It’s Sunday, there’s tea next to my laptop, and this week’s blogging prompt is a to-want list for the last quarter of the year. I’m writing mine right now, so this is a plan, not a result.
A to-want list is a list of goals you actually want, written as if they were already done and checkable by 31 December 2026. I’m borrowing the format from Judith Peters, who adapted it from the book “Das 12-Wochen-Jahr”. The idea is to treat a quarter like a short year, instead of waiting until December to panic.
This is my first to-want list on this blog, so there is nothing to tick off from a previous quarter.
Two decisions up front. First, I only list goals for this site, Code & Value. Second, many of my goals are “I find out X”. That fits the series: most posts here start as a weekend question. If a question is still open on 31 December, I’ll say so in my review.
Hermes and my four agents
I currently run four agents with Hermes: Ada, Ivy, Clark and Pi. Several questions from the very first post in this series are still open, especially which model fits which agent and what it all costs.
- I test which models, including open ones, work best for each of my four agents, and I publish my choice per agent with reasons.
- I compare what it costs to connect models to Hermes: vendor subscriptions, a Nous Portal subscription and the OpenRouter API. I use current prices from the providers’ own pages and my own real usage, and I publish the comparison with a date.
- I find a way to see what my agents do in the background (which tools they call, which subagents they start, which permissions they have), and I publish what I could see and what stays invisible.
- I connect a separate Google account to one of my agents, document every step, and try out how the agent can work with Google tools and which useful applications come out of it. I already know from my job radar post that Google accounts can be a small adventure 🙈.
- I test whether I can use the Gemini models in Hermes through an API key. As of today I have Google AI Pro (5 TB) again as a student, for 5.49 EUR instead of 21.99 EUR (Gemini for students). Before signing up I asked Gemini itself whether that works, and it said yes. That doesn’t count for much until I have tried it myself, so this goal is the test.
Pi and Clark
- I keep testing how helpful Pi is for the Advanced Mathematics modules of my M.Sc., and I publish a verdict, including the answers that were wrong.
- I watch Clark’s daily news briefing and publish a post on what it got right, what it missed and what it cost.
Claude app and Claude Code
- I rebuild something I saw in a YouTube video: a mini-training with video and quizzes.
- I try out mods in Claude Code, the small TypeScript functions that hook into Claude Code’s events and change what the agent does, and I publish what I tried. According to Anthropic, mods are not sandboxed and run with the same access to my machine as Claude Code itself, so I only use mods from sources I trust.
- I write an article about the SQL and Python trainer I built for my students: the idea, how it was built, how it is meant to be used in teaching, and its limits.
From my ideas list
- I try to build an interactive architecture diagram with Claude and Archify, test it with a real diagram, and report what was usable.
- I look at how frontier model intelligence has developed over time, using the Artificial Analysis comparison of intelligence against output tokens per task, and I put the date of the data into the post.
How these goals connect
A few of these goals are really one story. Numbers 1, 2 and 5 are all about the same question: which model for which agent, and how do I pay for it? Number 12 gives the bigger picture for that. Numbers 3 and 4 are about what my agents can see and do, which is the part I find both most exciting and most worth being careful about.
I also know that twelve goals in eleven weeks, most of them with a post attached, is a lot. This is a list of wants, not a promise to finish everything.
What’s next
On 31 December I’ll go through this list and mark each point as done, not done or let go. I’ll write about the ones I dropped, too.
If you have a (technical 🙃) to-want list of your own, I’d love to read it.
A note on how this post came about
I wrote this draft together with Ivy, my writing agent, which runs in Hermes on
anthropic/claude-sonnet-5.5through the Nous Portal. Ivy turned my notes and my ideas list into a first draft. I reviewed the goals, the wording, the links and every claim before publishing.
Sources