Roy Canani

@OpenRouter @OpenAI How you make those prices?

GPT-5.6 Terra and Luna just got price cuts from @OpenAI. OpenRouter is cheaper still. Our 50% exclusive discount applies on top, bringing Luna to $0.1/M in $0.6/M out and Terra to $1/M in and $6/M out. Scale your workloads with luna: openrouter.ai/openai/gpt-5.6…

@rauchg Would be interesting to hear about your agent harness, how to develop following best practices and fixing root cause and not just bandaging bugs

This will be the norm as software projects transition to agentic software factories. Issue → Agent → PR → Release 🔁 The job of the author / maintainer is to work on the loop that yields the highest quality product and sets the criteria for what should be worked on.

@github Amazing!

Took us a minute, but stacked PRs are now on GitHub 🥞 github.blog/changelog/2026…

Yes

@linear Why doing it on linear better than others?

New: Coding sessions on mobile Keep work moving from the Linear mobile app. Review code changes, comment on specific lines, and send feedback directly to the agent.

@kiaraplds Where does security come in mind in the process?

My current Cursor loop: Slack message → opens a Jira ticket → a Skill picks it up → Cloud Agent builds → PR lands. I set the automations up once. Now shipping from my phone (and ipad) is normal.

@ChmalSzymon To change the subject from outages to bad UI

Can somebody tell me why I'm seeing pancakes here? 😅

@theo Burning out is my stopping point of new threads 😅

Think I have enough threads going or should I spin up more?

@ClaudeDevs Reset soon I hope?

Two incidents over the last 24 hours resulted in elevated errors or reduced availability on Claude for some users. Multiple separate network failures cut into our capacity to serve Claude, and some requests failed while we rerouted traffic.

@NookaRules Amazing to see such fascinating results

When I built computer vision models, the worst failure mode was a confident misclassification - a muffin labeled as a chihuahua. 🧁🐕 In cybersecurity, hallucinations have a very different failure mode. We ran 162 experiments to understand what kind of context helps coding agents write secure code. Instead, we found agents that produced convincing plans, passing tests, and hardened-looking code - while still shipping the vulnerability underneath. Research: lnkd.in/dSQDS758 Presenting this at AppSec Village @ DEF CON 34. Come say hi 👋

Fantastic experiment, really exciting to see such impressive results!

Jinx now the third outage

@ClaudeDevs How two outages in one week not equal reset?

@theo As expected we don’t

Claude Code has been mostly down for half an hour. Really inconvenient, but I’ll forgive them if we get a reset

@thsottiaux You are great, but you invent even more than the Claude reset button.

One day we created the reset button and the rest is history.

@Prathkum But it’s created also new metrics

AI killed the illusion that being able to write code is what made you a good engineer.

@ClaudeDevs Isn’t it published a month ago?

MCP 2026-07-28 is live and it's the largest update to the protocol since launch. MCP is now stateless, making it easier to deploy and scale remote servers. claude.com/blog/bringing-…

@ThePrimeagen I think they kind of randomise which inference you are routed to.

It's crazy how much I dislike opus 5.

@exodus hi

We're following back anyone who replies to this tweet. Say hi! 👋

Why isn't it supported in the Claude CLI?