In This Newsletter

Why You Hate AI Coding (And What To Do About It)

Some developers absolutely love AI coding. The annoying part of their jobs is gone and they can just build. I am very happy for them!

But there are a good number of us (myself included) who miss the logic puzzles and the flow state. For us, working with AI coding agents makes us feel this weird combination of bored and overwhelmed at the same time.

AI coding is here to stay, so for those of us who don’t love it, what can we do? This week’s content is about that.

A few things I cover:

  • AI brain fry is legit. In a BCG survey published in Harvard Business Review, 14% of people using AI at work reported mental fatigue from overseeing AI tools, and productivity dropped off once people were juggling four or more (like a bunch of terminals at the same time).

  • Your willingness to put in effort drops before your ability does. A Johns Hopkins study found that tired people stopped wanting to do the hard task long before their performance actually declines. So brain fry shows up initially as "ugh, I really don't want to read this huge PR…"

  • There is something we can do to help with this feeling. My coaching client Lewis Bates built a simple system to help make the process more enjoyable. Check out the video for a breakdown.

Or read the companion guide here.

Last Call for the Launchpad

The AI/ML Career Launchpad is my group coaching community for people breaking into AI and ML, with coaching and accountability from me, skill roadmaps and templates, and a group of people going through the same transition. It’s a fun place, you should join us.

Enrollment closes tonight, Thursday, November 19, at 11:59 pm PT.

What I Learned at the AI Engineer Code Summit

Last week I was at the AI Engineer Code Summit in San Francisco.

And yes, I'm aware it's a little funny that I made a whole video about not liking AI coding, and now I'm sharing all the updates on how AI coding works. But the reality is it's here to stay, and my video is more about how those of us who don't do well with AI coding can flourish in this new world.

And so many people at the summit were so excited about the new things they could build! They're having the time of their lives. Just because I don't like it doesn't mean it's necessarily bad.

Here are the key takeaways:

  • Verification is the main engineering problem now. Once generating code got cheap, almost every talk had its own answer to "how do you know the work is right?" ZoomInfo showed an LLM judge accepting an agent's claim that it had scheduled an email when the tool results said otherwise, so this is important stuff to take seriously.

  • Everyone loves Jev. At least six talks used fast, structured models for routine choices like ranking, routing, and judging, without paying full LLM cost and latency. If you haven’t worked with Jev yet, now’s the time.

  • Slop control is moving into code. Lauren Tan argued that every correction you give an agent should become a deterministic check, like a lint rule, so nobody has to catch the same mistake again.

  • Harnesses are shrinking. Philipp Schmid (Google DeepMind) showed a bash-only harness costs about half as much per task as one with many tools, and older tricks matter less as models improve.

  • Humans stay in the loop. Speakers disagreed a lot on process, but they agreed humans stay at the decision points. Steve Yegge, who has 22 Claude Max subscriptions running at once (!!!), said more agents so far has meant more humans for him and his team.

  • Ambition is all the rage. Several keynotes pushed people to build bigger things, and each paired that with a warning about passive use of agents and unreviewed slop.

Here are the full livestreams if you want to check it out for free:

Forwarded this email? Sign up here.

Follow me on YouTube | Instagram | X | LinkedIn | TikTok

Note: This email may contain affiliate links. If you make a purchase I may make a small commission, at no cost to you. Thank you for your support!