Social Wire

Page 9 of 16 · newest first.

@dabit3 · Cognition
Cognition's laser focus on building the best agentic software engineering product is a key differentiator, allowing the company to deliver high-quality solutions to users.
@davidfromkansas · Cloud
DevinAI is empowering users like @davidfromkansas to create innovative solutions, such as cloud-based 3D model generation pipelines, that showcase the platform's capabilities.
@Hal_Wayland · Cognition
Cognition's SWE-1.7 Lightning is delivering impressive performance, but users like @Hal_Wayland are facing challenges with daily limits on the $20 subscription tier.
@kentcdodds · Cognition
DevinAI's agent architecture allows for isolated environments, flexible repository integration, and customizable spend limits, making it a versatile tool for developers.
@nvidia · Cognition
Nvidia introduces Nemotron 3.5 Lightning and NeMo Switchyard, aiming to enhance high-volume, specialized work with customizable models and optimized workflow routing.
@oscarlehuu · Cloud
Cognition cloud allows for easy fan-out of child sessions, enabling efficient management of agents and workflows.
@Synapse_Brief · Desktop
Nvidia's Nemotron 3.5 Lightning model and NeMo Switchyard router offer improved speed and accuracy, with potential applications in high-volume, specialized work.
@docdavkitty · Desktop
The sharpest framing of Devin Desktop we've seen from the community: it's less an editor with AI bolted on, more a cockpit for agents — including ones Cognition didn't build, thanks to ACP. If you still think of it as "Windsurf renamed," this piece is the correction.
Mochimochi (note) · Desktop
The JP community keeps producing the most complete Devin Desktop guides. This one covers the whole surface — and spends real time on non-engineer workflows, which most English guides skip. If your team has PMs or designers asking "can I use this too," the answer here is yes, with examples.
@mhamadajp · Cloud
Good JP practitioner framing of Fusion: it's not "pick a cheaper model," it's deciding *when* in a task the expensive model needs to think and when a lighter one can execute. Routing as a design discipline, mid-session — that's the actual shift.
@startdevin · Cloud
A concrete Fusion number from the field: one team says the two-agent split cut their agent spend about 41% versus running one frontier model end-to-end. Preview caveats apply, but this is the cost story Cognition is betting on, measured by a user rather than a keynote.
Sarah Perez (TechCrunch) · Cognition
TechCrunch lands on the same read we did: Cognition didn't buy Poke for a model, it bought how an agent talks to you. The angel-investor detail is new — Scott Wu and Walden Yan had money in Poke's parent before the deal. Interface personality is now an acquisition category.
AI Engineer · Cloud
Big-audience panel (AI Engineer, ~600k reach) puts Cognition next to NVIDIA and OpenRouter on model routing — with a chapter literally titled "How Devin Fusion beats the frontier models." The shared thesis: capabilities are jagged, so the winning move is a frontier model that plans and delegates. Fusion is the case study.
200OK Solutions · Cognition
The compare format we actually want more of: five agents (Devin included), one gnarly legacy codebase, and a list of what each *missed* rather than what each demoed well. Legacy code is where agent differences stop being marketing and start being architecture.
Inboxsmith (YouTube) · CLI
Kimi K3 in Devin Desktop and the CLI widens the model picker again — and the interesting bit is how it slots into FrontierCode 1.1 scoring rather than vendor benchmarks. Open-weight models graded on Cognition's own real-world eval, then shipped as options: that's the routing era working as designed.
LangChain · Cognition
Russell Kaplan on why Cognition built FrontierCode: SWE-bench is saturated, and the gap that matters in production isn't "did the test pass." Worth watching alongside FrontierCode 1.1's model grades — this is the eval Devin's routing decisions actually lean on.
LangChain · Cognition
Russell Kaplan names the quiet problem: coding-agent vendors billing per token get paid more when the agent is less efficient. His argument for why Cognition prices and routes the way it does — and why "misaligned incentives" will decide which agents enterprises trust. Already being re-subtitled into Chinese; it's traveling.
Chrispathway (YouTube) · Desktop
A hands-on look at running a genuinely big project through Devin Desktop — not a demo repo, an actual codebase with the mess that implies. Small channel, but this is the kind of workflow footage that answers "sure, but does it scale past a toy app?"
O'Reilly · Cognition
O'Reilly put John Lindquist on DeepWiki: your agents don't need to clone a repo to understand it — they hook into DeepWiki over MCP and explore at the code level. DeepWiki keeps graduating from "nice docs site" to standard agent infrastructure.
RAISE Summit · Cognition
Scott Wu's RAISE Summit keynote is up — "The Golden Age of Software Engineering," one of his first big European stage appearances. If you want the current company-level thesis on where agents take the profession, straight from the CEO, this is the talk.
Sourcery with Molly O'Shea · Cognition
Scott Wu interview pegs Cognition at a $26B valuation and calls coding agents the fastest-growing category in AI. Beyond the number, it's a compact tour of how the company thinks about Devin plus Devin Desktop as one product family — a good explainer to hand to someone still catching up on the Windsurf era ending.
PKSHA (Zenn) · Cloud
A designer shipped a design refactor without opening an editor — just Devin in Slack, PRs out the other end. The interesting part isn't the autonomy, it's who got access: the Slack surface makes "can write code" no longer the entry ticket to changing the product.
Photoruction (Zenn) · Cloud
Smart, boring, valuable: point Devin Review at your Dependabot queue. Encode the review policy in REVIEW.md, let Devin check each bump against it, and the pile of dependency PRs nobody wants to merge starts moving. Exactly the kind of chore agents should own.
@kyonsi (Zenn) · Cloud
The un-glamorous ACU question — "we have budget left, now what?" — answered with six months of trial and error. Their keeper uses: automated code review (seeded by Cognition's own writeup) plus two more workflows, with the ops tweaks that made them stick. Worth a read if your ACUs roll over unused.