I started building things for myself with Claude Code many months ago. I’ve built:

  • an RSVP system where my guests converse with an effusive “event planner” who trades letters with the guests to get their RSVP
  • a gallery for our digital art
  • a personal time manager that includes a daily briefing, goal tracker, writing helper, and workout coach
  • the first music project(s) I’ve worked on since I left Apple Music more than ten years ago

But over this holiday break I, like many others, took time to try to find the limits of what I could accomplish with Claude Code, and I couldn’t find the ceiling. I know you’re hearing heads spinning about this; the hype is deserved. I can’t think of an analog for what I’m feeling. The closest for me is the feeling of building Web sites and registering domains in the 1990s. It’s the feeling of a new kind of agency.

While Claude Code is busy doing what it does, I’ve been jotting down thoughts. Here are a few:

I’m glad I started playing with Claude Code before Sonnet 4.5, because feeling the jump from 4.0 to 4.5 at the end of September was a lesson in how fast things are improving. Claude Code periodically asks you how it’s doing and my answer went from always a 2 to always a 3 out of 4. I went from cursing at Claude Code for doing dumb shit to muttering “no. fucking. way.” to myself during every session. 

In addition to Claude Code I’ve been playing with Codex, which is no less impressive. It has a different personality and I’m not sure which one I prefer. Also, I’m using Vercel and Supabase for pretty much everything. It’s not just Claude Code, it’s the entire stack of Dev tools available to everyone today that allows you to ship working apps so quickly. 

I have not built any native mobile apps but I have turned my apps into mobile-optimized PWAs.

I find it tricky to get what I want from a graphic design perspective, but I’m getting better at it. I had Claude redesign my old blog and it’s pretty OK. I’ve avoided doing this for years simply because I haven’t wanted to waste the time. But when it becomes a fifteen minute task it suddenly becomes worth doing. The Claude Chrome plug-in was very handy for this (and many other things). 

I’m sure you’re thinking I spent a lot of time doing this over the holiday. I did not. What’s astounding is how much I accomplished in such little time actually spent at the keyboard. I never did any “flow state” sessions at the terminal. It’s more, “take a walk in the woods for an hour, let the ideas spin, then type the prompts in when you get home and approve/direct while you make a snack and read.” I can’t imagine what it’s possible to accomplish if you have fifteen sessions running in parallel like the people I’m reading about who are focused on building this way. A next step for me is getting a VPS set up with Happy connected so I can Claude Code while hiking. 

I don’t think AI kills coding as a skill. It changes it, yes. Every “coder” is now a manager of an army of coders, a pipeline optimizer. It certainly makes the roles of Product Manager and Engineer overlap much more. A product manager used to write specs, hand off to engineering, then wait for the code to arrive and see if it meets the spec. Now they will build what they envision and hand it off to engineering for integration, security review, hardening, and push to production. Similarly, engineers previously defferred to product managers for user research, product marketing and brand promise/fit, etc. AI is now a better partner for both product managers and engineers, and both can do more without a human hand-off. All that without saying anything about other functions such as QA, customer care, etc. 

It seems to me SAS will be transformed by this very quickly. What non-tech Enterprises (like LVMH, where I worked, or Dr Marten’s, where I’m a non-exec director) do with SAS has never made sense. They overpay for complicated systems such as ERP then overpay more for “integrators” (rarely staffed with 10x engineers) to make the system work within their specific environment. But now it can be cost-effective to have employees who know their systems and processes well build and continuously improve them. Given what Enterprises spend on SAS I can imagine a Claude Coder embedded in teams (Finance, HR, e-commerce), constantly building the features these teams need and ask for. Eradicating repetitive work flows will be a job. 

I do think Claude Code and Codex completely transform venture capital and startups. You needn’t have funding to build a prototype and show demand. You can now get past the prototype step in A DAY without a technical co-founder. I’m thinking of the failed startups in my past, both ones I tried to build as well as investments made, and realizing we now have the tools to rebuild these relatively small and simple single-purpose apps comparatively instantaneously, and the ability to completely rewrite the code ground up on a daily basis if necessary to pivot toward a customer. I can think of multiple companies where we built product with small teams, and Claude Code and Codex to build a better app (including documentation, APIs, unit test coverage, etc) than we were able to build over a couple of years in a day. These startup apps were usually solid ideas but not great apps. They were built with the skills we could afford, not necessarily the best teams, the tech debt was instantaneous and we could never afford to dig out of the holes we found ourselves in. The approach to all these hurdles would be completely different with these tools. The type of startups we see this year will be a new animal, both for better and for worse. 

Computer Science as a discipline isn’t dead, but ten years from now the kids who are starting to code today will say, “Can you believe my dad used to write and debug this shit by typing each character, parsing code with his eyes, and stepping through the debugger himself?!?! That’s insane!”

There have been a lot of “AGI is here” takes with so many people discovering they have superpowers with Claude Code. Here are a few worth reading IMHO:

https://www.oneusefulthing.org/p/claude-code-and-what-comes-next

https://www.transformernews.ai/p/claude-code-is-about-so-much-more

https://www.robinsloan.com/winter-garden/agi-is-here/

https://www.patreon.com/posts/does-work-still-147978167 (podcast) 

https://www.patreon.com/posts/why-everyone-is-147688638 (podcast) 

I’ve been devouring all the content about how to get the most out of Claude Code. I found these valuable to improve my setup and approach:

https://steipete.me/posts/2025/shipping-at-inference-speed

Also, everyone, regardless of your technical level or job function, should be doing this ten week “jump-start your AI skills in 2026” 10-week new year resolution challenge:

https://www.patreon.com/posts/ai-new-years-10-147076337

The mainstream media take is pretty funny and misses the point per usual: 

https://www.perplexity.ai/page/claude-code-sparks-developer-i-l.fodMTgRhS_awrXM1nnXQ

Please suggest other worthwhile reading I may have missed!

Is there an AI bubble? Of course there is. As Carlota Perez pointed out, all technological revolutions come with a bubble, a burst, then thirty years of sustained growth. Do you really think this could possibly be the first technological revolution that doesn’t? No way. It’s likely the biggest revolution in the history of humanity. So isn’t it more likely it’s the biggest bubble in the history of humanity? Does that take away from it’s revolution-ness, or define it? 

Building on the internet is so much more fun than consuming. 

My next PC:

https://x.com/i/status/2009973702314864680