Collected Thoughts

A few weeks back, I wrote a post arguing that AGI should not be a goal to achieve with LLMs. I wrote from the heart and at my own level as an experienced layman, and I didn’t hold back from expressing deeply-held beliefs that aren’t grounded in research. I’m satisfied with my writing—it has flaws, but I think it holds up as an impassioned opinion piece.

The incredible pace of change in AI turns today’s truths into tomorrow’s fictions (and vice versa), which creates a contradictory tension between two important principles. First, it’s more important than ever to maintain my grounding in closely held values. Second, I need to maintain an open mind and be ready to radically alter core beliefs. In that spirit, I’ve been reflecting on my post and wondering if I could build a more structured case for arguing against scaling LLMs into AGI. I’ve also been curious how much of my argument is easily refuted or biased.

I did a short Claude-assisted exploration that helped me discover some nuances and dig deeper into the issue—it’s a powerful demonstration of the current capabilities of frontier AI. I used Claude’s research mode to create a reasonably detailed document with citations.

After a fascinating experience working with Claude Code, I’m ready to share a prototype app for learning to code in an LLM-native way. I documented the beginning of this project, in which I used OpenAI’s custom GPTs, and the start of work on the prototype. This work is the result of hitting some of their natural limits without needing to resort to custom actions, along with my desire to have an all-in-one experience instead of a copy/paste workflow.

You can check it out here.

This is a proof-of-concept web app based on my musings about what it means to learn to code in 2025: an age when novices can accomplish quite a lot without knowing anything about programming. I firmly believe that learning programming still has value. When a novice hits walls with more complex projects, they’ll only be able to advance if they can reason about code written by an LLM, even if just at a high level. And if they can’t read or modify any code by hand, they’ll be more likely to remain stuck. I’ll write more about these thoughts later.

I just had a near miss in my job hunt. When you come so close without getting the nod, it can really hurt. This isn’t the first time, and that’s both bad news and good.

Having left my last job over a year ago and been in the job search for about nine months, I’ve grown accustomed to quite a few new rhythms. I am responsible for structuring my days and choosing how to fill my time. I create my own priorities and my own pace. I have to hold myself accountable and find people who can help.

There’s also a pattern in the job application process. The mechanical process of customizing résumés, writing cover letters, and filling in forms. The personal connections of networking with known contacts, interviewing, and emailing. The emotional journey from detached analysis to excitement as I progress through the initial rounds towards a final selection.

Applying to jobs is a game of percentages, so at this point I have a lot of fatigue from the many initial applications I’ve submitted, while the middle of the process feels well-worn without having lost all of its novelty. And unfortunately, while rejections at the last step of the process are pretty rare, I now have three under my belt. Enough for me to experience a twinge of familiarity.

I believe that we’re living through a technological revolution that rivals the dawn of the information age. I also believe it’s partially driven by hype, misinformation, hubris, and poisonous incentives. At its foundation is a technology that’s enormously powerful, generally misunderstood, deeply flawed, and (of course) controlled by far too few, with far too much unchecked power.

And that’s too bad. Modern machine learning techniques are enabling amazing breakthroughs across many domains. The revolution is real, and we’ll reap the benefits. But even leaving aside any existential risks—where I’m truly ambivalent—we’re risking enormous harms in the future, and we’re doing serious damage even now.

My core disagreement with the current narrative around AI is its focus on achieving artificial general intelligence or superintelligence (AGI/ASI). First, I dispute that these goals are even possible using today’s technology as I understand it. Second, I question whether AGI and ASI are worthy of pursuit: I contend that if we could achieve them using today’s technology, we would create something monstrous.

I’m listening to Rachmaninov’s second piano concerto, getting the goosebumps and euphoria that always come at the end of the second movement. No matter how far we push AI, this experience is quintessentially human, and something AI will never possess—at least as far as I can see, given the technological underpinnings and stated aspirations of AI researchers.

I made some serious strides with my “vibe-to-learn-to-code” project. In half a day. I did this with Claude Code.

I watched the video; I read quite a few things about it; but they didn’t entirely prepare me for the experience. As always, AI progress moves so quickly that it’s hard not to be astonished when you experience a step change in capability.

Prior to this, I’d used Aider for some coding automation, as well as interacting directly with both ChatGPT and Claude via their standard chat interfaces. Those experiences were powerful enough; I built this website from scratch using Claude with artifacts.

Agentic coding is not entirely new, but it’s really become mainstream within the past year, if not the past six months. Here are some of my initial thoughts.

The European Commission and the OECD just released an AI Literacy Framework for teachers, leaders, policymakers, and learning designers. Here’s their mission statement:

Artificial Intelligence (AI) has become a widely used tool in our everyday life, including for learning, personalized assistance, and entertainment. Therefore, young people must be able to understand how AI works, its societal impact, and how to use it ethically in order to be prepared for a society and economy in the age of AI.

I sympathize with Miriam Reynoldson’s skepticism of the term “AI literacy” and the place of AI in education writ large. Terminology aside, she and I agree that there is a real need for students to have critical literacy and build skills for good judgement, and nurturing those competencies requires us to understand and incorporate AI. In some ways, it’s no different than the impact of past technologies—the World Wide Web and social media, to name two.

I encourage you to read the whole framework. It synthesizes a lot of important concepts and reflects a lot of careful thought. If you have your own thoughts, you can submit feedback. I did, so I thought I’d share my ideas here.

I’ve spent many years teaching programming, and I’ve always tried to adapt my approach to fit the current technology and anticipate my learners’ needs. AI is having a huge impact on programming: Novices can now simply talk to an LLM and get it to output working code. On the whole, this is a good thing. Programming can be fun and rewarding, and vibe coding opens up possibilities for a new cohort of hobbyists and tinkerers who, otherwise, might have considered it beyond their capabilities.

But where does that leave the field of coding education, especially at the novice level? I believe that we have to adapt our approach to fit with the current tools and with the habits and aspirations of learners. But we also need to give them a scaffold that they can use to build a solid foundation, if they wish to. A base from which to explore programming and computer science in a more structured way.

I set myself a goal of posting to this blog once a week. For the most part, it’s worked out well. I keep a journal of post ideas, and I typically select one on Monday, start writing, and try to post it at some point on Wednesday.

One rationale for keeping this schedule is that the discipline is good for me—I’m in the middle of a job search and career pivot, and the landscape is uncertain, especially in my chosen fields of technology, coding, and education. Having a regular activity helps keep me grounded.

This week, the process broke down.

Apple seems to be having a moment—and not in a good way. The judge in the Epic v. Apple suit just excoriated top leadership for violating the ruling that required them to provide developers with a system to link users out to external payment processing. I won’t get into details here, but you should know that just about everyone outside Apple leadership viewed the order as completely reasonable. Yet Apple’s "compliance" was reverse-engineered to preserve their bottom line by imposing hefty fees on the external payments.

Then there’s the Apple Intelligence fiasco. They announced it with much fanfare last June at WWDC, and it looked like a reasonable (if ambitious) approach to integrating LLM technology into their platforms. And if they had done it right, it really would have been a game-changer for their platform and the industry. But they bungled it by announcing vaporware. They’ve released basically nothing of consequence over the past year, despite having used the moniker aggressively to market the latest generation of devices and operating systems.

From a consumer perspective, they’re doing OK, but developer sentiment is grim, and their diehard insider community has been distressed about their direction for at least a decade. I count myself among them. I’ve been an Apple user since the beginning, and much of what I see in Apple now pains me. But while the company seems to have turned to evil, more profit-extracting machine now than human-centered enterprise, there’s still good in it. They still make (some) amazing products. And, on balance, Apple employees are talented, creative, kindhearted, believe deeply in their work, and want to do good things in the world.

I saw a post on LinkedIn that got me thinking about the world I'm parenting my six-year-old in. It's equal parts scary and exciting. Things are changing way too fast for me to truly keep up, and it will only accelerate from here. Yet, fundamentally, my job as a father isn't all that different than it was for my father—it's just inflected with today's technology.

I'm the same age as the personal computer. It puts me in a unique position to understand today's technology, since as I grew up, personal computing grew along with me. I'm part of the first generation of humans who truly had the opportunity to be digital natives. I've witnessed every era of the personal computing revolution, and I've adapted to each one.

On one hand, AI is the latest in a chain of technological quantum leaps—from desktops, to laptops, to the public Internet and nascent World Wide Web, to the modern web, to smartphones. And yet AI is also different, a step change that defies a simple depiction as such. So what changes about parenting, and what remains constant?

Looking for a job is always challenging. Sometimes the struggle is rewarding and fun, and other times it's easy to succumb to darker thoughts. Here's an illustration in a single day.

The valley

I had a great job application process last week with a small startup for an iOS developer role. I talked to all three founders, and had deep, stimulating conversations with each one. We clearly mutually respected and liked each other, and I was feeling positive about my chances.

I came across Playlab as part of my exploration of educational technology as I build the next chapter of my career. They give teachers tools to build custom chatbot apps on top of all the main LLM platforms, and a community in which to share and remix them. In the universe of AI tech for education, I think their approach is unusual and thoughtful.

The interface provides a prompt construction area with guidance and templating, settings for persistent memory and documents for the LLM context window, and some other helpful tweaks to improve the standard experience of interacting with AI chatbots. (Imagine the Projects feature of Claude or ChatGPT, but thoughtfully designed specifically for education.) The community is pretty broad and has lots of interesting applications, from teaching tools like lesson planning, to student-focused ones like writing feedback.

I was in San Diego over last weekend for The AI Show, which preceded the ASU+GSV Summit. I'm considering career options in education technology, and I had a couple objectives for the weekend.

Get the lay of the land. I have a deep background in education, technology, and instructional design, always orbiting EdTech without ever being truly inside. How has it changed since I was a teacher myself? What influence is AI having?

Connect with people. To find the right opportunities, I need a broad network of people who are not in the industry, but operating at the forefront.

I love many of the projects at The Pudding, and they’ve inspired me to try telling a story with data, not just make some charts and graphs. There are more than a few challenges for me to tackle, but the principal one is finding the right data set. Without my emotional investment, any storytelling project will fall flat.

Since November’s election, and especially since January, I’ve struggled to find ways to engage and act in the face of the onslaught. I recognize how privileged my existence is, and even for me, everything is overwhelming on a daily basis. I’ve read many times that it’s important to stay active and do something, no matter how small, that helps you to feel that you’re part of a solution.

My love-hate relationship with searching for a job goes back to my first. On one hand, I'm excited by the possibility that lies in the unknown. But on the flip side, I struggle at times with focusing my research exactly because there are so many possibilities.

This time around, LLMs are in the mix. I'm trying to take advantage of them to tap into my excitement, while alleviating my struggles.

Simon Willison has a great post that—coincidentally—he published just a day after my own post on coding with LLMs. You should definitely check it out.

When I was working at Apple, they had a strict no-LLM policy, so I wasn't able to use them to accelerate my work. At the time, AI-assisted coding was considerably less powerful than it is now, but I'm sure even then I could have benefited from automating repetitive and simple coding and writing tasks.

Now I'm between jobs and exploring career options. Obviously AI is inevitably part of the process, in multiple ways. I'll write more about some other experiments and workflows in later posts, but here I'll get into my first impressions of AI-assisted coding.