wayanjimmy
ENID

Newsletter #25

· 2026-W25 · 2026-06-15 to 2026-06-21

Modern Engineering Values #

— Christoph Nakazawa (creator of Jest, ex-Meta)

tl;dr: The AI era shifts the primary challenge from “how to write code” to “how to make decisions.” Small teams of 2–3 people are now taking the lead. Product and design taste becomes the ultimate differentiator since AI can generate any code. Nakazawa himself barely writes code manually anymore—all his projects are 90–100% generated by AI, even in languages like Rust which he doesn’t know.

EngineeringCulture AI #

Software Engineering at the Tipping Point: Why Your Developer Ecosystem Won’t Survive 10x Unchanged #

— Adam Bender @ Google I/O 2026

tl;dr: AI accelerates code writing by up to 10x, but the supporting systems (code reviews, compilation, testing, releases, and bug fixes) are not yet ready to keep pace. Without a solid foundation, this speed increase risks piling up technical debt and chaos. The biggest challenge is no longer how to write code quickly, but maintaining human comprehension over increasingly complex systems.

SoftwareEngineering AI #

Mastering Tech Lead Management #

— Christoph Nakazawa

tl;dr: The Tech Lead Manager (TLM) role is a unique hybrid of team manager and hands-on technical contributor—not just a manager who can write code or a programmer doing admin work. His seven key tips:

  1. Prioritize human relationships over technical issues.
  2. Complete 1:1 syncs early in the week.
  3. Coordinate asynchronously via writing to protect deep work from meeting interruptions.
  4. Use code reviews to coach and mentor the team.
  5. Make decisions quickly without waiting for 100% consensus.
  6. Think holistically about the impact of workflow systems.
  7. Shield the team’s focus from external noise.
Leadership Management #

Is Go the Perfect Language in the World of AI? #

— Franziska Hinkelmann, PhD (@fhinkel)

tl;dr: In the AI era, a developer’s job shifts from writing code to acting as a code reviewer. Go’s rigid and “boring” structure becomes a major advantage over Python or JavaScript because the code is highly readable. Additionally, the Go compiler helps catch AI hallucinations instantly before runtime.

ProgrammingLanguages Go AI #

JuiceFS at Xiaomi: Unified Storage #

— JuiceFS Team

tl;dr: Xiaomi replaced HDFS, PFS, and NAS with JuiceFS as unified file storage. Cost savings range from 70% to 90%. Cache layer (NVMe + RDMA) acts as a game-changer for AI training. EB-scale running hundreds of billions of files—one of the largest JuiceFS deployments in the world.

Infrastructure Storage #

Stitch + Antigravity CLI: Frontend Design Workflow #

— Guillaume Laforge

tl;dr: A design-to-code pipeline built for backend developers. The DESIGN.md format (design tokens in markdown) is AI-readable and version-controllable. Stitch generates the UI from these tokens, while the Antigravity CLI allows iteration via natural language. Now backend developers can ship decent frontends.

Tools Frontend AI #

Google Gemma 4 12B - Encoder-Free Multimodal #

— @analogalok

tl;dr: Google’s first encoder-free multimodal model. Vision and audio are projected directly into the LLM backbone without a frozen encoder. 6.6 GB quantized, running at 21 tok/s on an RTX 4060 with a native 256K context under Apache 2.0. Runs locally on budget GPUs.

AIHardware Models #

Discobot Architecture Study #

tl;dr: A 738-line deep-dive into agent-in-container architecture. Features a per-sandbox MITM proxy for network isolation, a sudo gate for credential authorization, and sandbox orchestration. An isolated sandbox architecture communicating via proxies—an interesting pattern for production agent systems.

AgentArchitecture DeepDive #

Inside Evonic’s Memory System #

tl;dr: Evonic’s memory system uses SQLite + FTS5 BM25 + LLM extraction—without vector embeddings. The philosophy: memory is a compressed conversation, not a queryable knowledge graph. A refreshingly different approach from the mainstream vector database path.

AgentArchitecture Memory #

Subscribe