Social Wire
Community observations and coverage, newest first.
@SimiStern Ambassador · Desktop ·
Devin Desktop built a utility with ActivityWatch to track token usage and prompt quality across coding tools—agent-built observability for your own agent workflows.@Chris_Wozniczek Ambassador · Cloud ·
Review 2.0 auto-review workflow: enable always-on reviews, then merge when the bot clears it—automated quality gates without manual review queues.@bradshannon Ambassador · Cloud ·
Brad Shannon's reusable Devin Cloud prompt: point SWE-2 max subagents at the repo to spider scripts in parallel looking for implicit state machines, then come back with title, description, key files, and an optional doc. That's a Cloud subagent fan-out, not a single Desktop local run.@kinopee_ai Ambassador · CLI ·
Devin CLI now supports Agent Plugins, and Desktop picks them up through ACP—extending your agent's capabilities across both environments with shared plugin configuration.@willebrew Ambassador · Cloud ·
Cloud Agents simplify iOS development by reaching your Mac when needed—cloud orchestration with local build access in one workflow.@jensenloke Ambassador · CLI ·
CLI workflow idea: assess your repo and migrate the workflow up to Cloud Agents—turning local agent runs into orchestrated cloud sessions when the task scales.@Da7_Tech Ambassador · Cloud ·
Da7_Tech's hands-on in Devin Cloud: Kimi K3.1 Pro put up one of the most impressive results he's posted yet — a video of the run, not an availability tip. That's the model inside Devin's harness, not the Kimi chat app.@Da7_Tech Ambassador · Cloud ·
Da7_Tech: Kimi K3.1 Pro ran better on Devin Cloud than on Kimi's own sub (limits), Devin Max quota felt generous, and he's orchestrating multiple SWE-2 agents with Kimi as the conductor — Cloud multi-agent, not a single Desktop-local session.@devindevelopers · Cognition ·
Official @devindevelopers: moving from Conductor or another coding agent? Code DEVIN-TEAMS is 50% off Devin Teams for three months. The linked guide maps the other agent's rules into AGENTS.md and .agents/skills so Cloud and Local both pick them up — review and commit; don't copy secrets.@Da7_Tech Ambassador · Cognition ·
SWE-2 Flash runs at roughly 200 tokens per second inside Devin and stays free through November 30—fast inference for everyday agent work.@bnistordev Ambassador · Cloud ·
SWE-2 Flash fits weekly automation jobs: Devin finds issues, fixes them, tests the numbers, and waits for merge approval—cheap enough to run on a cron.@lgsreal Ambassador · Desktop ·
SWE-2 Flash is now available in Desktop and CLI: three times faster and five times cheaper than SWE-2 for local agent work.@kinopee_ai Ambassador · Cloud ·
Early hands-on with SWE-2 Flash on refactors and features: fast enough, solid accuracy, and roughly five times cheaper than SWE-2.@notjazii Ambassador · Cognition ·
SWE-2 Flash launch numbers: roughly three times faster and one-fifth the price of SWE-2, with Terminal Bench and FrontierCode scores reflecting its Inkling-Small foundation.@AtlasInference Ambassador · Cloud ·
Kernel developers eyeing SWE-2 Flash—low-level systems work gets a fast, affordable agent option with the same Devin harness.@ScarletKc Ambassador · Cognition ·
Ambassador deep dive on SWE-2 Flash's MoE architecture: 276B parameters with 12B active per token—trading some depth for speed and cost in everyday agent tasks.@therceman Ambassador · Cloud ·
Practical SWE-2 Flash ideas: spin up Workers for explore/build/test cycles, or run Board Agent monitoring—fast and cheap enough for everyday automation.@Fei2411 Ambassador · Cognition ·
SWE-2 Flash specs for Chinese-speaking developers: 276B parameters with 12B active, FrontierCode and DeepSWE benchmark scores, roughly 200 tokens per second in Devin.@notjazii Ambassador · Cloud ·
Early Flash tests against Tencent Hy4 show Flash notably faster on complex tasks—speed advantage holds when the work scales up.@kr0der Ambassador · Cloud ·
Six parallel Devin agents updated six iPhone Duo apps and secured App Store approvals in 24 hours—Cloud Agents handle iOS release velocity at scale.@itamarsharify Ambassador · Cloud ·
Itamar's Cloud split: Fusion for the develop pass, an OpenAI model for the plan, SWE-2 Flash for Git and small diffs, SWE-2 and Kimi when you're actually implementing. Same Devin Cloud session, different models for different jobs — not one model for the whole stack.@Chris_Wozniczek Ambassador · Cloud ·
Kimi K3.1 is now available in Devin—another model option for cloud and CLI workflows.@JustinGorya Ambassador · Cloud ·
Kimi k3.1 is available in Devin for testing—another model to evaluate inside the agent harness.@kinopee_ai Ambassador · Desktop ·
kinopee: Devin Local Hands (a Cloud session reaching your machine through Desktop) stops on an approval card before it runs a local command — read it, paste or confirm the risky ones yourself, then Allow once. That's the local safety gate; a Desktop-local session already on your box does not use Local Hands.