Skip to content

Project

Devin

Autonomous software engineer product operated by Cognition.

Known aliases

  • Devin agents
  • Devin AI
  • Devin by Cognition
  • Devin CLI
  • Devin Desktop
  • Devin Web

Current stories

buildOne report1 publisher

Texas Tech study finds coding-agent revert rates range from 6.1% to 14.5% by vendor

Obada Kraishan of Texas Tech tracked 37,623 GitHub pull requests and found 90-day revert rates of 6.1% for OpenAI Codex and 14.5% for Devin. A pooled 'AI code' rate from that sample mostly measures Codex, so each tool needs its own numbers.

Publishers:dev.to

Reality

Evidence55
Adoption
Not scored
Hype gap+10
Incentives35
Confidence50
investOne report1 publisher

Goldman Sachs will put new hires in charge of AI agents from day one

Goldman Sachs will have new finance hires supervise AI agents from their first day, Asia Pacific chief Kevin Sneader said on October 8. The only reported productivity gain, more than 20%, is in engineering, so juniors checking agents in compliance and accounting is still an untested bet.

Reality

Evidence35
Adoption55
Hype gap+30
Incentives
Not scored
Confidence40
buildOne report1 publisher

Cognition moves the Dioxus team onto Devin's virtual machines and testing systems

Dioxus engineers already had code running inside Devin's command line, and they arrive with a cloud harness that snapshots and forks virtual machines on five operating systems. Neither side called it an acquisition.

Publishers:runtimewire.com

Reality

Evidence55
Adoption40
Hype gap+10
Incentives65
Confidence55
buildConfirmed2 publishers

OpenAI opens its 28-day Codex pledge by speeding GPT-6 Astra and Sol to about 50 tokens a second

OpenAI's Thibault Sottiaux says GPT-6 Astra and GPT-6.1 Sol now stream about 50 tokens a second, up from 30, as day one of a 28-day ship-or-reset pledge. The figure is an approximate streaming rate, so agent run times improve only as far as text generation dominates each loop.

Reality

Evidence35
Adoption50
Hype gap+15
Incentives70
Confidence60
buildOne report1 publisher

Steve Weis factors RSA-896 in ten days with Claude and idle GPUs

Anthropic engineer Steve Weis factored the 270-digit RSA-896 in ten days with Claude and up to 2,048 idle GPUs. Cognition estimates that factoring one 1024-bit number costs about $30 million, a sum a hyperscaler could spend.

Publishers:dev.to

Reality

Evidence55
Adoption
Not scored
Hype gap0
Incentives60
Confidence50
leadershipOne report1 publisher

Khosla calls portfolio company Factory 'second tier' in a fight over a board observer's move to Cognition

Vinod Khosla publicly sided with Cognition against Factory AI, which his firm helped fund in a $200 million round last month. For founders who share investors with a rival, the episode argues for writing observer access and investor conflicts into the contract.

Reality

Evidence55
Adoption
Not scored
Hype gap+5
Incentives75
Confidence55
buildConfirmed2 publishers

Cognition reports 4.8x token throughput on Nvidia Vera Rubin in its own coding-agent test

Cognition says its SWE-2 model produced up to 4.8 times more total token throughput on Nvidia's Vera Rubin NVL72 than on GB200, in a test on CoreWeave. The company-run result points to more agent capacity per system, but whether long coding tasks get cheaper depends on what the new systems cost per hour.

Reality

Evidence40
Adoption30
Hype gap+35
Incentives85
Confidence60
leadershipOne report1 publisher

Factory says its board observer spent weeks talking to Cognition before joining it as revenue chief

Factory CEO Matan Grinberg says he terminated board observer Chris Degnan for weeks of talks with Cognition, the $48 billion rival hiring him as revenue chief. Degnan and Cognition deny wrongdoing, so the fight turns on when an observer owes a board notice of talks with a rival.

Reality

Evidence40
Adoption
Not scored
Hype gap+30
Incentives80
Confidence40
buildOne report1 publisher

Cloudflare's Workers Issues hands grouped production errors to Claude Code, Cursor or Devin

Cloudflare's Workers Issues, now in open beta, groups production errors from one config line and sends them to Claude Code, Cursor or Devin. Detection and dispatch happen inside Cloudflare; the fix, and any gate before it ships, live in the agent's workflow.

Reality

Evidence45
Adoption10
Hype gap+15
Incentives85
Confidence50
buildOne report1 publisher

AI-assisted pull requests wait 4.6 times longer for a first review than unassisted ones

LinearB's 2026 benchmark finds only 32.7% of AI-assisted pull requests accepted within 30 days, against 84.4% of manual ones. The figures are associations, but they put review cost in the queue and in repeat rounds, where faster diff reading helps little.

Publishers:dev.to

Reality

Evidence45
Adoption
Not scored
Hype gap+10
Incentives40
Confidence40
investOne report1 publisher

Claude and Devin push the public factoring record 34 bits in 16 days

Two researchers moved the public RSA factoring record from 862 bits to 896 bits in sixteen days using the same classical sieve as 1990s attempts, and only the first of the two runs has a published compute bill.

Reality

Evidence20
Adoption12
Hype gap+55
Incentives62
Confidence28
buildOne report1 publisher

Martian scores code review bots on whether a developer actually changed the code

Code Review Bench reconstructs the timeline of 16,017 open source pull requests and publishes precision and recall beside every F1. The top five tools sit within five points of each other, on samples of very different size.

Publishers:dev.to

Reality

Evidence66
Adoption45
Hype gap+12
Incentives55
Confidence55
investOne report1 publisher

Three of the week's four billion-dollar rounds outweigh everything those companies raised before

Crunchbase counted four U.S. rounds of $1bn or more in one week. Tunneling, truck fleets and reusable rockets took $5.3bn of the $7.3bn, and the United Arab Emirates led the biggest check of the four.

Reality

Evidence58
Adoption18
Hype gap+14
Incentives62
Confidence60
buildOne report1 publisher

Cognition put a cost penalty inside SWE-2's reinforcement-learning objective

The model averages 53 steps a run against SWE-1.7's 127, and Cognition says its mean rollout cost is 64% below Fable 5.1's on a leaderboard Cognition built, runs and grades. Reproducing that takes Devin's harness.

Publishers:runtimewire.com

Reality

Evidence30
Adoption25
Hype gap+30
Incentives80
Confidence45

Earlier coverage

  1. Same price, cheaper fast mode: Opus 4.8 argues on unit economics

    Leadership · August 26, 2026 · One report1 publisher

  2. Slack Code makes coding agents taggable teammates. Your merge-approval policy is now overdue

    Product · August 20, 2026 · Confirmed4 publishers

  3. SpaceX went shopping for a second AI coding company five days after closing Cursor

    Build · August 19, 2026 · Confirmed2 publishers