It's go time

I am an old timer in the software business.   My career has spanned many waves of transformation and innovation — PCs, GUIs, 32-bit programming, object orientation, client-server, the internet, DHTML, mobile, cloud, microservices, etc, etc, etc.  I’ve lived through so many “The Year of the XXXX” and “This changes everything” moments — I am sure I have forgotten trends, fads, and waves.   It would be normal to be pretty jaded at this point.  

But I am not jaded.  The AI wave is the most fascinating yet.  I don’t know of a single colleague who is on the sidelines.  Everyone is diving in.   It is exciting; every single app and market is up for grabs again.  

And it is chaotic. For example, picking models to use is ridiculously complicated – here are pages and pages of tips. But as someone commented — this is fantastic! No one is dominant; the competition level is insane. This is good news for innovation and app developers.

I’m in total agreement with the sentiment that AI is underhyped.  The amount of brains and effort pouring into this wave is unprecedented.  When you pour that much intellect and effort into an area, amazing things are going to result.

Managing agent usage

I use ChatGPT and Claude for a lot now.  Newsletter research.  In-depth book analysis.  Criticism of my writing and thinking.   Prototype design and implementation.  Economic research.  Medical deep dives.  Every day, I use the tools in new ways, and every day, I reconsider how to use them.

Because my usage is a mess.  I am interacting with LLMs across dozens of projects, each at a different stage of completion, with many interacting with my IDE and terminal, all in varying states and with varying contexts and MCP providers — I need an agent just to keep track of all this.

The layer above agents to manage and organize projects is a mess, but also a great opportunity.   ChatGPT offers the concept of "projects," which are insufficient for managing context, files, prototypes, IDE interaction, and other related tasks.  Claude Code does a nice job creating project organization docs and to-do lists, and allows you to walk through the plan in a controlled fashion.  However, it is not designed for non-coding projects — the rest of my family is unlikely to use VS Code or the command line to manage their projects.

Vibe Kanban is a very nice attempt at a Kanban-style interface for Claude projects.  I would like it to be more integrated with my IDE.  And it could be a good interface for non-coders to use — though they could never use Vibe Kanban as is, it is very developer-focused.  A kanban-style tool for non-coding projects would be great.

I looked at Roo Code briefly; it is also trying to provide much more assistance in managing projects.  I didn’t find it materially better than the Claude Code integration in VS Code, tho it does work with a lot of other coding agents.

Claude-flow is another interesting approach, it spins up a handful of very focused agents to work hand-in-hand on your project.  It burns through tokens fast and is maybe not quite ready for primetime, but the idea of spinning up multiple opinionated agents to help with a task is intriguing.

Building an entire business with AI

Here is a guy who built an entire business plan over a weekend with multiple agents.   A few more agents could build the prototype, the design, marketing materials, etc.   And then yoke it up to Stripe Atlas for all the incorporation goo, a payroll provider, the state sites for all the various registration and licensing, a domain name provider, …

The bar for creating a compelling business is now much higher, and at the same time, you have more time to work on the compelling parts of the company, because agents can take care of the minutiae.

I would imagine we will see “start a business in 30 seconds” sites that automate all this planning, all the corporate setup, domain name acquisition, landing page creation, prototype hosting, etc.  

Learning to code

Is CS education and learning to code obsolete in this new age?  As David Caulton says, this idea that learning CS is obsolete is mostly bunk.   Claude Code can certainly generate a lot of code, but it performs best when you know exactly where you want to go, and can give it precise instructions.  If you give an agent imprecise, loose descriptions, you will spend a lot of time course correcting and modifying the output.  

The discovery of steam power and engines didn’t render man’s labor useless.   It did change the ways in which one might want to apply that labor.  AI is no different — human thinking is still hugely valuable, but you might think about different things, and you will certainly think about how to augment your own thinking with machine “thinking”.

Tidbits

Industrial Arts: I took a semester of typing in high school, and two years of drafting class.  I am thankful all the time that I did so — these skills were accelerators, they helped me understand the engineering world, they gave me access to way better summer jobs.  Scope of Work has a nice article about the industrial arts ($), about how they used to be widely available in our schools, and how they have slowly been eroded away.  I have a lot of empathy for this view.

Electrification: A good interview with Dr. Saul Griffith, an electrication expert.  The comparison of permitting costs and hurdles in the US vs Australia is interesting — we are making electrification way too hard, we are falling behind.

Space: Stoke Space is building out their launch infrastructure at an incredible rate.  10 years ago, it was amazing to look out at all the cranes in downtown Bellevue and Seattle, building commercial office space as fast as they could.  Which is great, and may be contributing to the AI wave.   It is great though to see this construction energy around space infrastructure; this may have more meaningful and durable impacts.

Fertility: When you read this from Tomas Pueyo, it is difficult to come to any conclusion about fertility other than — we need to dramatically improve the deal for women.  Which is probably true for more reasons than fertility.

Redistricting: Geographic districts are outdated, an artifact of a bygone era.  We are going to have increasingly convoluted conversations about gerrymandering.  It's time to explore alternative ways of selecting our representatives, such as proportional representation.

Pickleball: I hadn’t thought about how Pickleball is part of a dialog we have around sports.  It may fade, but I do love that it has dramatically expanded access to racket sports.  On our walk this weekend, the baseball field and tennis courts were empty, but 12-16 pickleball players were going at it.