>_ Eng Coffee Sips

All editions · EP.010 · 2026-10-19

Eng Coffee Sips EP.010

Ambition

Eng Coffee Sips is a short factual based newsletter for articles that might interest engineers or engineering leaders, not just AI news. Short and to the point to skim over your morning coffee.

In this edition

By Pete Shima · October 19, 2026

Today’s edition is 620 words, a 3-minute read.

  1. You need to be more ambitious with AI models Felix Rieseberg, who leads engineering for Claude’s desktop apps, splits ambition into Speed, Altitude, Depth and Range, and skips Speed as solved.

    My Take

    If you are working with AI models this is pretty much a must read. Outside of just token limits, Ambition could be the primary thing holding you back from progress with AI. I like the Altitude, Depth, and Range framing, basically increasing in all 3 areas. Act as an expert I use fairly often and it works pretty well, Altitude I can sometimes struggle with because AI can go off the rails in what I want and can end up spending more time correcting things than building. Range is interesting, haven't really tried that yet.

    ( felixrieseberg.com / @felixrieseberg / Hacker News / be-ambitious skill (GitHub) )

  2. Anthropic replaced git with an in-house version control system after merges hit 30,000 a day Anthropic’s monorepo went from about 2,500 merges a day to more than 30,000, and git mirroring became the bottleneck. A seven-person team built Scion, a server-side, always-online VCS, and migrated off git in five months. It is not being open-sourced. Anthropic calls it unsuitable as a general git replacement.

    My Take

    I like this article because it is AI use at scale solving a real problem. I do wonder what the fate of git will be. I am sure it will stick around for a very long time but I wonder what will emerge as challengers in the next 6-12 months. Scion is built to be much more AI friendly and helps with very large monorepos, being able to pull only the diffs you need when you need it. It is also designed to be always online, a fundamental departure from git.

    ( Anthropic Engineering / Hacker News / sunshowers: heavily inspired by Mononoke / @AnthropicAI )

  3. OpenCode’s hosted Console was breached and users’ stored API keys were drained Attackers got into a legacy system behind OpenCode Console, the hosted service where teams store OpenAI, Anthropic and Google keys. OpenCode says 1.4% of Console users were hit and ties it to a large-scale distillation attack. Users reported hundreds of dollars of unauthorized usage; Anthropic revoked affected keys and OpenAI is refunding drained credits. The CLI isn’t affected.

    My Take

    This is pretty scary. We've seen some scares in the past like with hugging face but I don't recall seeing a breach like this. What is even scarier is if this is what we are hearing about, what are all the cases we aren't hearing about or haven't even been discovered yet. With how many people are using these AI systems it seems like breaking trust over security issues could completely derail your product if it was large enough. I wonder how much of a hit this will be to OpenCode or if it will even survive this.

    ( @opencode / @thdxr: Console, not OpenCode itself / @thsottiaux: OpenAI refunds / The Hacker News )

  4. A streamer’s “hive coding”: chat prompts, agents ship overnight Max Blade launched Kong Royale, a 50-player browser battle royale built in three weeks on two Claude Max accounts. His “hive coding” loop: viewers type fixes into the stream chat, a couple of dozen agents filter them overnight, and a morning build lands. He says one night shipped 90 changes.

    My Take

    Vibe coding is cool but what about hive coding? Is this actually the future of product development? I don't think you can just build whatever users ask for, but maybe with the guard rails and human oversight you could?

    ( @_MaxBlade / @_MaxBlade: 25 agents + top chat messages / Kong Royale / The AI Daily Brief )

  5. Claude Platform now emits OpenTelemetry traces for the Messages API and Managed Agents Anthropic’s API can now send OpenTelemetry spans for every model request and tool use to any OTLP endpoint, with no code change: an admin adds a destination in the Console. Default spans carry duration, status, token counts, model and tool names; a “full content” mode also exports prompts and responses.

    My Take

    Just kind of interesting to see this sort of integration.

    ( Claude Platform docs / Anthropic / @ClaudeDevs )