Digital Colliers Daily Briefing — May 28, 2026
The AI coding agent thesis crystallized further on Wednesday, with capital, regulation, and revenue evidence all pointing the same direction. Cognition raised $1 billion at a $26 billion post-money valuation, validating the independent agent-lab model; Illinois passed the country's most stringent frontier AI safety bill, introducing third-party auditing of safety practices; and Simon Willison published a closely sourced analysis arguing that Anthropic and OpenAI have finally found product-market fit — with enterprise API pricing and a Cisco case study to back it up. Together, the three stories sketch a market that is simultaneously consolidating commercially and beginning to face structural oversight.
1. Cognition's $1B Round Validates the Independent Agent-Lab Thesis
Cognition, the maker of the Devin autonomous software engineer, has raised more than $1 billion at a $25 billion pre-money / $26 billion post-money valuation, according to TechCrunch. The round was led by Lux Capital, General Catalyst, and 8VC, with participation from Ribbit Capital, Atreides, Layer Global, Founders Fund, Elad Gil, and others. That marks a 2.5x markup from the $10.2 billion post-money valuation Cognition closed in September 2025 — just eight months ago. The company disclosed $492 million in annualized revenue run-rate, with enterprise usage of Devin growing 50% month-over-month for six consecutive months, and named Mercedes-Benz, NASA, Goldman Sachs, and Santander as customers. Latent Space's AINews recap reported Cognition is now projecting more than $1 billion ARR by year-end.
Why it matters. A year ago, the consensus assumption was that foundation-model labs would absorb the coding-agent layer themselves — a thesis reinforced by Anthropic's Claude Code, OpenAI's Codex, and Google's acqui-hire of Windsurf. Cognition's round is a direct bet against that absorption thesis. Latent Space frames Cognition as "officially the largest remaining independent agent lab in AI," and the ARR curve suggests harness, workflow, and enterprise integration can be defensible even when the underlying models are commodity inputs from competitors.
Who is affected. Enterprise buyers gain a credible independent vendor at scale; Cursor, Replit, and smaller agent startups now face a clearer pricing and hiring benchmark; and model vendors must decide whether to compete directly with their largest API customers or accept the middleware layer.
What to watch next. Whether Cognition's run-rate actually clears $1B by EOY, whether more enterprise logos surface as references, and whether Anthropic or OpenAI escalate their own coding agents to compete head-on with Devin on long-horizon tasks.
Sources:
2. Illinois SB 315 Mandates Third-Party Audits of Frontier Labs
The Illinois House passed SB 315 on Wednesday, sending to Governor JB Pritzker — who has said he will sign — a bill that requires frontier AI developers including OpenAI, Anthropic, and Google DeepMind to submit their safety practices to independent third-party audit. Wired reports that California and New York currently set the high-water mark with transparency and incident-reporting requirements; Illinois extends that baseline by introducing external verification that labs are actually adhering to the safety commitments they publish. Scott Wisor of the Secure AI Project, which supports the bill, told Wired that audits would likely be performed by the Big Four accounting firms (Deloitte, EY, KPMG, PwC) or by members of the AI Evaluator Forum, which includes METR, Transluce, and Averi.
The politics are notable. Anthropic claims it was the first frontier lab to back SB 315; OpenAI, after walking back earlier support for a separate Illinois bill containing a liability shield, endorsed SB 315 as well — consistent with OpenAI policy chief Chris Lehane's recent framing that the company is now organizing its policy effort around passing similar state laws. Chamber of Progress, a trade group backed by Google, Apple, Amazon, and Andreessen Horowitz, opposed the bill, arguing it would expose sensitive systems to "untested auditors." Separately, The Verge reports that OpenAI-funded super PAC Leading the Future has spent millions targeting New York assemblyman Alex Bores in the NY-12 Democratic primary, raising his profile as an AI-safety advocate in the process.
Why it matters. "AI companies grade their own homework," as Wisor put it; SB 315 ends that arrangement at the state level. Bill sponsor Rep. Daniel Didech explicitly framed Illinois as a "testing ground" for federal legislation — a notable posture given that the Trump administration has spent the year unwinding federal AI rules to avoid a "patchwork."
Who is affected. OpenAI, Anthropic, and Google DeepMind face direct compliance obligations once the law takes effect; the Big Four audit practices and the smaller AI Evaluator Forum nonprofits stand to absorb a new line of work; and other state legislatures gain a model text to copy.
What to watch next. Pritzker's signing timeline, the rulemaking process defining what an "audit" actually entails, and whether the federal executive branch tries to preempt state-level AI safety rules.
Sources:
3. Coding Agents Hit Product-Market Fit as Enterprise Pricing Shifts to API Rates
Simon Willison published an analysis on Wednesday arguing that Anthropic and OpenAI have, at last, found product-market fit — and that April 2026 marks the inflection point at which it becomes visible in revenue. The mechanism: both labs have quietly migrated enterprise pricing away from heavily-discounted seat plans to API-token-metered billing. Per The Information, Anthropic's enterprise plan moved to $20/seat/month plus API pricing in November 2025; OpenAI's Codex rate card was updated April 2, 2026 and extended to all existing ChatGPT Enterprise plans on April 23. Both labs also released costlier frontier models in the same window — GPT-5.5 at 2x the API price of GPT-5.4, and Opus 4.7 roughly 1.4x Opus 4.6 after tokenizer adjustments.
The financial picture has shifted accordingly. Anthropic is rumored to post its first profitable quarter on roughly $10.9 billion in Q2 revenue, up from $4 billion total in August 2025. SpaceX's recent S-1 disclosed a Cloud Services Agreement under which Anthropic will pay $1.25 billion per month through May 2029 for COLOSSUS and COLOSSUS II inference capacity — a single-vendor line item that hints at the scale of underlying demand. Willison also notes that 32.6% of OpenAI's 703 open jobs and 26.9% of Anthropic's 390 are enterprise sales and field-engineering roles.
OpenAI's Cisco case study, published the same day, makes the demand side concrete: 95%+ of new AI features in Cisco's AI Defense product are now written by Codex; a CodeWatch defect-remediation workflow on C/C++ codebases delivers 10–15x throughput; and cross-repo build optimization saves more than 1,500 engineering hours per month across 15+ interconnected repositories. A companion post on Thrive Holdings' Tax AI describes a Codex-driven self-improvement loop that lifted return accuracy from 25% to 86% reaching 75% field completion in six weeks across 7,000 returns.
Why it matters. As Willison observes, charging $10–$20/month to consumer subscribers cannot amortize trillion-dollar infrastructure commitments; charging $200–$1,000+ per professional user per month plausibly can. The pivot also has competitive consequences: Anthropic's Claude Code directly competes with its largest API customers, Cursor and GitHub Copilot — which last year accounted for roughly $1.2 billion of Anthropic's then-$4 billion revenue. Cursor is now investing in its own models in response.
Who is affected. Large enterprises signing annual contracts are absorbing materially higher AI line items — the Uber and Microsoft cost stories Willison addresses are early symptoms. Middleware vendors (Cursor, Copilot, smaller agent startups including Cognition) face margin compression as model labs move downstream. Cloud providers building inference capacity stand to benefit on the scale of Anthropic's SpaceX deal.
What to watch next. The Anthropic and OpenAI S-1 filings, which will convert rumored figures into audited ones; whether the Q3 enterprise renewal cycle produces a wave of budget-shock stories; and whether Cursor's in-house models meaningfully blunt Claude Code's encroachment.
Sources:
- I think Anthropic and OpenAI have found product-market fit — Hacker News
- Anthropic and OpenAI seem to have finally found product-market fit with coding agents, which are quickly becoming daily drivers for highly paid professionals (Simon Willison/Simon Willison's Weblog) — Techmeme
- I think Anthropic and OpenAI have found product-market fit — Simon Willison
- Cisco and OpenAI redefine enterprise engineering with Codex — OpenAI Blog
- Building self-improving tax agents with Codex — OpenAI Blog
The three stories interlock. Willison's revenue evidence explains why Cognition can credibly project >$1B ARR by year-end and why VCs underwrote a $26 billion valuation — coding agents are now metered, sticky, and expensed by procurement rather than experimented with by individual developers. The same scale that produces those revenue curves is what made Illinois SB 315 politically viable: frontier labs are now consequential enough that auditing them looks less like premature regulation and more like overdue corporate governance. The open question for the second half of 2026 is whether the audit regime arrives in time to shape the IPO disclosures that will, for the first time, put real numbers behind today's rumored ones.