Social Wire

Page 2 of 49 · newest first.

@Chris_Wozniczek Ambassador · Cloud
Cloud Agents as LLM training orchestration: Devin manages VPS provisioning, GPU rentals, Harbor registries, and spend caps—turning training ops into an agent-driven workflow.
@Chris_Wozniczek Ambassador · Cloud
One cloud session running since October 25, orchestrating tools, GitHub, Slack, and external VPS plus GPU infrastructure—proving Cloud Agents scale for persistent multi-service workflows.
@amirmxt Ambassador · Desktop
PSA: the Devin Mac app has improved significantly—worth revisiting if you tested an earlier version and hit friction.
@MatheusPagani77 Ambassador · Cognition
Hack Brasil participants: Cognition's $20 coupon toward a paid Devin Pro plan this month is in the event panel — not a public code — plus unlimited SWE on Desktop and CLI for the online phase, and $20k in credits for winners.
@kinopee_ai Ambassador · Cognition
Devin's notification system got a quiet upgrade, and early Mobile access users are impressed—solid groundwork before the public mobile launch.
@Da7_Tech Ambassador · Cognition
Devin mobile app demo shows sessions, agent machine control, and terminal steering straight from your phone—full agent access wherever you are.
@notjazii Ambassador · Cloud
notjazii ran the same high-reasoning prompt in Devin Cloud: SWE-2 (K3-based) ~25 minutes at $0.0 on the free window; Inkling 2 Small ~20 minutes at $0.5 and underperformed his SWE-2 expectation. One timed run, not a leaderboard — harness plus quota still decide what you feel.
@kinopee_ai Ambassador · Cloud
Cloud Outposts in action: Devin connects from a cloud session straight to your MacBook to debug from a screenshot, then opens a fix PR—no manual handoff, just seamless cloud-to-local autonomy.
@kinopee_ai Ambassador · Cloud
Outposts lets Devin Cloud reach external VMs and devices—turning cloud agents into a hub that orchestrates your local machines when the task calls for it.
@Da7_Tech Ambassador · Cloud
Review 2.0 tackles the PR review bottleneck: one-fifth the cost, catches more bugs, and runs 1.5× faster—no more queue pileups.
@kr0der Ambassador · Cloud
Review 2.0 runs $0.25–$0.75 per PR on average with a 94-second median latency—affordable enough to review every PR without the old speed and cost bottleneck.
@notjazii Ambassador · Cloud
Review 2.0 in practice: five times cheaper, faster, catches issues before merge, and can fix them—turning review from bottleneck to quality loop.
@JustinGorya Ambassador · Cloud
Review 2.0's five-times-cheaper, faster reviews mean you can lean into PR-driven workflows when running multiple agents—quality gates without cost spirals.
@VictorMotricala Ambassador · Cloud
Review 2.0 is 82% cheaper and 1.5× faster—but set a per-PR spend limit before turning it on for high-volume repos, or costs can still add up.
@bnistordev Ambassador · Cloud
Review 2.0 at $0.25–$0.75 per PR and 1.5× faster means you can pair it with Devin to surface only the decisions you need to make—automation handles the rest.
@kinopee_ai Ambassador · Cloud
Review 2.0 is live in the Web UI with per-repo settings and a new Ultra inference mode—higher cost, deeper reasoning for PRs that need it.
@ScarletKc Ambassador · Cloud
ScarletKc on Opus 5.5: inside Devin Cloud it felt like the raw API — sharp. Same model in Claude Code Desktop felt dumber to him; the Claude Code CLI a bit smarter. His point isn't the weights — it's system prompt, tools, and harness. Use them together: ACP / Devin CLI can sit next to Claude Code when you want Devin's harness on the same box.
@Chris_Wozniczek Ambassador · Cloud
Devin Secrets, automation cards, and GPU monitoring tie together for training workflows—cloud agents handle credentials, triggers, and real-time resource tracking in one layer.
@JoePro Ambassador · Cloud
Ambassador shipping real work: Devin reworked a Space copilot overnight and opened the PR—production agent output, not demos.
@therceman Ambassador · CLI
Terra Medium performs well in Devin CLI but struggles in Codex—a reminder that model scores reflect the whole agent harness, not just the LLM.
@Chris_Wozniczek Ambassador · Desktop
SWE-2 remains free through the end of November in Desktop local agents and CLI—full access to the flagship model for another few weeks.
@therceman Ambassador · Cloud
After Argon, therceman put SWE-2 Flash Max in the Devin Cloud lead seat — it has opinions, it argues, and it's fast. Argon he kept for delegating and talking inside Devin. Same account, different roles: the harness plus where you seat the model changes how the agent behaves.
@therceman Ambassador · Desktop
therceman's multi-PC Desktop setup: Outpost on every machine as the control center, SWE-2 Flash Max as the lead (still free through November), Devin Secrets PIN-gating critical actions, and he's wiring browser/agent automation on top. That's Outposts + Secrets on Desktop, not a Cloud-only stack.
@cognition · Cognition
SWE-2 Flash offers 5x lower cost and 3x higher speed than SWE-2, thanks to its optimized inference stack and smaller size