Social Wire

Page 29 of 49 · newest first.

@EvanDeKim · Cognition
SWE-2 model users are seeing impressive results, with one user reaching the 10b token mark and noting consistent p50 145 output tok/s
@fazedordecodigo Ambassador · Cloud
56 PRs shipped with Devin in one day hit both Devin Cloud quotas and GitHub Actions pipeline ceilings. The bottleneck shifted from dev throughput to CI capacity.
@Gunawardena_P · Cognition
Be aware of the costs: some users are finding that Devin's web app can burn through credits quickly, even with SWE-2
@robinebers · Desktop
Devin Desktop's cloud experience and Fusion model are winning over users, but the desktop app needs improvement
@silasalberti · Cloud
Free SWE-2 in Devin Cloud through Oct 8. Desktop and CLI demand hit unprecedented levels, prompting Cognition to scale up compute infrastructure.
@luongnv89 · Cognition
A user reports successful use of qwen3-coder-next for coding tasks, highlighting its efficiency, lack of usage limits, and privacy benefits as a local/self-hosted AI tool.
@cognition · Cognition
Cognition gave Devin a Ramp card and an open-ended goal: make money. Devin ran cold outreach, built payment portals, iterated on a business plan, and generated $75—showing both agent autonomy when given real-world tools and the challenge of open-ended optimization versus narrow task success.
@frmdomi Ambassador · Desktop
Devin Desktop's SWE-2 feature is getting positive feedback from users, who appreciate its capabilities in designing websites.
@Jfreee23 · Cognition
DevinAI is getting integrated into various projects, including a T3 Code fork, which may expand its user base.
@kr0der Ambassador · Cloud
Devin Cloud now includes Mac VMs, and with @rudrank's asc CLI tool, Devin can submit iOS apps directly to App Store Connect for review—a concrete new capability unlocked by the Mac Cloud infrastructure launch.
@luxus · Desktop
After experimenting with cloud agents, a user now sees why Cognition and Cursor AI are pushing this technology, citing increased productivity and efficiency.
@miguelg19877 · Desktop
A user finds SWE-2 capable of handling huge tasks with minimal revisions, but notes that Devin Desktop's performance, particularly scrolling, could be improved.
@luongnv89 · Cognition
The sleek-ui project has received a major update thanks to Devin from Cognition, now supporting customization of terminal environments like Pi, Ghostty, and Warp, and even web apps.
@muvonteam · Desktop
Devin Desktop has undergone rebranding and changes, including the introduction of quotas and a price increase for Pro, which may impact user preferences.
@SystematicEdge1 · Cognition
Devin's unlimited token usage and comprehensive goal mode are getting attention from users, who are also curious about the choice of Luna as a model for compacting.
@the_mobius_lab · Cloud
Devin and Qwen3.8 27B are being put to the test by users, who are finding the latter surprisingly capable for its size, picking up the slack in various tasks.
@AbdulTheBuilder · Cognition
Devin's platform allows seamless switching between accounts, making it easy to manage multiple projects
@cognition · Cognition
Cognition's Code Scans enable comprehensive, goal-oriented audits across entire codebases, leveraging Agentic MapReduce for efficient and accurate analysis and automated PR creation
@Da7_Tech Ambassador · Cloud
Da7 ran a detailed Fable 5.1 vs Astra vs SWE-2 shootout inside Cloud sessions, breaking down each model's strengths and failure modes—concrete routing data for teams choosing between speed (Astra), depth (SWE-2), or Fable's balance.
@DustinDavis · Cognition
MacOS users can now request support for cmd+v paste functionality in the TUI, enhancing their development experience with Cognition
@Fanfulladev · CLI
Developers see potential in integrating Devin with Omarchy, especially with SWE-2 available on the $20 Pro plan
@jaredpalmer · Cloud
DevinAI introduces Code Scans, a powerful tool for deeply auditing and reviewing entire codebases against specific goals, enhancing development efficiency and accuracy
@JustinHerman · Cognition
With Code Scans, teams can work at 'Frontier speed' with expert accuracy, ensuring coding standards are met and issues like dead code and untested flows are addressed efficiently
@kinopee_ai Ambassador · Cloud
Cloud quirk spotted: kinopee's Code Scan runs twice forced SWE-2 sessions into implementation (instead of Fusion), while normal sessions start Fusion without issue. If your scans skip Fusion unexpectedly, check whether the scan workflow is overriding your model picker.