The $26 Billion Coding Agent Verdict.

Mazen Al Sabbagh

— 6 min read
Share

TL;DR

  • Cognition AI raised over $1 billion at a $26 billion valuation, more than doubling its price tag in eight months, as its autonomous coding agent Devin grew from $37 million to $492 million in annualized revenue in a single year
  • Illinois passed the strongest AI safety bill in the U.S., requiring third-party audits of frontier AI labs — Governor Pritzker will sign it, setting a precedent that goes further than California or New York
  • The OpenAI Foundation committed $250 million to help workers and economies navigate AI-driven job displacement, its first major philanthropic initiative targeting automation's labor consequences
  • Amazon greenlit three AI-generated animated children's series under a new GenAI Creators' Fund, drawing immediate backlash from Hollywood unions and animation professionals
  • Dell reports Q1 earnings after the close today with a record $43 billion AI server backlog and expectations for roughly $13 billion in AI server revenue this quarter alone

Lead Story: Cognition Raised $1 Billion — and the Coding Agent Thesis Got Real

Cognition, the company behind autonomous AI software engineer Devin, raised more than $1 billion in a Series D round at a $26 billion post-money valuation, the company announced Wednesday. That is a 2.5x leap from the $10.2 billion it was worth when it closed a $400 million round in September 2025. Lux Capital, General Catalyst, and 8VC led the round, with participation from Founders Fund and Ribbit Capital.

The numbers behind the valuation tell the story. Cognition's annualized revenue run rate reached $492 million — up from $37 million a year ago, a 13x increase. The company told investors it intends to cross $1 billion in ARR later this year. Goldman Sachs, Mercedes-Benz, and the U.S. government are among its enterprise customers. The Next Web reported that 89% of Cognition's own codebase is now written by Devin — the product eating the company's own software engineering in real time.

The raise landed the same evening Salesforce reported first-quarter results that beat estimates on the top line but issued second-quarter guidance below Wall Street expectations. Salesforce stock, already down 33% in 2026, slipped further after hours. The juxtaposition is hard to miss: the company that built its business selling software to enterprises is losing investor confidence at the exact moment an autonomous coding agent more than doubled its valuation by promising to write the software itself.

In Other News

Illinois passed America's strongest AI safety law. The Illinois House voted 110-0 on Wednesday to send SB 315 to Governor J.B. Pritzker's desk, where he has confirmed he will sign it. The Artificial Intelligence Safety Measures Act is the first state law to require annual independent third-party audits of safety practices at frontier AI labs — targeting companies with $500 million or more in revenue, which catches OpenAI, Anthropic, and Google. Wired called it the most significant state-level AI regulation yet, going beyond California's compute-threshold approach and New York's disclosure requirements. The Chamber of Progress, which represents Google, Apple, and Amazon, lobbied against it. The White House also opposed it. That the bill passed unanimously despite coordinated industry resistance — in a federal vacuum left by Trump's postponed AI executive order — suggests state-level AI regulation has its own momentum now.

The OpenAI Foundation pledged $250 million for AI-displaced workers. The nonprofit that controls OpenAI announced Wednesday its first major philanthropic commitment: an initial $250 million for grants, research partnerships, and direct programs aimed at helping workers and economies adapt to AI automation. The timing was deliberate — it arrives as tech layoffs in 2026 have surpassed 114,000 (per Layoffs.fyi), with companies from Cloudflare to Intuit citing AI as a factor. Sam Altman has said he does not expect a "jobs apocalypse," but the foundation's own framing acknowledges that the disruption is real enough to require quarter-billion-dollar intervention. The IBTimes noted that the commitment is "initial," implying more funding could follow.

Amazon greenlit three AI-animated children's series under a new GenAI Creators' Fund. Amazon MGM Studios and AWS unveiled the fund Wednesday, ordering animated shows from former Nickelodeon executive Albie Hecht, animator Jorge Gutierrez, and BuzzFeed Studios. The productions will use AWS's "Project Nara" AI filmmaking tools across previsualization, asset generation, and production workflows. Variety reported that the fund provides both financing and access to AI production infrastructure. The move places Amazon alongside Netflix, which recently launched its own genAI animation pipeline. Animation unions and artists responded sharply — Cartoon Brew noted the projects land amid ongoing tensions between studios and creative workers over AI's role in production.

Cisco research found that multi-turn conversational attacks succeed against 88% of frontier AI models. The study, covered by Help Net Security, tested 15 flagship models from OpenAI, Anthropic, Google, and xAI using conversational escalation strategies that systematically break alignment guardrails. Unlike single-prompt jailbreaks, multi-turn attacks build context over several exchanges before extracting harmful outputs — making them harder to detect and filter.

X / Social Pulse

The Cognition raise dominated AI Twitter, with developers debating whether a $26 billion valuation for a coding agent signals a permanent shift in how software gets built or a repeat of the "GitHub Copilot will replace developers" cycle from 2022 — except this time with 13x revenue growth backing the claim. The Illinois bill generated a split reaction: safety advocates praised the third-party audit requirement as the template other states should follow, while industry-aligned accounts warned it would drive AI companies out of Illinois. The OpenAI Foundation's $250 million pledge drew skepticism from labor advocates who pointed out that OpenAI's own projected 2026 losses exceed $14 billion, asking whether a quarter-billion in philanthropy constitutes a serious response or a press release. Amazon's AI children's shows drew the sharpest emotions — animators posted side-by-side comparisons of GenAI output versus hand-drawn work, and several prominent creators publicly refused to participate.

One to Watch

Dell reports Q1 earnings after the close today — and it may be the clearest read on where enterprise AI spending actually stands. Analysts expect roughly $35.7 billion in revenue (up ~50% year-over-year) with $2.96 in EPS. The real number to watch is AI server revenue, expected around $13 billion this quarter, and whether Dell's $43 billion AI server backlog grew or shrank. Dell guided to $50 billion in AI revenue for fiscal 2027 and recently signed a $1.6 billion deal with IREN for Blackwell AI systems — one of the largest infrastructure builds outside hyperscaler data centers. The stock has surged 135% year-to-date. If Dell's AI margins hold and the backlog keeps growing, it confirms enterprise demand is real and accelerating. If margins compress or orders plateau, the question becomes whether $43 billion in backlog was the peak.

Quick Hits

  • Anthropic released a self-hosted Claude sandbox and security guidance plugin, adding to the 28 enterprise security integrations with CrowdStrike, Palo Alto Networks, Okta, and Wiz announced last week
  • Cerebras (CBRS) is trading as one of 2026's most closely watched AI chip IPOs, though the stock has pulled back from opening highs on customer concentration concerns tied to its OpenAI deal
  • Baseten is in talks to raise $1 billion at an $11 billion valuation, adding another inference-layer startup to a week that already includes Fireworks AI at $15 billion and OpenRouter at $1.3 billion
  • Mistral CEO Arthur Mensch publicly pushed back on Pope Leo XIV's AI encyclical, arguing that Europe cannot afford to slow down while adversaries accelerate — the first major lab executive to break with the Vatican's framing
  • Anthropic's Mythos model detected 23,000 potential vulnerabilities across 1,000 open-source projects, per SecurityWeek, an update from the 10,000 figure reported earlier in Project Glasswing's expansion

The thread running through today's news is how AI's consequences are becoming too concrete to ignore — and how different institutions are responding. Illinois decided to regulate. OpenAI decided to write a check. Amazon decided to greenlight AI-produced entertainment for children. Cognition decided the best proof of concept is having its AI write 89% of its own code. And Dell, which reports tonight, may offer the most useful signal of all: not what anyone hopes AI will become, but what enterprises are actually willing to pay for right now. The coding agent thesis, the safety audit thesis, the worker displacement thesis, and the AI entertainment thesis are all running simultaneously — and they are no longer hypothetical.

Sources

Read more