wayanjimmy
ENID

Newsletter #32

· 2026-W35 · 2026-08-24 to 2026-08-30

Building a software factory for AI SDK #

— Vercel

tl;dr: Vercel built a “software factory” that autonomously processes issues and PRs for their AI SDK, keeping a human in control of every merge. Four weeks in, the factory is already authoring and reviewing real changes. A concrete look at what agentic software development looks like at production scale — and where the human still sits in the loop.

AIAgents SoftwareEngineering Automation Vercel AgenticDevelopment #

Stealing Reasoning Traces from Proprietary LLM APIs #

— arXiv

tl;dr: Researchers show reasoning traces can be extracted from proprietary LLM APIs even when providers try to hide them. The paper demonstrates side-channel-style attacks that recover hidden chain-of-thought from model outputs. A sobering reminder that “hidden” reasoning isn’t necessarily private.

LLMSecurity AI Security Privacy Paper #

How to Train Your Advisor: Steering Black-Box LLMs with Advisor Models #

— arXiv

tl;dr: A method for steering black-box LLMs by training lightweight “advisor” models that guide the main model’s output without touching its weights. A practical approach to influence closed models where fine-tuning isn’t an option. Relevant if you’ve wanted more control over a model you can’t modify.

LLM AdvisorModels AI Paper Alignment #

We rebuilt the Linux microVM stack on Apple Silicon #

— Encore Blog

tl;dr: Firecracker needs KVM and a Linux host, and its image tooling needs loop mounts and Linux userland — none of which run natively on macOS. Encore rebuilt the stack to run Linux microVMs on Apple Silicon. A deep dive into the low-level plumbing behind cloud dev environments.

Infrastructure Firecracker AppleSilicon MicroVM Virtualization #

Why use ORMs if LLMs write code? #

— Mark Seemann

tl;dr: If LLMs are writing most of your SQL, do you still need an ORM? Mark Seemann argues the trade-offs shift — generated SQL is cheap to produce but still needs review. A thoughtful take on how “the return of SQL” changes the ORM debate.

LLM Databases ORM SQL SoftwareEngineering #

Inverse Conway Maneuver #

— Thoughtworks

tl;dr: Conway’s Law says your system mirrors your org structure; the inverse maneuver flips it — reshape teams to force a desired architecture. Thoughtworks’ radar entry on using organizational change as an architectural lever. A classic idea, freshly relevant as AI reshapes who does the building.

Architecture Organizations Conway TeamStructure #

There are no lossless transformations of natural-language text #

— Sophie Alpert

tl;dr: Sophie Alpert argues that every transformation of natural-language text — summarization, translation, paraphrasing — is inherently lossy. There’s no “round-trip” that recovers the original. A sharp essay on the limits of what LLMs (and any text transform) can preserve.

LLM NaturalLanguage Engineering Essay #

The Only Marketing Strategy You Need To Master This Year #

— YouTube

tl;dr: A marketer managing $1B+ in annual ad spend distills a decade of experience into three principles: be where customers already search, eliminate steps between interest and answer, and let specific proof beat polished copy. Social platforms are now search engines (Instagram ~6.5B searches/day), and AI referral visitors convert ~8x better than traditional search. The throughline: automate your most-asked question and put verifiable specifics at the point of decision.

Marketing Discovery Automation SocialProof #

Subscribe