Model

Cognition launches Devin Bots, AI managers that built an Android app in 44 days

The $48 billion startup launched Devin Bots on November 17, AI managers for fleets of its Devin coders, staking its case that enterprises will pay for management above the models. It lands as investors offer $70 billion to $80 billion for a new round.

Vincent JiangVincent Jiangthree-minute read
Share
Scott Wu, center in glasses and a name badge, stands between two teammates in red shirts in front of a Welcome Participants banner at the 2016 ACM-ICPC World Finals
Scott Wu, center, at the 2016 ACM-ICPC World Finals in Phuket, Thailand, nine years before Devin Bots. He co-founded Cognition and is its chief executive.

One message, 44 days, an Android app

On September 28, Samyok, an engineer on Cognition's~$10B — Cognition, private, latest valuation ~$10B Bots team, gave a new "Android Bot" a single instruction: product manage the Android app, autonomously 1. Forty-four days later Cognition had an Android app in open testing on Google Play, with 1,134 merged pull requests, 344,651 lines of Kotlin, 2,841 screenshot tests and 1,217 tickets the engineer never filed 1.

He sent 183 messages over the run, about one for every six merges, and on 22 of the 44 days he sent none 1. That is the demo behind Devin Bots, launched November 17: persistent agents with memory that take a goal, break it into tickets and run teams of Devin coding agents in parallel, reporting in Slack 12.

The Bots have a mark to defend: 48 times revenue

Devin Bots are the first product evidence for a valuation that has climbed from $10.2 billion in September 2025 to $26 billion in May and $48 billion, where a $2 billion Series E priced it on September 8 3.

Cognition said on September 25 it had passed $1 billion in annualized revenue 3, which puts the last mark at roughly 48 times run rate, while TechCrunch, citing The Information, put possible cash burn this year at $800 million 5.

Business Insider reported on October 28 that Cognition has received offers at $70 billion to $80 billion and is seriously considering a round before year-end 4. The bet being priced: enterprises pay for the management layer above the models, not the models.

Cognition's valuation went from $10.2B to $48B in a year, and offers now reach $80B

  • Post-money valuation
  • Estimate
$0B$20B$40B$60B$80BSept 2025May 2026Sept 2026Oct 2026, talks$48BReported offers, not a pricedround
Data
Post-money valuation
Sept 2025$10.2B
May 2026$26B
Sept 2026$48B
Oct 2026, talks (estimate)$75B ($70–80B)
Post-money valuations announced by Cognition ($10.2B, September 2025; $26B, May 2026; $48B, September 2026) and the midpoint of the $70B to $80B offers Business Insider reported on October 28, 2026, shown as a range because no round has priced. Sources: Cognition blog; Business Insider.3,4

The receipts are all homegrown

Every number in the launch post is Cognition reporting on itself. The studio that went from about 50 to 1,900 weekly pull requests runs on roughly 20 Bots and one human, Jenny Xu, who then joined Cognition full time 1.

Outside evidence is thinner: usage at early-access customers including Decagon~$1.5B — Decagon, private, latest valuation ~$1.5B, Fireworks AI~$1.1B — Fireworks AI, private, latest valuation ~$1.1B and Reducto grew tenfold in two weeks, and Citigroup−2.18% — Citigroup, down 2.18 percent today told Fortune that 14,000 of its engineers use Devin 6. Those are usage counts, not audited outcomes. Meanwhile the labs that supply Devin's models shipped first: OpenAI's$1.18T — OpenAI, private, latest valuation $1.18T Dots on September 29, Anthropic's$350B — Anthropic, private, latest valuation $350B Duet on October 21 78.

Routing around the labs, at $80 to $200 a seat

Chief executive Scott Wu's answer is routing, a Bot picking models per task and leaning on cheaper ones Cognition trains in-house 6, on seats of $80 to $200 a month with extra usage billed at API rates 9.

One engineer sent 183 messages while the Bot filed 1,217 tickets and merged 1,134 PRs

01,0002,0003,000Screenshot tests2,841Tickets filed1,217Merged pull requests1,134Engineer messages183
Data
Value
Screenshot tests2,841
Tickets filed1,217
Merged pull requests1,134
Engineer messages183
Four counts from Cognition's own account of the 44-day Android Bot run, from September 28 to the app's open test on Google Play. Units differ (messages, pull requests, tickets, tests) and are not summed. Source: Devin blog, Introducing Devin Bots.1

Nobody has priced the supervision yet

The open question is the supervision bill. In a January 2025 test, Answer.AI found Devin failed 14 of 20 tasks and "couldn't discern any pattern" to predict the failures 7. A McKinsey survey of 334 product and engineering leaders in the second quarter of 2026 found 30% of organizations reporting declining productivity 7. Cognition announced a McKinsey alliance on November 16 to help enterprises prove AI's value 10. Wu concedes the control: "it's always up to the human where we're driving to" 2.

A smaller total bill is what would settle it

One number would settle the valuation debate: a public customer beating human cost per accepted change, review time included. The strongest customer claim so far, vice president of engineering Will Laughlin's report of about 1,600 Bots-merged pull requests in a week, sits on Cognition's own page 1. A quiet inbox is a demo; a smaller total bill is a business.

Get Model by email, free

The AI labs, their models, and the business behind them. In our daily email, every weekday.

Loading C

Reader comments

0 comments

    Model

    HeyGen's Avatar 6 scored 84% on realism, but its 42.2% pick rate was below chance

    Avatar 6 launched November 17, 2026, with an 84% realism figure and first place in a blind arena; the score doubles a below-chance pick rate, and the arena capped 97% of clips at 35 seconds. That is the evidence base under a company claiming $200 million of annual revenue.

    Model

    MiniMax hit a record low, about 90% below its January peak, as a US rule gates its rented Nvidia compute

    Washington's new remote-access rule puts a US license between China's listed model makers and the offshore Nvidia capacity they rent. MiniMax, the purest listed bet on Chinese foundation models, holds the price tag.

    Model

    OpenAI cuts flagship pricing 20% to match Anthropic as heavy users call GPT-6.2 Astra unusable

    OpenAI cut its flagship's price to match Claude Fable 5.2's rate, and the users who run models hardest rejected it, 24 hours before Anthropic priced a $1.4 trillion IPO on the assumption the quality gap holds. The independent labs say the gap is real, for now.

    Model

    Microsoft books Azure revenue on Chinese open models running on Baseten hardware it does not own

    Microsoft now bills customers' committed Azure dollars for open models from Chinese labs running on Baseten's hardware, a toll booth earning rent on compute it does not own.

    Model

    Alibaba stops giving away its best AI model, three days before earnings

    Alibaba launched Qwen4-Max on November 16 and kept the 2.4-trillion-parameter flagship behind its own paid API, promising open weights only for smaller models. Thursday's report is the first test of whether Qwen's three billion downloads become cloud revenue, and it lands two days after a US rule put the chips behind the model under license.

    Model

    Perplexity claims $1 billion of revenue on five times less spend than Anthropic's $1.4 trillion IPO

    Perplexity says it hit $1 billion in recurring revenue on a fraction of Anthropic's spend, three days before Anthropic prices a $1.4 trillion IPO on the opposite assumption. Whose economics hold will only be settled by a filing Srinivas now says could come in early 2027.

    Subscribe to TheInference for free