing: MathGPT.ai, the ‘cheat-proof’ tutor and teaching a...
    Analyzing: Investors are loving Lovable...
    Analyzing: AI hires or human hustle? Inside the next frontier...
    Analyzing: How a 16-year-old company is easing small business...
    Analyzing: Maisa AI gets $25M to fix enterprise AI’s 95% fail...
    Analyzing: The Trick That Finally Makes Soft Bodies Walk!...
   Analyzed 15 items
📝 Generating daily synopsis...
✅ Daily synopsis saved to synopsis_2025-08-28.md
📓 Updating Logseq journal...
✅ Updated Logseq journal: <path-to-my-logseq-graph>/org/journals/2025_08_28.md
✅ Logseq journal updated for 2025-08-28

Final Thoughts

🎯 Reality Check: Now obviously this approach isn’t perfect. The process over all is very lossy.

What I’ve done is optimize for prompt length along the way so that at each step Claude is given the least amount of context possible. This is because model performance degrades with longer prompts. However there’s still a ton of room for improvement.

The bottom line: Where using RSS directly was feeling super overwhelming, this approach has made it much more manageable so far. Overall, this has helped quite a bit.

Back to blog listing