This week’s review is out.
open.substack.com/pub/alteredc...
29.12.2025 20:43
👍 0
🔁 0
💬 0
📌 0
Reflecting on 2025, a year dedicated to recalibration, I am excited about the plans for 2026, which will focus on accelerating into AI research and writing.
open.substack.com/pub/alteredc...
Latest edition of Altered Craft’s weekly AI review
open.substack.com/pub/alteredc...
Some thoughts towards adapting to an AI involved software development workflow
open.substack.com/pub/alteredc...
Latest post from Altered Craft. Quick Skills guide for those using Claude Code
open.substack.com/pub/alteredc...
You would expect the output of `ind` to suffice. It this just a "try everything" mentality, or does that visual layout resonate in some way with Opus 4.1.
Or just random noise in a LLM's "reasoning"??
That is curious, I witnessed Claude Opus 4.1 in a task to "review a folder's contents" using `tree` to "get a better understanding of folder structure" 🤔
I recently ran a scan of all @github.com repos in the range of 450k stars to 200 stars to see what real world coding agent use looks like.
open.substack.com/pub/alteredc...
This will enable your agent to explore large JSON files without the need to brute force read most or all of the file (token 🔥)
Additionally maintain an annotated schema file for any core JSON files enabling even more precise examination of the files by the coding agent.
☝️Helping coding agents concur JSON
If you work with a lot of JSON data, install `jq` and ensure your agent is aware of it's presence. Any frontier agent will know how to use it, but you can also hint the agent to examine `jq --help` if you think that will help.
This weeks AI review for developers is out
Agent kit plus much more
open.substack.com/pub/alteredc...
Latest post from my newsletter
Levering AI itself to help adapt to the disruption of AI in software development
open.substack.com/pub/alteredc...
My new fav mini-prompt for working with Claude (Code)
- After Claude suggests a "fix":
"""
In your opinion is this "simple fix" addressing a bug or is it a hack? We want to ensure it brings us closer to best practices and it not just addressing a symptom of a deeper arch issue in the code base.
"""
Claude Code then provided the exact commands to create a new IP pool and add it to the cluster, bypassing the need to involve networking specialists."
They fed screenshots of dashboards into Claude Code, which guided them through Google Cloud’s UI menu by menu until they found a warning indicating pod IP address exhaustion.
ScreenShotOps
www-cdn.anthropic.com/58284b19e70...
Kubernetes debugging with screenshots
When Kubernetes clusters went down and weren’t scheduling new pods, the team used Claude Code to diagnose the issue.
#ScreenShotOps
The 2025 Ai engineers Worlds fair is in progress. For those who could not attend (like me), the event organizers where kind enough to put much of the content on YouTube. Great collection of talks!
youtube.com/@aidotengin...
Some really interesting features and approaches for dagger's LLM integration.
MCP support and and elegant approach to agent loops. Best of all, open source!
Peter Mbanugo Has pulled together a nice collection of articles speaking to the advantages and frustrations of pair programming with an AI
pmbanugo.me/blog/peer-p...
#GenAI #SoftwareDevelopment
Here’s the official Google summary (not all ai but most at least AI adjacent) blog.google/technology/...
What a week for generative AI. With conferences from Microsoft, Google, and Anthropic
For those of you using premium AI tiers, how do you measure the return on your investment?
Would love to hear your thoughts! What metrics make the cut for you when evaluating high-tier AI subscriptions, especially when weighing them against the value from the ~$20/mo options?
Google AI Ultra: $250/mo
ChatGPT Pro: $200/mo
Claude Max: Starting at $100/mo
"Pro" Tier AI Access Hitting "Sports Fanatic Cable Bill" Rates... 🤔
These price points for top-tier AI are definitely climbing, especially when many of us are familiar with the pretty solid capabilities of the ~$20/mo tiers.
FireCrawl is hiring Agents
Bit of a publicity stunt (they are also hiring the human who created the agent)
Is this a glimpse of future roles:
” _____ agent operator” (ex: coding agent operator)
What do you think?
techcrunch.com/2025/05/17/...
Went through through @DeepLearningAI's short course on MCP this weekend. Highly recommended! It's led by Elie Schoppik of @AnthropicAI Opens with a high-level overview then details implementation of servers and clients.
#MCP #GenAI #SoftwareDevelopment
learn.deeplearning.ai/courses/mcp...
Excited for the unofficial start of summer with the holiday weekend in the US! After that, we'll have our second Portland AI Engineers meetup, this time in partnership with WeWork.
Sign up today
www.portlandai.engineer/#events
And lastly, Microsoft is open sourcing the Copilot VSCode copilot extension
code.visualstudio.com/blogs/2025/...
Google announced early access to what appears to be a Codex clone
jules.google/
