Hacker News: zmccormick7

Context is the bottleneck for coding agents now

zmccormick7 — Fri, 26 Sep 2025 15:06:42 +0000

Article URL: https://runnercode.com/blog/context-is-the-bottleneck-for-coding-agents-now

Comments URL: https://news.ycombinator.com/item?id=45387374

Points: 196

# Comments: 187

New comment by zmccormick7 in "Show HN: Runner – the anti-vibe coding agent"

zmccormick7 — Fri, 12 Sep 2025 14:10:14 +0000

Good to know. I've heard great things about Context7, but haven't experimented with it yet.

New comment by zmccormick7 in "Show HN: Runner – the anti-vibe coding agent"

zmccormick7 — Fri, 12 Sep 2025 14:09:21 +0000

As in the download itself didn't happen when you clicked the download button, or the installation failed?

New comment by zmccormick7 in "Show HN: Runner – the anti-vibe coding agent"

zmccormick7 — Fri, 12 Sep 2025 14:08:02 +0000

Cool, I hadn't heard of Traycer. That does look quite similar!

Completely agree. I basically built Runner to codify the process I was already using manually with Claude Code and Gemini. A lot of developers seem to be settling on a similar workflow, so I'm betting that something like Runner or Traycer will be useful for a lot of devs.

I'll be curious to see how far the existing players like Cursor push in this direction.

New comment by zmccormick7 in "Show HN: Runner – the anti-vibe coding agent"

zmccormick7 — Thu, 11 Sep 2025 22:26:31 +0000

I agree that's a major problem. It's not something I've solved yet. I suspect a web research sub-agent is likely what's needed, so it can pull in up-to-date docs for whatever library you need to work with.

New comment by zmccormick7 in "Show HN: Runner – the anti-vibe coding agent"

zmccormick7 — Thu, 11 Sep 2025 22:22:53 +0000

Gemini is required for the context management sub-agent. You can use any of OpenAI, Anthropic, or Gemini for the main planning and coding agents, but GPT-5 performs the best in my experience. Claude 4 Sonnet works well too, but it's about twice as expensive.

Show HN: Runner – the anti-vibe coding agent

zmccormick7 — Thu, 11 Sep 2025 18:59:40 +0000

Now that AI is capable of writing large volumes of production-quality code, our role as developers is changing. Our primary job is no longer writing code. It’s planning and communicating software design and architecture. We have to do this collaboratively with agents and then review and iterate on their implementations.

IDEs were not built for this workflow. So about three months ago I decided to try to build what I thought this new interface should look like.

Runner is a coding agent purpose-built for this new “plan and review” workflow. It’s not for vibe coding. It’s for professional software developers who are responsible for the code they ship.

It encourages and supports a more structured and controlled workflow than other coding agents. It’s built around the concept of tasks. A task is a small, clearly scoped change. The planning agent creates and edits task specs, and then you can assign them to coding agents once you’re happy with the plan. When the coding agent finishes, you can review the changes via the built-in diff viewer. If you’re happy with them you can approve the changes, which will trigger a git commit.

Runner is available as a free BYOK beta for MacOS right now. You can learn more and download it here: https://runnercode.com/. You will need at least a Gemini API key, and for best performance also an OpenAI API key.

Comments URL: https://news.ycombinator.com/item?id=45214955

Points: 22

# Comments: 10

Bootstrapping a Coding Agent

zmccormick7 — Tue, 26 Aug 2025 21:42:47 +0000

Article URL: https://runnercode.com/blog/bootstrapping-a-coding-agent

Comments URL: https://news.ycombinator.com/item?id=45032676

Points: 2

# Comments: 0

New comment by zmccormick7 in "Show HN: MinDB – an extremely memory-efficient vector database"

zmccormick7 — Mon, 23 Sep 2024 01:02:31 +0000

The main thing we need to add is metadata filtering, as that's required for a lot of use cases. We're also thinking about adding hybrid search support and multi-factor ranking.

New comment by zmccormick7 in "Show HN: MinDB – an extremely memory-efficient vector database"

zmccormick7 — Fri, 20 Sep 2024 21:01:08 +0000

We've only done full benchmarking with the FIQA dataset, comparing minDB with Chroma. We're going to try it with Qdrant and Weaviate soon too, since they both have support for quantization, which will be a more apples-to-apples comparison with our approach.

We did test uploading and querying a Wikipedia dump, which was ~35M vectors. Query latency was around 150ms and peak memory usage was 1.5GB. We couldn't test recall, though, because we didn't have queries with ground truths.

Show HN: MinDB – an extremely memory-efficient vector database

zmccormick7 — Fri, 20 Sep 2024 07:15:19 +0000

Article URL: https://github.com/D-Star-AI/minDB

Comments URL: https://news.ycombinator.com/item?id=41599605

Points: 17

# Comments: 5

Show HN: Retrieval engine with SOTA performance on challenging RAG benchmarks

zmccormick7 — Thu, 22 Aug 2024 16:39:34 +0000

Article URL: https://github.com/D-Star-AI/dsRAG

Comments URL: https://news.ycombinator.com/item?id=41322022

Points: 2

# Comments: 0

New comment by zmccormick7 in "Solving the out-of-context chunk problem for RAG"

zmccormick7 — Thu, 25 Jul 2024 15:08:31 +0000

Agreed that thresholds don't work when applied to the cosine similarity of embeddings. But I have found that the similarity score returned by high-quality rerankers, especially Cohere, are consistent and meaningful enough that using a threshold works well there.

New comment by zmccormick7 in "Solving the out-of-context chunk problem for RAG"

zmccormick7 — Wed, 24 Jul 2024 16:13:21 +0000

Agreed. Retrieval performance is very dependent on the quality of the search queries. Letting the LLM generate the search queries is much more reliable than just embedding the user input. Also, no retrieval system is going to return everything needed on the first try, so using a multi-step agent approach to retrieving information is the only way I've found to get extremely high accuracy.

Solving the out-of-context chunk problem for RAG

zmccormick7 — Mon, 22 Jul 2024 13:44:19 +0000

Article URL: https://d-star.ai/solving-the-out-of-context-chunk-problem-for-rag

Comments URL: https://news.ycombinator.com/item?id=41034297

Points: 260

# Comments: 89

Embeddings are not all you need

zmccormick7 — Wed, 03 Jul 2024 18:09:41 +0000

Article URL: https://d-star.ai/embeddings-are-not-all-you-need

Comments URL: https://news.ycombinator.com/item?id=40868547

Points: 3

# Comments: 0

New comment by zmccormick7 in "Show HN: I made a search engine for Hacker News"

zmccormick7 — Wed, 03 Jul 2024 16:29:14 +0000

I had the same issue when searching for specific companies/products. It feels like a pretty basic vector search with no hybrid search component or reranking.

New comment by zmccormick7 in "Show HN: SpRAG – Open-source RAG implementation for challenging real-world tasks"

zmccormick7 — Sun, 05 May 2024 18:06:50 +0000

That should work well with the default parameters, so you shouldn't have to do anything special.

New comment by zmccormick7 in "Show HN: SpRAG – Open-source RAG implementation for challenging real-world tasks"

zmccormick7 — Fri, 03 May 2024 17:53:31 +0000

That description is a little vague, so I need to improve that. The use cases we're focused on are ones with 1) dense unstructured text, like legal documents, financial reports, and academic papers; and 2) challenging queries that go beyond simple factoid question answering. Those kinds of use cases are where we see existing RAG systems struggle the most.

New comment by zmccormick7 in "Show HN: SpRAG – Open-source RAG implementation for challenging real-world tasks"

zmccormick7 — Fri, 03 May 2024 17:48:55 +0000

I think the difference is that they're building for end users, not developers.