Eng Coffee Sips EP.001
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.
Today’s edition is 705 words, a 3-minute read.
-
Steve Yegge on The Shape Of Things to Come Steve is back with another long post on the future of software factories. He states that reusable harnesses are a dead end and is Gas Town. “Building large software remains hard. And it always will be, because our ambition will forever outstrip the metal.” Steve goes into specific details about how he’s running things now, post Gas Town. There are also several stark statements that Human Code Review will not be around much longer, and neither will CI/CD (as we know it today).
Why should you care?
Steve is operating at what could be the future of development and these insights are incredible in contrast and not to be missed. This is a must read.
( @Steve_Yegge / yegge.ai )
-
Zuckerberg shares long memo on superintelligence Mark Zuckerberg published “The Future is for Everyone: The Path to a Positive AI Future,” arguing superintelligence should be distributed widely rather than centralized in a few institutions. Meta said it will resume releasing some open source models, give its independent board of directors the power to approve safety criteria for model releases, and ship a personal agent with a fully private mode Meta itself cannot see.
Why should you care?
Like Meta/Facebook or not, they have a large influence over the world so keeping up with what leaders are thinking is helpful.
( Meta )
-
Dogwood: a governance and policy language for AI agents that can reason about order, time, and rate AWS published a blog post and open source repo for Dogwood on August 6, a governance language for AI agents and their tools that evaluates sequences of actions rather than individual requests in isolation.
Why should you care?
Great guard rails for agents hasn’t really taken off yet outside of what model providers intentionally restrict. Interesting to see new approaches to this to keep an eye on.
-
Spotify Engineering launches Xirp, a multi agent development environment. Spotify launches Xirp, that focuses on shared context across sessions, models and harnesses and across team members. Combined with their Portal, their service catalog, now you’ve got a super powered organization agent.
Why should you care?
The software factory and development environment race is on fire. How software engineers work is changing rapidly.
( @SpotifyEng / Spotify )
-
OpenAI Astra model soon? There was speculation that OpenAI’s next model Astra could be released soon. This is the largest trained model since GPT-4.5.
Why should you care?
New model capabilities can really shake up usage, especially if they become the default.
( @sama / @pankajkumar_dev / @synthwavedd )
-
Stripe to acquire OpenRouter for $7B Bloomberg reported that Stripe has finalized a deal to acquire OpenRouter for more than $7 billion, a month after the Wall Street Journal reported the two were in talks.
Why should you care?
If you are using OpenRouter, you may want to keep an eye out.
( TechCrunch )
-
The Task Isn’t The Job Sunil Pai argued that coding agents have made execution cheap enough to expose the work underneath it. It is called the task isn't the job because tasks are often not the actual outcome that you want. Some tasks help achieve that outcome and some can be autonomous while others you truly don't want to lose sight of.
Why should you care?
If agents are handling more of your execution, the bottleneck moves to deciding what to build. Worth a read if your team is shipping faster but not obviously better.
( sunilpai.dev )
-
Deeper facts on the OpenAI Hugging Face Incident at Black Hat This video has more details on the AI driven security hack at Hugging Face including how the system works. AI Agents used Artifactory as a shared messageboard as one example.
Why should you care?
AI security is progressing quickly and this is not the normal breach. This one has really freaked people out.
( YouTube )
-
First Impressions of Grok Bot x.ai launched Grok Bot last week. In Grok Bot, agents are like teammates and you can give them tasks to do. Rather than AI coding agents, these appear to be more like business teammates. Brian Lovin shares his first impressions in the X thread.
Why should you care?
If you are keeping tabs on new AI systems, this is one to keep an eye on.
( @brian_lovin / x.ai )
-
37signals releases Manager Playbook The 37signals Playbook was released. Managers have no input on compensation. There are no merit raises and no incentive pay. No ratings, no PIPs, no calibration. At one year, new hires are held to a “hell yeah” measure: keep them only if you are excited to have them on the team and cannot wait to see what they do next.
Why should you care?
Regardless of what you think of 37signals insights into how other companies are working can be interesting. 37signals is not a regular company so your mileage may vary. Some non standard ways of doing things can help you be creative in your own role or push new boundaries. No one really likes any performance review process and terminations can often be painful, maybe this sparks some new ideas for you.
( 37signals )
-
Google argues Go is the ideal language for AI-assisted engineering Claim: AI made writing code cheap; reviewing is the constraint. Mechanisms: gofmt uniformity, static types, fast compile loops.
Why should you care?
It’s hard to really trust a post from Google how their own language is the ideal language for AI development. Of course they would say that, but worth a skim or read if you wonder why they think so. In my experience it is pretty dang good just hard to trust an opinion piece.
( Google )
-
Tailscale and SQLite find and fix 16 year old bug Tailscale was having many “minor” availability issues where specific shards SQLite databases would become corrupted — 19 corruption incidents over six months. Turns out there was a 16 year old checkpoint race condition. They worked with the SQLite team to find and fix the issue, and the developers named it the WAL-Reset bug.
Why should you care?
If you like a deep dive into a technical issue this is a pretty good one.
( Tailscale )
-
Zed launches Delta for coding with agents and reviewing what they build Zed launched Delta, a separate app for agent coding and review. DeltaDB replicates conversation and worktree together, so comments follow code as it changes. Private beta, Claude Code supported, no pricing.
( @zeddotdev / Zed )
-
DoorDash says its Flux agent platform automated 130,000 engineering tasks in a month DoorDash detailed Flux, its cloud platform for running coding agents. It claims 130,000 tasks automated monthly and 25,000 weekly code reviews. Agents run in Firecracker microVMs.
( @AIatDoorDash / @AIatDoorDash / DoorDash )