Social Wire

Page 3 of 49 · newest first.

@cognition · Cognition
Introducing SWE-2 Flash, a fast and efficient coding model that outperforms larger American open-weight models
@cognition · Desktop
Streamline your workflow with Autofix, now available in Devin Cloud and Desktop, and save time on code reviews
@cognition · CLI
Get SWE-2 Flash for free until the end of the month with Devin Desktop and Devin CLI, and experience secure weight access
@cognition · Cloud
Introducing Devin Review 2.0, with up to 5x cheaper, 1.5x faster, and new features and integrations for improved code review
@cognition · Cloud
Devin Review 2.0 is here, offering faster, cheaper, and better performance - learn more about the upgrade and its benefits
@chaserobbinss Ambassador · Cognition
Chase Robbins after 14 days on t3 code: he is going back to DevinAI, which he calls light years ahead. He still lists t3 pluses — bring-your-own subscription, no GitHub app required, and a nice UI.
@cognition · Cognition
Cognition: Head of Security Products Jack Cable published A Practical Guide to Security in November 2026. He recaps the Binance $6.34B laptop-to-wallet-signing theft (fake job interview, tampered installer), the Oct 29 malicious Claude Code Cursor extension, PHP2Shell, GitLab RCE, eMudhra's key, a KVM escape, and Linux MPTCP CVE-2026-98432 — and tells teams to use phishing-resistant MFA, get secrets off laptops, move work to cloud coding agents / remote env, set a 7-day package age, and alert on high-signal events. https://cognition.com/blog/practical-guide-to-security-november-2026
@Da7_Tech Ambassador · Cloud
Da7_Tech ran a video render test clip on Devin in the cloud: under 2% of his weekly limit, Mac stayed cool, no throttling.
@glaucia_lemos86 Ambassador · Cloud
Glaucia Lemos is building her January workshop site with Devin on GLM-5.3 Max — she says it is cheap and comparable to Opus 5.5 Medium.
@kinopee_ai Ambassador · Cognition
kinopee on Cognition Fusion: Opus 5.5 as lead and GPT-6 Luna as sidekick scored higher than Claude Code + Opus 5.5 at 31% lower cost per task, in the example he cited.
@kinopee_ai Ambassador · Cognition
kinopee recommends Poke's free plan as a personal agent: no VM, but a sandbox for simple coding, with bigger development handed off to Devin.
@kinopee_ai Ambassador · Cognition
kinopee also uses Poke, which Cognition acquired: strongest at real-time notifications and good Japanese. He is curious how Cognition will combine it with Devin's VM tech.
@kinopee_ai Ambassador · Cognition
kinopee: the SWE-2 unlimited campaign through Dec 1 works in both Devin Desktop and Devin CLI — turning on Fast mode is billed.
@lgsreal Ambassador · Cognition
Luiz Real is mentoring at Hack Brasil (async via Discord) and building with tools like Devin by Cognition. He flags R$1M+ in prizes, free registration, teams of 1–4.
@Chris_Wozniczek Ambassador · Cloud
Chris Wozniczek's getting-started path: run Devin Cloud agents in a macOS virtual environment for iOS or macOS apps, and ask Devin to test and return a recording plus a test report. Treat it like an engineer on Slack; also try Desktop and CLI (cloud and local), SWE-2, and Fusion. His closer: you don't need every feature — tell Devin what you need and it will propose a setup.
@Chris_Wozniczek Ambassador · Cloud
Chris Wozniczek built an LLM-knowledge curriculum — tokenizers, transformers, KV cache, architectures, training — with Devin, including lab measurements on VMs. He published it at llm-curriculum.chris-wozniczek.dev (search plus DeepWiki) and asked Devin for a video demo.
@Da7_Tech Ambassador · Cognition
Da7_Tech's Devin vs Cursor note: more usage across more models on Devin, with SWE-2 free, plus cloud Mac, Linux, and Windows VMs. His stack is plan with Opus 5.5, build with SWE-2, verify with something cheaper like Sol 6.1 — mix the models instead of burning one quota.
@Da7_Tech Ambassador · Cloud
Da7_Tech: Devin VMs run Linux, Mac, and Windows with 32GB RAM (he contrasts that with Grok Bot's 16GB). He claims SWE-2 is about 11x cheaper per task once you count price, and that you pick the model — his comparison, not a Cognition figure.
@Chris_Wozniczek Ambassador · Cloud
Chris Wozniczek's Devin Cloud tip: after Devin builds an app, ask it to open a sub-session on a macOS VM to create and render a demo video — he says macOS rendering is much faster than the default Ubuntu VM.
@Da7_Tech Ambassador · Cloud
Da7_Tech posted a side-by-side video of GPT-6.1 Terra versus Claude Sonnet 5.5 running inside Devin.
@Da7_Tech Ambassador · Cloud
Da7_Tech says Gemini 4 Argon works well inside Devin, especially for writing, and traces a lot of Argon criticism — including his own — to the Gemini app and Antigravity rather than the model in Devin.
@gching Ambassador · Cognition
Gavin Ching's Vancouver BIOS SPHERE at Science World, powered by Cognition/Devin (Jake King, Orchid, keiki, Jory Braun) — luma.com/biossphere. His Nov 2 recap: "ai visibility and doing it right in Vancouver," thanks Cognition, Orchid/keiki, and jkudish.
@glaucia_lemos86 Ambassador · Cloud
Glaucia Lemos: Opus 5.5 on medium mode in Devin is more than enough for excellent results in her use.
@jensenloke Ambassador · Cognition
jensenloke's recap of Agentic Builders Collective meetup #11 (Oct 30): they shared their Devin workflow — GitHub issues, Microsoft Teams follow-ups, auto-created PRs, 300+ cloud sessions a week — plus a Qwen 3.8 Max walkthrough in Claude Code. Also Amp, Codex, and Grokbot from other speakers. Next: agenticbuilders.sg