My dearest gentlest reader, I felt a little bit creative, so here goes nothing:

The longest days are finally here,

The brightest month of the year,

A delight to feel the warmth of June,

With melting ice cream on a spoon.

I hope the above poem livened your day today! I also hope June has been kind to you.

Another fortnight, another issue that reveals what’s going on in the AI scene. Anthropic’s new Claude Fable 5 was found to be the strongest publicly accessible AI model across multiple benchmarks, beating OpenAI’s GPT-5.5, which was taken offline following a U.S. government directive. The directive was based on security concerns due to its complex reasoning. There is growing tension between government oversight and AI advances, which means the best-performing model might not be available to the general public.

I have been following a series of System Design questions by Joud, which you should too. I am currently on Day 38. The question explores how AI systems handle documents that are larger than the LLM’s context window. The exercise focuses on tradeoffs between four architectural approaches:

• RAG(Retrieval Augmented Generation): retrieve relevant chunks at query time to reduce token usage and improve scalability

• Summarisation pipelines: recursively summarise the document into smaller chunks that fit the context window

• Hierarchical retrieval: combine summaries and chunks so that the model can navigate the document at multiple levels

• Long-context models: use models with larger context windows, which trade simplicity for high cost and memory usage.
  
  The objective is that there is no universal solution. It is subjective. A good answer would highlight retrieval quality, latency, risks such as costs and hallucinations, and how different types of documents influence the architecture to be used. RAG works when users need answers to specific questions. Summarisations are useful for understanding the overall content, and hierarchical retrieval uses the document's breadth and depth, while long-context models are the simplest but often the most expensive.

Writing is the Thinking argues that writing is both a means of communicating ideas and a process by which ideas are developed and refined. Intellectual work happens when you challenge and improve your ideas through revision, which shows up in problem-solving, intellectual growth, and learning.

I Had No Idea Building a Cart Was This Deep. E-Commerce shopping carts have complexities that require you to account for discounts/coupons, payment options, tax calculations, shipping rules, and guest or authenticated users. This article is a reminder that software complexity often shows up from the countless exceptions, integrations and operational requirements needed to meet business expectations and ensure reliability in production.

Why AI is Replacing Junior Developers and How Mid-Level Engineers Can Survive. AI is automating the habitual tasks that were traditionally performed by juniors, which has now shifted the value of software engineers toward architecture, problem-solving and business understanding. The engineers who survive will be the ones who make better technical and product decisions rather than simply write code.

Why AI hasn’t replaced software engineers, and won’t. AI is increasing engineers' productivity by automating menial tasks, but it has not yet replaced software engineers because software engineering is primarily about making trade-off decisions, accountability, and delivering reliable, sustainable systems. Value has shifted from writing code to problem definition, architecture and ownership.

Doing nothing at work. Doing nothing is not always unproductive. Maintaining time for slack reduces mistakes and creates time and space for thinking. Productivity is about solving the right problems rather than staying busy every minute.

Python 3.14.6 and 3.13.14 are now available!

AI Job Grief: The Unnamed Psychological Crisis Hitting Tech Workers. People are experiencing grief, stemming from seeing the skills, expertise, and professional identities that took years to build suddenly become less valuable as AI becomes more capable.

Tailwind is great. Tailwind sucks. Tailwind is not a silver bullet. Tailwind is also not a mistake. Both critics and adopters are right, depending on their use cases. Tailwind is a trade-off between speed and consistency on one hand, and cleaner markup and greater flexibility on the other. The right choice depends on your project’s priorities, not if Tailwind is inherently good or bad.

The word of the day is auspicious. Auspicious means to be favoured by fortune or promise success.

Example in a sentence:

The company's record-breaking first-quarter sales provided an auspicious start to the fiscal year.

I am making my grand departure into the unknown.

Take care of yourself!

Until the next fortnight, my treasured reader, go forth, and may the odds be ever in your favour! 👏 🤖 ✊ ☠️ 🏹 🪖

Do you enjoy the latest issues of my newsletter? Buy me a piping cup of hot chocolate today!

Cup of Hot Chocolate