I’m challenging myself to build 10 things in 10 days using agentic AI!

I’ve been … ambivalent… about Ai tools for quite a while. This morning I was talking to @mitch_nz who suggested I should try out different tools and build 10 things in 10 days.

For day 1 I decided to resurrect my website at http://www.lukesleeman.com

Tools I used

Visual studio code, with the Clien plugin using the claude 4.0 sonnet model.

The website is hosted using GitHub pages.

What went well

• The websites back up and running!
• The tool was able to do a surprisingly broad range of things. It helped me diagnose DNS issues, and searched for nice looking blog themes.
• I was able to get a surprisingly large amount of things done, even though I was “coding” with stacks I knew very little about.
• Right at the end I decided to try and build a mastodon to blog bridge, that reposts my toots to the website - I was able to build it from scratch in just over an hour!

What went badly

• The tool can do a wide range of things, many of which involve executing commands. It surprised me by running a git commit and push to main, sending its work straight out into production. YOLO I guess 🤷‍♂️
• It costs 💰💰💰! By the end of the day I had burnt through $30 of tokens
• It hallucinates and makes things up, and makes mistakes. This isn’t as bad as would you expect - you just need to keep an eye on what it’s doing and poke it in the right direction.
• At various times it crashed, errored out, etc

What I learnt

• The tools are surprisingly powerful and not to be underestimated. You can do useful work quickly.
• To get production level code you need an engineer who can write production level code overseeing and guiding it.
• You don’t need production level code for everything. I’m kind of fine with the mastodon to blog bridge thing being a file of hacky JavaScript.

Originally posted on Mastodon