field archive · channel dogbed · est. oct 2026
An unauthorized, over-decorated, fully static record of Cognition — the lab that went from three IOI gold medalists in a San Francisco living room to the company behind Devin, Windsurf, and the SWE model line. Every page on this node was compiled by dogbed: knap template in, textile through the middle, HTML out. No CMS. No JavaScript. No edit button.
██████╗ ██████╗ ██████╗ ███╗ ██╗██╗████████╗██╗ ██████╗ ███╗ ██╗
██╔════╝██╔═══██╗██╔════╝ ████╗ ██║██║╚══██╔══╝██║██╔═══██╗████╗ ██║
██║ ██║ ██║██║ ███╗██╔██╗ ██║██║ ██║ ██║██║ ██║██╔██╗ ██║
██║ ██║ ██║██║ ██║██║╚██╗██║██║ ██║ ██║██║ ██║██║╚██╗██║
╚██████╗╚██████╔╝╚██████╔╝██║ ╚████║██║ ██║ ██║╚██████╔╝██║ ╚████║
╚═════╝ ╚═════╝ ╚═════╝ ╚═╝ ╚═══╝╚═╝ ╚═╝ ╚═╝ ╚═════╝ ╚═╝ ╚═══╝
>> the lab, the agent, the models — an exhibit in thirteen transmissions <<
This is a museum built out of plain text. Somewhere upstream there is a file called page.knap, a pile of JSON objects, and a single Zig binary named dogbed. The binary chewed the templates, spat out Textile, pushed it through oliver, and soldered the result into the document you are reading. The entire site — the scanlines, the ticker, the bezel — is one stylesheet and zero lines of JavaScript.
The subject matter deserves the treatment. In under three years, Cognition went from unveiling a demo of “the first AI software engineer” to training reinforcement-learning models at multi-trillion-parameter scale, rescuing an orphaned IDE in a weekend, and shipping a coding model that sits within a point of the frontier at a fraction of the cost. These are the annals of that run.
| transmission | contents |
|---|---|
| COGNITION | the company — founders, funding, the burn ledger |
| DEVIN | the AI software engineer, from viral demo to everywhere |
| WINDSURF | the weekend acquisition that doubled the company |
| PRODUCTS | the museum — DeepWiki, Kevin-32B, the API, the small machines |
| SWE-1 | the first model family — SWE-1, -lite, -mini |
| SWE-1.5 | the speed era — 950 tok/s on wafer-scale silicon |
| SWE-1.6 | the model-UX era — trained to be pleasant, not just smart |
| SWE-1.7 | the harness-native era — RL inside Devin itself |
| SWE-2 | the flagship — multi-trillion-parameter RL, effort levels |
| TIMELINE | the whole chronology, 2023 to the present |
| LEDGER | the operator’s burn ledger — tokens by model, day, project |
| COMMITS | every Devin-authored commit on this machine |
| 2023 | 13.86% | $40B+ | 950 tok/s |
| founded late | devin’s first swe-bench claim | reported valuation talks, aug 2026 | swe-1.5 peak serve rate |
Within our lifetime, engineers will go from bricklayers to architects, focusing on the creativity of designing systems — and delegating the construction to agents. — Scott Wu, CEO
All figures are as publicly reported by Cognition, press coverage, and product docs at the time of archival (October 2026). Benchmarks are vendor-reported unless noted. This node is an exhibit, not an offer to sell securities.
— END OF TRANSMISSION · CTR-(ok)VERIFIED% —
+++ COGNITION ANNALS +++ DEVIN · THE AI SOFTWARE ENGINEER +++ SWE-1 → SWE-1.5 → SWE-1.6 → SWE-1.7 → SWE-2 +++ WINDSURF FOREVER +++ COMPILED BY DOGBED, NOT WRITTEN BY HAND +++ NO JAVASCRIPT WAS HARMED IN THE MAKING OF THIS ARCHIVE +++
rendered by dogbed — a document compiler, not a CMS. one binary, one template, thirteen documents.