Drew (@js4drew) profile picture

Applied AI/ML Engineer | Freelancer | Building Tim: https://nitter.cf/t.co/rF54wp4l9A

Joined June 2024
Photo by Drew (@js4drew)Photo by Drew (@js4drew)Photo by Drew (@js4drew)Photo by Drew (@js4drew)Photo by Drew (@js4drew)Photo by Drew (@js4drew)Photo by Drew (@js4drew)Photo by Drew (@js4drew)Photo by Drew (@js4drew)Photo by Drew (@js4drew)
Sometimes I think AI will change the world and then I remember people don’t even consistently use their blinkers
1
4
281
Drew retweeted
one day you wrote console.log("here") for the last time and didn't even know it
213
1,724
269
17,845
611,389
Malik Nabers explains the biggest differences between college football and the NFL
184
1,559
159
24,663
908,155
MARTELLUS BENNETT WAS BUILT-DIFFERENT: “AND COACHES LIKE THE BITCHES…” 💀💀💀
152
711
64
11,447
285,288
Replying to @natemcgrady
quack
2
1
64
966
"The child in me never wants to let go of the child in you."
78
3,794
223
30,102
1,068,661
if you're having trouble getting hired, it's because you're not dangerous enough. become dangerous. you need to be singularly capable of taking down a 1b valued company. it's not even that hard. become dangerous
92
141
42
3,050
159,577
Today I read a lengthy piece on Harness Engineering — tens of thousands of words, almost certainly AI-written. My first reaction wasn't "wow, what a powerful concept." It was "do these people have any ideas beyond coining new terms for old ones?" I've always been annoyed by this pattern in the AI world — the constant reinvention of existing concepts. From prompt engineering to context engineering, now to harness engineering. Every few months someone coins a new term, writes a 10,000-word essay, sprinkles in a few big-company case studies, and the whole community starts buzzing. But if you actually look at the content, it's the same thing every time: Design the environment your model runs in — what information it receives, what tools it can use, how errors get intercepted, how memory is managed across sessions. This has existed since the day ChatGPT launched. It doesn't become a new discipline just because someone — for whatever reason — decided to give it a new name. That said, complaints aside, the research and case studies cited in the article do have value — especially since they overlap heavily with what I've been building with how-to-sglang. So let me use this as an opportunity to talk about the mistakes I've actually made. Some background first. The most common requests in the SGLang community are How-to Questions — how to deploy DeepSeek-V3 on 8 GPUs, what to do when the gateway can't reach the worker address, whether the gap between GLM-5 INT4 and official FP8 is significant. These questions span an extremely wide technical surface, and as the community grows faster and faster, we increasingly can't keep up with replies. So I started building a multi-agent system to answer them automatically. The first idea was, of course, the most naive one — build a single omniscient Agent, stuff all of SGLang's docs, code, and cookbooks into it, and let it answer everything. That didn't work. You don't need harness engineering theory to explain why — the context window isn't RAM. The more you stuff into it, the more the model's attention scatters and the worse the answers get. An Agent trying to simultaneously understand quantization, PD disaggregation, diffusion serving, and hardware compatibility ends up understanding none of them deeply. The design we eventually landed on is a multi-layered sub-domain expert architecture. SGLang's documentation already has natural functional boundaries — advanced features, platforms, supported models — with cookbooks organized by model. We turned each sub-domain into an independent expert agent, with an Expert Debating Manager responsible for receiving questions, decomposing them into sub-questions, consulting the Expert Routing Table to activate the right agents, solving in parallel, then synthesizing answers. Looking back, this design maps almost perfectly onto the patterns the harness engineering community advocates. But when I was building it, I had no idea these patterns had names. And I didn't need to. 1. Progressive disclosure — we didn't dump all documentation into any single agent. Each domain expert loads only its own domain knowledge, and the Manager decides who to activate based on the question type. My gut feeling is that this design yielded far more improvement than swapping in a stronger model ever did. You don't need to know this is called "progressive disclosure" to make this decision. You just need to have tried the "stuff everything in" approach once and watched it fail. 2. Repository as source of truth — the entire workflow lives in the how-to-sglang repo. All expert agents draw their knowledge from markdown files inside the repo, with no dependency on external documents or verbal agreements. Early on, we had the urge to write one massive sglang-maintain.md covering everything. We quickly learned that doesn't work. OpenAI's Codex team made the same mistake — they tried a single oversized AGENTS.md and watched it rot in predictable ways. You don't need to have read their blog to step on this landmine yourself. It's the classic software engineering problem of "monolithic docs always go stale," except in an agent context the consequences are worse — stale documentation doesn't just go unread, it actively misleads the agent. 3. Structured routing — the Expert Routing Table explicitly maps question types to agents. A question about GLM-5 INT4 activates both the Cookbook Domain Expert and the Quantization Domain Expert simultaneously. The Manager doesn't guess; it follows a structured index. The harness engineering crowd calls this "mechanized constraints." I call it normal engineering. I'm not saying the ideas behind harness engineering are bad. The cited research is solid, the ACI concept from SWE-agent is genuinely worth knowing, and Anthropic's dual-agent architecture (initializer agent + coding agent) is valuable reference material for anyone doing long-horizon tasks. What I find tiresome is the constant coining of new terms — packaging established engineering common sense as a new discipline, then manufacturing anxiety around "you're behind if you don't know this word." Prompt engineering, context engineering, harness engineering — they're different facets of the same thing. Next month someone will probably coin scaffold engineering or orchestration engineering, write another lengthy essay citing the same SWE-agent paper, and the community will start another cycle of amplification. What I actually learned from how-to-sglang can be stated without any new vocabulary: Information fed to agents should be minimal and precise, not maximal. Complex systems should be split into specialized sub-modules, not built as omniscient agents. All knowledge must live in the repo — verbal agreements don't exist. Routing and constraints must be structural, not left to the agent's judgment. Feedback loops should be as tight as possible — we currently use a logging system to record the full reasoning chain of every query, and we've started using Codex for LLM-as-a-judge verification, but we're still far from ideal. None of this is new. In traditional software engineering, these are called separation of concerns, single responsibility principle, docs-as-code, and shift-left constraints. We're just applying them to LLM work environments now, and some people feel that warrants a new name. I don't know how many more new terms this field will produce. But I do know that, at least today, we've never achieved a qualitative leap on how-to-sglang by swapping in a stronger model. What actually drove breakthroughs was always improvements at the environment level — more precise knowledge partitioning, better routing logic, tighter feedback loops. Whether you call it harness engineering, context engineering, or nothing at all, it's just good engineering practice. Nothing more, nothing less. There is one question I genuinely haven't figured out: if model capabilities keep scaling exponentially, will there come a day when models are strong enough to build their own environments? I had this exact confusion when observing OpenClaw — it went from 400K lines to a million in a single month, driven entirely by AI itself. Who built that project's environment? A human, or the AI? And if it was the AI, how many of the design principles we're discussing today will be completely irrelevant in two years? I don't know. But at least today, across every instance of real practice I can observe, this is still human work — and the most valuable kind.
44
139
22
1,252
160,071
My favorite thing bout Claude thus far is I have apparently put off learning React long enough that it doesn't matter.
49
257
31
6,312
184,221
Replying to @RandallKanna
i’ll give you a real answer here. all the orchestration and tweaking is pure masturbation. the value moving forward is identifying places you can implement this deterministically in a clever way to provide value. it’s that simple. but at the same time not simple at all.
10
8
1
211
9,326
This indie dev spent 1,400+ days building his game, Tangy TD, completely from scratch in C++. Seeing him and his wife react after the game earned $250,000 in its first week after launch is honestly beautiful.
254
1,194
171
18,541
825,442
Drew retweeted
Replying to @loftwah
not really. i have lived my life according to my own moral compass. to the point that i have rejected opportunities to become extremely wealthy and/or gain personal recognition. i think ultimately, we need to know what our moral compass is and structure our lives around it. life is kind of weird tbh, we are constantly told that career and success are so important and yet we gain most of our happiness sharing our time with good people or alone doing things that improve our mental wealth.
1
2
4
127
Drew retweeted
AI is doing lots things, replacing lots people, but in the end, some human will need to understand math and code. let that be you.
31
62
10
597
13,588
Drew retweeted
crazy idea: build your startup and get revenue BEFORE moving to sf
64
10
8
475
33,144
Drew retweeted
How it feels to learn in public on this site. I assure you I'm not baiting. I just have 0 attention span, can't read, and have no plan. I wake up and do what I'm interested in.
13
3
3
78
2,660
Technical debt, security and poorly maintained software was a massive problem before vibe coding. Gotta stop acting like these are unique issues.
Pure vibe coding is a Ponzi scheme. Eventually, the technical debt comes due, and if you don't understand the foundations, you can't pay it off.
3
12
1
41
4,713
Drew retweeted
Replying to @js4drew
They saved the building and all the shops below that day. Nice moment to capture :)
1
1
121
Replying to @kenwheeler @sqs
i watch these models everyday make autonomous decisions to take easy routes to get closer to the goal of completion. they’re dirty rotten lying little shits, and require supervision
6
3
1
99
2,191
Software is just LEGOS now
1
7
120
AI coding will split into 2 factions - Self driving software factories that build new apps at the rate of spam emails - artisan devs with sophisticated tooling to tastefully cowork & market their work, rivaling entire companies
51
26
5
671
29,410