Social Wire

Page 4 of 49 · newest first.

@jumperz Ambassador · Desktop
jumperz's $20 stack this month: Devin Pro first because SWE-2 is free on the $20 plan in Devin Desktop and CLI and does not touch quota, with Claude, GPT, and Gemini still available. Claude Pro sits behind it as the $20 "brain" he uses to write handoff files and guide other models — Devin for the coding loop, Claude for the handoff.
@kinopee_ai Ambassador · Cloud
kinopee: Devin Automations can launch child sessions and target macOS and Windows VMs — a combination he calls unique.
@kinopee_ai Ambassador · Cloud
kinopee, wiring bots: Devin's API polls session state, so you write a watcher routine; Cursor Cloud Agents' API streams, so you can go event-driven without polling. Different integration shapes for the same class of cloud agent.
@kinopee_ai Ambassador · Cloud
kinopee published the "Devin man" Grok Bot template: it hands a task to Devin through a PR merge, watches the session, handles CI failures and review comments, and merges only after owner OK. Import steps in Japanese and English: https://devin-man-import-guide-dvad.pages.grok.bot/
@kinopee_ai Ambassador · Cloud
kinopee calls Devin from Grok Bot instead of Cursor Cloud Agents when the work is native macOS or Windows app development — pick the tool for the job.
@notjazii Ambassador · Cloud
notjazii's zero-shot comparison: same prompt, max reasoning — he ran GPT-6 Astra in Devin; a friend ran GPT-6.2 Astra in Codex. He says the newer Astra looks better on this prompt; more prompts still to come. Two models in two harnesses, not a single-agent bake-off.
@kinopee_ai Ambassador · Cloud
kinopee: Devin's `gh` is a safety wrapper that checks the base branch with real gh before merge. On Windows VM sessions without a real gh/token, merges stop at "Merging is blocked"; Linux sessions can merge with owner approval. He says merging VM-worked PRs yourself is faster.
@realmtbman Ambassador · Cloud
Yohann Calpu built a 16 MB browser downhill MTB game in four days with Devin — Three.js, a 900 m trail, ghost racing, night and rain. Play it at https://devkick.com/downhill/ (open source; 3D via Meshy, audio via ElevenLabs).
@ScarletKc Ambassador · Cloud
ScarletKc tried Muse Wonder (standard, not Full) inside Devin. Coding: average, roughly DeepSeek V4.1 Flash 1017 tier in his hands — fine for simple tasks, strained on harder ones. Chat and text felt more natural. His review, not a benchmark.
@taiyo_ai_gakuse Ambassador · Cloud
Taiyo Kimura expects Grok Bot power users to try Devin: better VM performance than Grok Bot, and a ChatGPT subscription can be used with it. He also flags a steep learning curve until it clicks.
@cognition · Cognition
RT @ArtificialAnlys: Announcing Artificial Analysis Intelligence Index v5, featuring 2 new benchmarks and 3 upgrades. For the first time, t… Worth tracking on the Devin Central desk.
@cognition · Cognition
RT @ArtificialAnlys: FrontierCode 2.0 is the latest update of @cognition's private coding benchmark. Launched earlier today, it comprises 1… Worth tracking on the Devin Central desk.
@cognition · Cognition
Cognition is committing hundreds of millions of dollars in Devin credits to provide free deep scans with frontier cyber models for enterprise customers and remediation credits for new customers.
@cognition · Cognition
Learn more about Devin Security and how it works, with a link to a blog post that provides an in-depth introduction to the technology.
@cognition · Cognition
Each task on FrontierCode 2.0 was built by a maintainer of the open-source repo it comes from, with 40+ hours of work per task, and audits led to improved reference solutions and even discovered a security vulnerability.
@cognition · Cognition
FrontierCode 2.0 provides a clearer separation of model tiers than its predecessor, with some models ranking differently than expected, such as Muse Wonder 1.0 and Gemini 4 Argon.
@cognition · Cognition
Explore an example task from FrontierCode 2.0, where Claude Opus 5.5 successfully fixes a user-reported bug in the Mock Service Worker JavaScript library.
@cognition · Cognition
The price vs performance evaluation on FrontierCode 2.0 reveals clusters of models, with Opus 5.5 and GPT-6 Astra leading in intelligence, and GPT-6.1 Sol leading in cost efficiency.
@cognition · Cognition
Learn more about the methodology and scoring criteria behind FrontierCode 2.0, Cognition's private coding benchmark, and explore sample tasks on their blog.
@cognition · Cognition
Check out the full leaderboard for FrontierCode 2.0, Cognition's private coding benchmark, to see how different models stack up against each other.
@cognition · Cognition
Cognition announces the acquisition of Corridor, emphasizing the importance of securing the world's software as a top priority in technology today.
@bnistordev Ambassador · Cognition
bnistordev: Devin Security upgrade Monday, Corridor today — the pitch is catching vulnerabilities while the agent is still writing the code, before there is a PR to review. Inside Devin that is no longer a separate step.
@Chris_Wozniczek Ambassador · Cognition
Chris Wozniczek: Corridor is Cognition making the right market move — security for agents will get hotter, and he expects others to follow in a few months.
@Chris_Wozniczek Ambassador · Cloud
Chris Wozniczek: his first cloud-agent use was Devin, and he has been posting about cloud agents since July — the unlock is out-of-the-box (security, macOS VMs, review, testing) so you skip setup and just run tasks.