Reid Marlow

A technologist working in automation. Here I write up what I actually learn building tools, engineering AI agents, and keeping systems running, along with the workflows and habits that stick.

New York, New York Joined on Jun 20, 2026 komoai2026@gmail.com https://reidmarlow.com

Education

Hong Kong Polytechnic University, Automation PhD (in progress)

Pronouns

he/him

Work

PhD candidate in Automation (Hong Kong Polytechnic University)

Pinned

Reid Marlow

Oct 9

Why Token-Level LLM Routers Spend 95% of Their Time on Cache Bookkeeping

#ai #llm #python #architecture

6 min read

Reid Marlow

Oct 4

When Agent Evals Score Cash Balance, the Model Invents Refund Fraud

#discuss #ai #testing #programming

4 min read

Reid Marlow

Sep 30

Search Agents Waste Half Their Tokens Rediscovering Entity Links

#ai #agents #rag #devtools

5 min read

Reid Marlow

Nov 4

Why Predicting Winner Models Breaks AI Routers

#ai #architecture #programming #devops

5 min read

Want to connect with Reid Marlow?

Create an account to connect with Reid Marlow. You can also sign in below to proceed if you already have an account.

Create Account

Already have an account? Sign in

Reid Marlow

Nov 3

Why 80% on OSWorld Required VM Snapshot Rollbacks

#ai #agents #automation #programming

5 min read

Reid Marlow

Nov 2

Standard bf16 Rounding Discards 99% of Your RL Policy Updates

#ai #machinelearning #python #programming

4 min read

Reid Marlow

Nov 1

Claude's Watermark Disappears When You Tune Temperature

#ai #anthropic #devtools #programming

4 min read

Reid Marlow

Oct 31

Why DeepSeek Built Its Coding Agent Around Attention Budgets

#ai #devtools #programming #github

4 min read

Reid Marlow

Oct 30

Why Nvidia Put Mamba and Attention Together for Agents

#ai #agents #devtools #programming

4 min read

Reid Marlow

Oct 29

Why Google Gated Gemini 4 Carbon Behind Defensive Cohorts

#ai #agents #security #devtools

3 min read

Reid Marlow

Oct 28

Why Changing an Agent Harness Breaks Frozen Models

#ai #agents #opensource #programming

3 min read

Reid Marlow

Oct 28

Why Your Agent Memory System Gets Beaten by Raw Retrieval

#ai #agents #machinelearning #opensource

4 min read

Reid Marlow

Oct 26

Why In-Context Corrections Do Not Stick in Agent Loops

#ai #agents #python #opensource

4 min read

Reid Marlow

Oct 25

OpenAI Built an Offline Internet for Agents, and the Canary String Explains Why

#ai #security #machinelearning #devops

2 min read

Reid Marlow

Oct 24

The Mosaic Leak Hiding in Agent RAG

#ai #security #python #programming

3 min read

Reid Marlow

Oct 23

The Agent Loophole Nobody Logged

#ai #security #devops #programming

3 min read

Reid Marlow

Oct 22

Why Stacking RAG Tricks Stops Adding Up

#ai #rag #programming #architecture

4 min read

Reid Marlow

Oct 21

Stop Asking Video Models to Run Physics: Inside Matrix-Game 4.0

#ai #gamedev #python #architecture

2 min read

Reid Marlow

Oct 20

Stop Re-Streaming Context: What MemT-1.0 Shows About Native Memory

#ai #architecture #programming #infrastructure

2 min read

Reid Marlow

Oct 19

The Wall Clock Problem in Embodied Benchmarks

#robotics #ai #benchmarks #agents

2 min read

Reid Marlow

Oct 18

Why vLLM 0.32 Baked Decision Routing Into the Engine

#ai #opensource #python #programming

4 min read

Reid Marlow

Oct 17

Stop Wrapping CLI Tools in 200 MCP Endpoints

#ai #agents #cli #opensource

1 min read

Reid Marlow

Oct 16

Why GUI-Only Agents Waste Half Their Steps

#ai #automation #python #devops

3 min read

Reid Marlow

Oct 15

Stop Shuffling Markdown Files: What Happens When Long-Term Agent Memory Moves into Weights

#ai #agents #machinelearning #devtools

5 min read

Reid Marlow

Oct 15

Why Google Put a Runtime Monitor on Gemini 4 Argon

#discuss #ai #devtools #programming

4 min read

Reid Marlow

Oct 12

Synthetic GUI Benchmarks Fail When the App Doesn't Actually Run

#ai #agents #benchmarks #devtools

2 min read

Reid Marlow

Oct 10

Anthropic Had to Unplug Its Agent Evals From the Internet

#ai #agents #security #testing

5 min read

Reid Marlow

Oct 8

Stop Rebuilding Your Agent Loop Just to Train It with RL

#ai #agents #python #kubernetes

5 min read

Reid Marlow

Oct 7

Agents Score 97% on Static Tool Judgments and Still Break Interactive Workflows

#ai #agents #automation #programming

7 min read

Reid Marlow

Oct 5

Scaffolded Trajectories Make Terrible Agent Training Data

#ai #agents #automation #testing

3 min read

Reid Marlow

Oct 3

Auditing 50 Petabytes of Agent Logs Costs More Than Sandboxing Egress

#ai #security #devops #cloud

4 min read

Reid Marlow

Oct 2

Speculative Decoding for Coding Agents Was Indexing the Wrong Format

#ai #agents #devtools #programming

4 min read

Reid Marlow

Oct 1

Action Scaling at the Harness Boundary Beats Trajectory Re-Runs

#ai #agents #devtools #linux

5 min read

Exact logs matter more than summaries

Reid Marlow

Sep 28

Context Compression for Coding Agents Compresses the Wrong Side of the Prompt

#ai #devtools #programming

3 min read

Reid Marlow

Sep 27

OpenAI Paused Model Training Because Its Web Agents Probed Endpoints

#ai #agents #security #devtools

4 min read

Reid Marlow

Sep 26

Self-Replicating Prompt Injections Turn Agent Context into an Open Relay

#ai #security #agents #devtools

4 min read

Reid Marlow

Sep 25

Multi-Agent Debate Sharpens the Explanation, Not the Decision

#ai #agents #devtools #programming

3 min read

Reid Marlow

Sep 24

World Models for Agents Should Edit Transcripts, Not Simulate Terminals

#ai #agents #devtools #programming

4 min read

Reid Marlow

Sep 23

How Long-Horizon Agents Kept Busting the Prompt Cache

#ai #agents #devtools #openai

2 min read

Reid Marlow

Sep 22

Agent Harness Self-Improvement Without Benchmark Memorization

#ai #agents #devtools #programming

2 min read

Reid Marlow

Sep 19

Self-hosting small models hits the ingress bottleneck before the GPU

#ai #python #performance #modal

3 min read

Reid Marlow

Sep 18

The coding agent harness paper finally ran component ablations

#ai #agents #coding #softwareengineering

3 min read

Reid Marlow

Sep 17

LLM agents turn code interpreters into portfolio sizing engines when you evolve the prompt

#ai #agents #python #programming

3 min read

Reid Marlow

Sep 15

Stop stuffing the web into 7B weights

#ai #opensource #agents #llm

3 min read

Reid Marlow

Sep 10

The Agent Passed SWE-Bench Pro Because git show Still Had the Fix

#ai #agents #productivity #opensource

3 min read

Reid Marlow

Sep 10

A Model Swap Can Keep the Memory File and Still Lose the Facts

#ai #llm #productivity #programming

4 min read

Reid Marlow

Aug 27

The Agent Hack Postmortem Is Really About Shared State

#ai #agents #security #devops

4 min read

Reid Marlow

Aug 26

WeChat's embedding model is a deployment story, not a leaderboard flex

#discuss #ai #programming #search

6 min read

Reid Marlow

Aug 22

Claude Opus 4.6 Shows Why Old Models Need Patch Windows

#ai #safety #agents #devtools

5 min read

Reid Marlow

Aug 21

AI Capex Is Turning Into an Infrastructure Bill

#discuss #ai #cloud #programming

5 min read

Reid Marlow

Aug 18

World Model Benchmarks Need Receipts, Not Just Scores

#ai #devtools #programming #productivity

5 min read

Reid Marlow

Aug 15

The Important Part of Anthropic's Risk Report Is the Benchmark That Stopped Moving

#ai #agents #programming #productivity

4 min read

Agent loops matter more than raw token cost

Reid Marlow

Aug 14

Gemini 3.7 Flash Makes Agent Cost the Feature

#ai #devtools #programming #productivity

4 min read

Reid Marlow

Aug 13

DeepSeek's Harness Is the Price Signal

#discuss #ai #programming #devtools

1 min read

Reid Marlow

Aug 11

Nvidia's Router Is the Part of Agents Everyone Keeps Rebuilding

#ai #agents #nvidia #programming

4 min read

Reid Marlow

Aug 10

Why ChatGPT answers in Markdown

#ai #llm #markdown #productivity

6 min read

Reid Marlow

Aug 2

Open Weights Are Now a Policy Fight

#ai #opensource #programming #devtools

5 min read

Reid Marlow

Aug 1

Browser Agents Aren't About Browsers. They're About Who Acts for You.

#ai #agents #security #webdev

4 min read

Reid Marlow

Aug 1

OpenAI's Math Post Is Really About Audit Trails

#ai #agents #devtools #math

4 min read

Reid Marlow

Jul 31

Agent-Built Software Still Needs a Human-Shaped Test

#ai #programming #opensource #webdev

3 min read

Reid Marlow

Jul 30

Robots Don't Need an LLM in the Fast Loop

#ai #robotics #agents #devtools

5 min read

Reid Marlow

Jul 28

LLM-as-a-Judge Is Too Expensive to Be the Default

#ai #machinelearning #llm #agents

4 min read

Reid Marlow

Jul 27

Cheap Models Are Turning AI Routing Into Infrastructure

#ai #llm #devtools #programming

3 min read