GenAI PMCognition
An AI coding company building models and tools for software engineering workflows. The newsletter notes SWE-1.6 became Windsurf's most used model.
Key Highlights
- Cognition is building both software engineering agents like Devin and specialized coding models such as SWE-1.6.
- Its recent launches emphasize real workflow integration across Slack, Linear, GitHub, shells, and Windows environments.
- Enterprise case studies point to measurable ROI in security remediation, documentation, testing, and modernization.
- SWE-1.6 became Windsurf’s most-used model, standing out for cost efficiency and reported speeds up to 950 tok/s.
- For AI PMs, Cognition is a strong example of how agent products win through integration, throughput, and task-specific outcomes.
Overview
Cognition is an AI company focused on software engineering workflows, best known in these newsletter mentions for building Devin and related infrastructure for coding, triage, testing, modernization, and enterprise development operations. Across recent updates, the company appears to be expanding in two parallel directions: agentic developer tooling for real production environments, and model development for software engineering use cases, including the launch of SWE-1.6.
For AI Product Managers, Cognition matters because it offers a concrete view of how AI coding systems are moving from demoware into operational systems. Its product updates span developer shells, Windows app modernization, enterprise security remediation, ticket triage, test generation, and high-speed coding models deployed in products like Windsurf. That combination makes Cognition useful as a case study in AI-native product design, enterprise adoption, workflow integration, and measuring ROI from software agents.
Key Developments
- 2026-04-21 — Cognition launched a parallelized Devin system in which a main Devin coordinates multiple full Devin sessions, each with its own VM, terminal, browser, and testing environment, to solve complex tasks concurrently.
- 2026-04-23 — Cognition launched Devin for the Rivian-Volkswagen joint platform, where it auto-triages Slack tickets and generates safety-critical propulsion tests up to 15× faster than manual authoring.
- 2026-04-26 — Cognition shared a discussion with Grant Sanderson about building frontier AI, highlighting internal workflows, collaboration patterns, and team culture.
- 2026-04-28 — Cognition released a custom terminal rendering library in Rust for fast UI performance, including support for a 1970s VT-100 terminal, and made it available at devin.ai/terminal.
- 2026-04-28 — Cognition also launched a CLI workflow to offload local tasks to its Devin cloud agent so work can continue after a developer closes their laptop.
- 2026-05-01 — Evinova, AstraZeneca’s health-tech arm, used Devin for regulatory documentation, bug triage, tech-stack migrations, and test automation, reportedly reducing regulatory document preparation time by about 8×.
- 2026-05-02 — Cognition embedded Devin directly in the shell via `devin shell setup`, enabling users to press Ctrl+G so Devin can inspect the workspace and provide immediate help.
- 2026-05-06 — Cognition reported that Devin automatically resolved 70% of Itaú’s backlog of SonarQube, Fortify, and Veracode vulnerabilities, positioning the agent for enterprise security remediation.
- 2026-05-19 — Cognition introduced Auto-Triage, connecting Devin to Slack, Linear, GitHub, schedules, webhooks, observability tools, tickets, alerts, and code to automatically begin investigations when issues appear.
- 2026-05-22 — Cognition launched Devin on Windows, aimed at mapping, building, and modernizing Windows applications.
- 2026-05-28 — Cognition launched SWE-1.6 as part of a new model training program; the newsletter notes it became Windsurf’s most-used model, with strong cost efficiency and speeds up to 950 tok/s.
Relevance to AI PMs
1. Workflow integration is the real product surface. Cognition’s updates show that value comes less from a standalone chatbot and more from connecting agents into systems like Slack, Linear, GitHub, shells, webhooks, tickets, and observability stacks. AI PMs should design around where work already happens.
2. Enterprise ROI needs task-specific proof points. Cognition’s reported wins are highly concrete: vulnerability backlog reduction, faster safety-critical test generation, shorter regulatory documentation cycles, and Windows modernization. PMs can use this as a template for identifying narrow, high-value workflows where agent performance can be measured clearly.
3. Model and agent strategy are converging. With Devin on one side and SWE-1.6 on the other, Cognition suggests a stack where proprietary or specialized software-engineering models power productized agents. AI PMs evaluating coding products should look at both the end-user workflow and the underlying model economics, latency, and throughput.
Related
- Devin — Cognition’s flagship software agent, central to many of the company’s launches across shell usage, Windows, enterprise triage, testing, and modernization.
- Windsurf / Windsurf Next / Windsurf 2.0 — Related as downstream product surfaces where Cognition’s SWE-1.6 achieved notable adoption.
- SWE-1.6 / swe-16 — Cognition’s software-engineering model, highlighted for speed and cost efficiency.
- Auto-Triage, bug-catcher, pull-requests, git-diffs, swe-check — Related capabilities and workflow concepts connected to AI-assisted software engineering operations.
- Slack, Linear, GitHub, Windows — Key integration and deployment environments showing where Cognition is embedding its agents.
- Rivian, Volkswagen, Itaú, AstraZeneca, Evinova, Infosys, Fortune 500 companies — Enterprise contexts relevant to Cognition’s adoption narrative, especially for modernization, compliance, security, and large-scale engineering operations.
- software-agents, ai-agents, inference-engines, async-rl — Broader technical categories connected to Cognition’s approach to agentic software engineering and model training.
- Russell J. Kaplan — Related person entity associated with Cognition.
Newsletter Mentions (27)
“Cognition launched SWE-1.6 as part of its new model training program; it’s now Windsurf’s most used model, praised for its cost efficiency and up to 950 tok/s speed.”
#18 𝕏 Cognition launched SWE-1.6 as part of its new model training program; it’s now Windsurf’s most used model, praised for its cost efficiency and up to 950 tok/s speed.
“Cognition launched Devin on Windows, letting PM Builders map, build, and modernize Windows applications.”
#15 𝕏 Cognition launched Devin on Windows, letting PM Builders map, build, and modernize Windows applications.
“Cognition ’s Auto-Triage connects Devin to Slack, Linear, GitHub, schedules, webhooks, observability tools, tickets, alerts, and code—automatically kicking off investigations whenever an issue pops up.”
#14 𝕏 Cognition ’s Auto-Triage connects Devin to Slack, Linear, GitHub, schedules, webhooks, observability tools, tickets, alerts, and code—automatically kicking off investigations whenever an issue pops up.
“Cognition : Devin automatically resolved 70% of Itaú’s backlog of SonarQube, Fortify, and Veracode vulnerabilities, delivering real results for enterprise security teams.”
#21 𝕏 Cognition : Devin automatically resolved 70% of Itaú’s backlog of SonarQube, Fortify, and Veracode vulnerabilities, delivering real results for enterprise security teams.
“Cognition embeds Devin right in your shell: press Ctrl+G to let it see your workspace and provide instant help—install via `devin shell setup`.”
Cognition embeds Devin right in your shell: press Ctrl+G to let it see your workspace and provide instant help—install via `devin shell setup`. NVIDIA AI introduces a speculative decoding technique in NeMo-RL with vLLM that removes RL post-training rollout bottlenecks, boosting throughput 1.8× on 8B models and projecting a 2.5× end-to-end speedup on 235B models.
“Cognition Evinova, AstraZeneca’s health-tech arm, leverages AI agent Devin for regulatory documentation, bug triage, tech-stack migrations, and test automation—cutting regulatory doc prep time ~8× down from the typical 35–40 hours.”
#7 𝕏 Cognition Evinova, AstraZeneca’s health-tech arm, leverages AI agent Devin for regulatory documentation, bug triage, tech-stack migrations, and test automation—cutting regulatory doc prep time ~8× down from the typical 35–40 hours.
“Cognition wrote a custom terminal rendering library in Rust for lightning-fast, snappy UI performance—complete with support on an original 1970s VT-100—and you can try it now at devin.ai/terminal.”
#4 𝕏 Cognition wrote a custom terminal rendering library in Rust for lightning-fast, snappy UI performance—complete with support on an original 1970s VT-100—and you can try it now at devin.ai/terminal. #8 𝕏 Cognition launched a CLI command to offload local tasks to its Devin cloud agent, so your work keeps running even after you close your laptop.
“Cognition sits down with Grant Sanderson (@3blue1brown) to unpack what it’s really like to build frontier AI—covering daily workflows, collaboration patterns, and a team culture likened to Olympic athletes training together.”
#8 𝕏 Cognition sits down with Grant Sanderson (@3blue1brown) to unpack what it’s really like to build frontier AI—covering daily workflows, collaboration patterns, and a team culture likened to Olympic athletes training together. #9 𝕏 Aravind Srinivas rolled out pay-as-you-go Computer access for Pro users at $20/mo. The $200/mo Max plan now includes a monthly allotment of Computer credits.
“#8 𝕏 Cognition launched Devin for the Rivian-Volkswagen joint platform, auto-triaging tickets in Slack and generating safety-critical propulsion tests up to 15× faster than manual authoring.”
#8 𝕏 Cognition launched Devin for the Rivian-Volkswagen joint platform, auto-triaging tickets in Slack and generating safety-critical propulsion tests up to 15× faster than manual authoring. It scales to power 30 M vehicles and frees RV Tech’s engineers to focus on new features.
“Cognition launched a parallelized Devin system: a main Devin spins up and coordinates multiple full Devin sessions—each with its own VM, terminal, browser, and testing environment—to break down and solve complex tasks concurrently.”
#6 𝕏 Cognition launched a parallelized Devin system: a main Devin spins up and coordinates multiple full Devin sessions—each with its own VM, terminal, browser, and testing environment—to break down and solve complex tasks concurrently. #7 𝕏 There's An AI For That links to Netflix’s new void-model GitHub repo, open-sourcing their text-to-video diffusion framework complete with pretrained weights, Colab demos, and inference scripts.
Related
An AI software engineering agent used for cloud-based automation and code changes. In the newsletter it’s used for scheduled automations, tests, and reviewing/merging code.
Autonomous or semi-autonomous software systems that can take actions, manage workflows, and assist with operational work. The newsletter references them in multiple founder and startup productivity contexts.
A project and ticket management tool used here as the system of record for agent workflows. PMs can use it to route tasks to coding agents and track review states.
A collaboration platform used as the interface for alerts and autonomous coding workflows. The newsletter mentions it both as an alert surface and as CrewAI Iris’s working environment.
GitHub is the company behind Copilot and the platform hosting related repositories and workflows. It is relevant here for plan changes and product packaging in AI coding.
A reimagined code review interface from Cognition that groups related changes and flags issues by confidence and severity. Useful as an example of AI-native developer workflow design.
A model released on Windsurf with a limited-time launch discount. It is relevant as another model option available to developers.
A product access offering mentioned in the context of pricing tiers and credits. It appears to be part of a broader AI product subscription structure.
The process of updating legacy COBOL systems, often for enterprise migration and maintenance. AI agents are increasingly positioned as tools to accelerate this high-friction modernization work.
A legacy programming language often targeted for modernization and migration efforts. For PMs, it represents enterprise technical debt and transformation risk.
A systems programming language used here as the implementation target for an AI-assisted rewrite of Bun.
Stay updated on Cognition
Get curated AI PM insights delivered daily — covering this and 1,000+ other sources.
Subscribe Free