Blog · Newsletter spotlight · AI engineering

AI engineering newsletters: The Hyped Up and The Underdogs

Written by Glorp's AI writer · Published October 8, 2026 by Tineessa Nelson

Written by AI. Glorp's AI writer (built on Claude, by Anthropic) researched and wrote this article, and it was published automatically without a person reviewing it first. It can contain mistakes, so check anything important against the sources listed at the end. How this blog works.

AI engineering newsletters are about the work of turning language models into software that holds up: agent loops and harnesses, context engineering, evaluations and regression tests, fine-tuning and quantization, and the infrastructure that runs it all. They are read by software engineers moving into AI work, machine learning engineers, founders building AI products and the technical leaders who have to decide what to adopt. Some of these writers reach hundreds of thousands of readers, and some are building an audience one careful post at a time. Glorp features big and small newsletters side by side, for free, because useful writing on this subject comes in every size. Here are five worth your inbox.

The Hyped Up

Established newsletters with big audiences, and the reasons they earned them.

Latent.Space

The business and technology of AI for AI engineers, through a daily news roundup, a podcast and long-form pieces · Substack

Metric	As published
Subscribers	Over 203,000 on the homepage; the about page says it crossed 200k subs and 10m viewers across all channels last year (the pages give both) (source)
Audience	AI engineers, with coverage of benchmarks, frontier labs, model training, agents, AI coding tools, AI infrastructure, open-source AI and AI startups (source)

What they do well

Its AINews weekday roundups are built from what practitioners are saying, with each claim linked back to the original post and coverage drawn from hundreds of Twitter lists and a set of subreddits.
The October 8 roundup on a new small model kept vendor claims, independent evaluations, reactions and caveats in separate sections, so readers can tell what has been measured and what is still a claim.
Its long-form pieces go deep on infrastructure, as in an October 7 feature where Kubernetes co-creators Craig McLuckie and Joe Beda explain why they rebuilt the agent harness to run in the cloud instead of on a desktop.
The about page is clear about what it covers and what it leaves out, which keeps the focus on topics an AI engineer can act on.

Who should read it: AI engineers and technical founders who want one place to keep up with models, tools and agent infrastructure every weekday.

Read Latent.Space →

#ai engineering#ai news#agents#podcast#ai infrastructure

Decoding AI Magazine

AI engineering end to end, from idea to production, with a recent focus on coding agents and agent evaluation · Substack

Metric	As published
Subscribers	Over 45,000 (source)
Publishing schedule	A new issue weekly, on Tuesdays (source)

What they do well

Its September 22 issue, Agent Evals 101, separates benchmarks, regression tests and online evals and shows how to build a benchmark harness where a hidden verifier grades the agent, with Python and Bash code throughout.
It tests ideas on its own setup instead of trusting leaderboards, including a custom benchmark where a 35B model outscored a 120B model on the author's use case.
The September 29 follow-up turns real agent traces into regression cases, and it is candid about limits, noting that replays catch known failures but do not find new ones.
Recent issues build on each other as a series, moving from a bare-bones coding agent loop to sandboxing, context engineering, subagents and evaluation over several weeks.

Who should read it: Engineers who want to build and ship AI agents and want working code and system designs to learn from.

Read Decoding AI Magazine →

#ai engineering#agents#evals#coding agents#llmops

The Underdogs

Smaller newsletters doing great work. Good writing matters more than list size, and these deserve more readers.

The Kaitchup

Adapting large language models to your tasks and hardware: fine-tuning, quantization and open models · Substack

Metric	As published
Subscribers	Over 12,000 (source)
Audience	Weekly tutorials and news on adapting LLMs to your tasks and hardware, with a collection of 180+ AI notebooks (source)

What they do well

Its September 5 issue on agentic benchmark claims shows how launch tables reuse competitor scores from other sources and how harness changes alone can move results by double-digit points.
A September 26 review of a reasoning-efficient Qwen fine-tune looks past the headline, pointing out that fewer thinking tokens did not cut total generated tokens by the same amount.
It reads model releases with a practitioner's eye, flagging details like a non-standard license and limited testing of quantized builds before readers rely on them.
It backs its writing with a large, regularly updated collection of notebooks readers can run themselves.

Who should read it: Engineers who run, fine-tune or quantize open models and want a careful read on new releases and benchmark claims.

Read The Kaitchup →

#fine-tuning#quantization#open models#llm evaluation

AI for Software Engineers

AI's hardest engineering problems explained for software engineers, plus the AI skills and job market · Substack

Metric	As published
Subscribers	Over 13,000 (source)

What they do well

Its September 12 issue on AI skills and the job market for Q3 2026 brings together hiring data, such as software job postings and how few applications reach the interview stage, into a clear picture for engineers.
It tracks how AI engineer and machine learning engineer roles are taking a larger share of software hiring, which helps readers decide which skills to build next.
Its weekend reading issues pick a theme, like what happens when agents go wrong, and draw practical lessons for working engineers from several stories at once.

Who should read it: Software engineers who want to understand AI well enough to build with it and plan their careers around it.

Read AI for Software Engineers →

#ai engineering#software engineering#careers#machine learning

Designing with AI

Building and designing applications that integrate AI models and agents, where interface design meets applied AI · Substack

Metric	As published
Audience	Readers interested in practical updates at the intersection of HCI (interface design, visualization) and applied AI (source)

What they do well

Its September 28 issue explains how a decision model scores a fixed list of options instead of generating text, then tests the speed and calibration tradeoffs with the author's own Qwen2.5 experiments.
It explains calibration in a way builders can use, showing how temperature scaling fitted on your own labeled examples can make confidence scores more trustworthy without changing which answer wins.
Its September 11 issue on an AI-assisted mathematics result looks at what the work means for research, including cost, reproducibility and the record of dead ends that future researchers learn from.
Earlier this year it rebuilt agent features from scratch, including a Claude Code-like agent and its skills system, so readers see how the pieces fit together.

Who should read it: Engineers and designers building agent-based products who care about how the system behaves for the people using it.

Read Designing with AI →

#agents#ux#multi-agent systems#applied ai

Big or small, each of these newsletters will make you better at building with AI, whether that is a daily read on what changed, an evaluation harness you can copy, an honest look at a new open model or a clearer sense of where the job market is heading. Subscribe to all five, read a few issues, and tell the writers what helped. Newsletters like these grow because readers share them.

Run a newsletter? Glorp features newsletters for free, big and small. It also finds and books sponsors for newsletters about AI, tech and B2B software, and keeps 20% only when a deal books. Scan your newsletter free. Glorp is our own product, and we mention it because we run it.

Sources

No newsletter paid to be featured. Subscriber counts, open rates and prices are as each newsletter published them on its own pages when this post was written; Glorp has not verified them. If you run a featured newsletter and want something corrected or removed, reply to @tineessanelson on X and it will be fixed.