Seroter's Daily Reading — #754 (April 1, 2026)

By gzfl…hqfj April 4, 2026

State of Java 2025, uncomfortable truths about AI coding agents, triple debt model, cloud maturity gaps, LLM inference optimization, KubeCon EU 2026, ADK agent skills, junior developer futures, PostgreSQL investments, and RSAC 2026 CISO perspectives.

Episode Image

🎧 Listen to this episode

Richard opens Daily Reading List #754 with a note that despite some downbeat reading items, it was a good day. Sometimes you need those cautionary words to keep you sharp. Today’s list digs into AI’s uncomfortable truths, the efficient frontier of inference, Kubernetes as AI infrastructure, and what junior developers should focus on next.

JetBrains’ State of Java 2025 report reveals some surprising data: China dominates Java usage at 37% of respondents, the developer population skews young, and Java 21 has overtaken Java 17 as the most-used version with 40% regular usage. Java 8 finally dips below a third. A notable trend: more developers are forgoing frameworks entirely, a shift for a language historically defined by its framework ecosystem. On AI tooling, ChatGPT leads at 40%, GitHub Copilot at 29%, and JetBrains AI Assistant at 16%.

Google Cloud’s deep dive on five techniques to reach the efficient frontier of LLM inference borrows from portfolio theory: given a fixed hardware budget, there’s an optimal curve trading latency for throughput. The five techniques — semantic routing across model tiers, disaggregating prefill and decode phases, quantization, speculative decoding, and continuous batching with paged attention — represent what’s available today to move from sub-optimal to frontier performance.

A blog post on uncomfortable truths about AI coding agents makes the case for a blanket ban on AI agents in professional production code, citing four concerns: skill atrophy as developers become pure code reviewers, artificially low costs that may not last, prompt injection vulnerabilities, and unresolved copyright and licensing questions. The author acknowledges it might be an “old man yells at cloud” moment but raises legitimate long-term concerns.

Related, a research digest on new forms of AI debt covers Dr. Margaret-Anne Storey’s “triple debt model.” Beyond technical debt, there’s cognitive debt — the erosion of shared understanding when AI writes code developers don’t fully internalize — and intent debt — the absence of externalized rationale in artifacts. The paper calls uncritical acceptance of AI output “cognitive surrender” and argues teams focused only on code quality are managing one-third of their software health risk.

A CIO Dive report on cloud maturity finds only 14% of enterprises have reached the highest level of cloud maturity, per NTT Data’s survey of 2,300 decision-makers. Cloud immaturity now threatens AI deployment plans, and nearly half of cloud leaders cite a lack of AI skills as a key blocker.

A comparison of Cloud Run Jobs vs. Cloud Batch helps pick the right Google Cloud service for run-to-completion workloads. Cloud Run Jobs offers serverless simplicity; Cloud Batch provides hardware control with up to 8 GPUs per VM and inter-task communication via MPI for HPC workloads.

Intuit Engineering’s six takeaways from KubeCon EU 2026 highlight Kubernetes becoming the AI workload foundation — ready or not. Inference is the main event, Google and Anthropic admitted Kubernetes wasn’t designed for AI but they’re using it anyway, and open source communities are struggling with AI-generated PR floods. Kyverno graduated as a CNCF project with contributors up 39% year over year.

Google’s guide to building ADK agents with skills introduces progressive disclosure: instead of monolithic system prompts, skills break knowledge into three levels (metadata, instructions, resources), cutting baseline context usage by roughly 90%. The post covers four patterns culminating in a “skill factory” where agents write their own new skills at runtime.

In a fun companion piece, Guillaume Laforge demonstrates ADK for Java 1.0 with a Comic Trip agent that transforms travel photos into pop-art comic strips using Gemini, Google Maps, and a multi-agent architecture on Quarkus with Java 21 virtual threads.

An InfoWorld article on junior developers argues code is now a commodity and the skills that matter are clear communication, systems thinking, and the judgment to know whether something actually works. The provocative take: “Markdown is the new programming language” and English majors might have a surprising edge.

Google’s open source blog details investments in PostgreSQL, focusing on contributions toward active-active replication through automatic conflict detection, logical replication of sequences, and upgrade resilience improvements including a fix that reduced Large Object upgrade times from days to minutes.

Finally, Google’s Cloud CISO perspectives from RSAC ’26 covers the three-stage AI adoption journey, the rise of AI-powered attacks including autonomous tools like Hexstrike AI, and advice for CISOs to be multi-model, multicloud, and prepared for AI-powered threats.

Source: Daily Reading List – April 1, 2026 (#754) by Richard Seroter

Articles covered: