At the last ClickFunnels hackathon, I gave a spontaneous talk on skills and local engineering setup, and asked the team to roast me on my setup. One roasting comment was whether one of my new favorite skills is really that expedient: /grill-me

Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk down each branch of the design tree, resolving dependencies between decisions one-by-one. For each question, provide your recommended answer.

Ask the questions one at a time, waiting for feedback on each question before continuing. Asking multiple questions at once is bewildering.

If a question can be answered by exploring the codebase, explore the codebase instead.

During the roast, I went the consultant route and responded with something along the lines of "it depends", and I think it still holds: In some cases, you know exactly what you want but expect some unknowns. In that case, /grill-me is useful to nail down the details. In other cases, you are confident that the machine will do the job or want it to be creative, in which case the grill skill would be redundant or too constraining.

[AICN#26] AI Content Nuggets week#26

/ultracode

I felt like I and some other good folks misused the ultracode effort in Claude Code (burning tokens and token limits and extra usage and all that good stuff that claude.ai came up with to have us generate and use tokens, one way or another). So, I skimmed through YouTube this week. There is a, let's say, "generous" number of videos. I'm not sure if there is a good quality one, but I think you will save yourself some time if you go directly to the source and learn there:

Orchestrate subagents at scale with dynamic workflows - Claude Code Docs
Dynamic workflows orchestrate many subagents from a script Claude writes and you can rerun. Use them for codebase audits, large migrations, and cross-checked research.
Claude Code Docs

The point of this is that to understand Claude Code's ultracode you will need to understand Claude Code's workflows. Once done, ultracode I just a way to ask to dynamically invoke their workflows on every turn when you have the ultracode effort enabled, so Claude will make an effort to run agent orchestration and complex task execution where it sees fit. I've been using the mode substantially for the last few weeks and subjectively feel like it's good results.

Every question your AI asks is a system gap 

(Todd Dickerson, @ToddDickerson) - Kill manual handoffs whenever you can. Treating agent friction as a systems problem.

OpenAI benchmark drop shared from X 

Allegedly, the new GPT model is competitive with Opus at roughly a third of the output tokens. Stealing China the show?

Building a Sustainable Future in APIs with Kin Lane

A refreshing radical anti-AI chat for my AI psychosis bubble :D

Plus known Builders from the industry working on making software with AI in a more sustainable way. Who knows, maybe this is the path many companies will take in the future.

Building a Sustainable Future in APIs with Kin Lane — APIs You Won’t Hate — Overcast
Overcast

just for fun

(I'm having fun so you don't have to)