Eng Coffee Sips EP.008
My Dots, My Dots
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
Today’s edition is 671 words, a 3-minute read.
-
OpenAI DevDay replaces GPT-6 Sol after a week, ships always-on Dots and a Decisions API aimed at Jev OpenAI’s DevDay ran to more than 20 announcements. GPT-6.1 Sol replaces the GPT-6 Sol that shipped a week earlier. Dots are always-on agents on GPT-6 Astra with their own cloud computer and browser; the first is included on Pro. A Decisions API on Luna picks one answer from a fixed list in about 150 ms. A $500 Pro tier gets Ultrafast, and the reopened $200 Pro plan comes with half the usage it had.
My Take
DevDay had a lot of announcements but Dots being the headline just felt like old news even before it was announced. Heard mixed things on dots, my own experience was alright compared to regular ChatGPT interface. On stage there were several issues. One voice dot demo failed and had to be pivoted away from. Then Sam is on stage talking about reliability only to have the next demo fail with an error message. Apparently the system had just gone down an hour before. I’m left confused. If they knew it was down why continue with the demo? Overall an interesting DevDay showcase. Ultrafast seems cool but I guess you have to be in the token burner don’t worry about money club to enjoy it.
( OpenAI: DevDay 2026 recap / OpenAI: introducing GPT-6.1 Sol / OpenAI: introducing Dots / OpenAI Help Center: getting started with your dot / Artificial Analysis / Hacker News: GPT-6.1 Sol / Hacker News: Dots / OpenAI Help Center: about ChatGPT Pro tiers / The New Stack: the Pro 200 cut / @sama / @thsottiaux: the Pro 200 change / @agentnative_: the DevDay list )
-
Cloudflare’s Birthday Week: K2 ships, KV Instant opens Quicksilver, Clef takes on Jev, and Cloudflare applies to be a certificate authority Cloudflare’s Birthday Week ran to 46 posts. K2, the event stream we spotted unannounced last week, is official: a durable log on R2 in public beta for Workers Paid, Kafka-client support still on the roadmap. Workers KV Instant opens Quicksilver, the store behind nearly every Cloudflare request, with claimed p99 reads of 1.62 ms. Clef and Clef-flash are open-weight, Jev-compatible decision models; independent tests still favour Jev on speed and price. Cloudflare has also applied to become a public certificate authority, buying a GlobalSign root, and Artifacts, its Git storage for agents, is in open beta with a $25,000-credit contest.
My Take
Cloudflare is just shipping which is amazing to see. Event streams are pretty cool, and Quicksilver could be interesting. They seem to keep on shipping the "new" developer platform. Love to see it and I’m thinking about trying the Artifacts GitHub contest.
( Cloudflare Blog: Birthday Week wrap-up / Cloudflare Blog: K2 / Cloudflare Blog: Workers KV Instant / Cloudflare Blog: Clef / Cloudflare Blog: certificate authority / Cloudflare Blog: Artifacts and the Git contest / Cloudflare Docs: K2 limits / Cloudflare Docs: Workers AI pricing / SumGuy’s Ramblings: Clef vs Jev, 440 items / Hacker News: K2 / Hacker News: Clef / Hacker News: the CA / @ashleypeacock: day four roundup / @MattieTK: the CA )
-
Anthropic ships Sonnet 5.5 at Sonnet 5’s price, opens Claude Code to mods, and puts $100M into training deployed engineers Claude Sonnet 5.5 is the second model in the 5.5 family; Anthropic says it is 30%+ faster and up to 30% cheaper per task than Sonnet 5. Claude Code gained mods, TypeScript functions that run inside the process to rewrite prompts, block tool calls and gate permissions, shipped as plugins. And the Claude Frontier Academy will spend $100M training 10,000 “Frontier Deployed Engineers” by the end of 2027.
My Take
Not as major news from Anthropic this week. Claude Code mods seem pretty neat but I haven’t built any yet.
( Anthropic: Claude Sonnet 5.5 / Claude Platform docs: Sonnet 5.5 / Artificial Analysis / Claude Blog: Claude Code mods / Anthropic: Claude Frontier Academy / Hacker News: Sonnet 5.5 / @claudeai )
-
Two practitioners on working with agents: Yanacek’s part two, and Sridharan’s rule for debugging Two weeks ago his essay said the job was never about coding; this one lists what replaced it. Agent onboarding docs get updated daily instead of every few months, the deterministic quality gates he once hated now run the loop, every human code-review comment is treated as a defect in the auto-reviewers, and a local stack with simulated dependencies is mandatory. A part three is promised. Cindy Sridharan, separately, says agents debug well when held to two tiers: immutable facts gathered by tooling, which the model may analyse but never dispute, and everything inferred from them, which is where models are “most often wrong and most convincing” and so gets the scrutiny.
My Take
Yanacek - Sr. Principal with 15 years at Amazon talking about software factories. But beyond that just real detail on working in an agent world from a senior engineer. A great read. Cindy also posts that a big part of AI debugging is grounded in facts first. This is interesting 2 phased approach that I hadn't thought about.
( David Yanacek: so what would you say… you do here? / David Yanacek: it was never about coding (part one) / @dyanacek / @copyconstruct: the two-tier framework )