Three AI Stocks the Market Has Already Marked Down

BlackLine stock closed at $30.14 on September 8, down 44.8% over 12 months, while AI-stock screens still flag it as an automation play. Pagaya Technologies closed at $22.08 the same day, down 41.86% over that span, and Docusign closed at $65.08, down 20.4%.

Anthropic’s economics team published an interactive model on September 9 that puts dollar figures on how far algorithms could push growth, jobs, and the split between wages and capital. That is the map these three software names are being read against, and it is a harsher map than the buyer screens imply.

Anthropic Mapped the Job These Three Stocks Sell

The model treats the U.S. economy as a pile of tasks. Software can leave a task alone, help a person do it faster, take the task over, or create a new one. BlackLine sells that substitution inside the close. Pagaya sells it inside credit files. Docusign sells it inside the contract.

Anthropic’s team, including economists Anton Korinek and Chad Jones, published three scenarios for U.S. GDP through 2030 at 2025 prices and attached no probabilities to them. A companion survey of 10,980 U.S. adults in August produced a typical answer close to the middle path: output about 10% higher than a no-AI path, and unemployment around 5%. About 10% of those respondents lined up with the extreme case.

ANTHROPIC’S 2030 CASES

In the modest case, AI looks like the internet: real, and easy to miss in the macro data. In the substantial case, the model can do half of knowledge work by 2030, most of it on its own, and the economy grows at twice its normal rate while knowledge-worker pay stays flat. In the extreme case, annual GDP growth hits 15%, the economy doubles every 4.5 years, and knowledge-worker wages fall more than 10% by 2030.

Today about 60 cents of each dollar of output goes to workers and 40 cents to capital. Even the middle case lifts capital’s share 3.9 points, to 43.9%. The extreme case flips the split, sending 54.8% to capital. Owners of models, data centers, and software sit on that side of the ledger. The open question for these three tickers is how much of that capital share they keep after the labs take their cut.

BlackLine’s Agents Spread Faster Than Its Contracts

BlackLine, the Los Angeles firm that automates record-to-report and invoice-to-cash work for the office of the CFO, is the cleanest example of the squeeze. Annual recurring revenue was $719 million in the second quarter, up 6%. GAAP revenue was $187.8 million, up 9.2%. The company kept its full-year GAAP revenue range at $765 million to $769 million.

Those are solid software prints. They are not the multiples the AI label once bought. The shares’ $1.75 billion market cap sits against a 52-week range of $24.70 to $59.57. Trailing-twelve-month revenue is $732.45 million.

I believe the first half of 2026 will prove to be the most consequential period in BlackLine’s history. AI is reshaping the Office of the CFO, and we are meeting that moment.

Owen Ryan, CEO, August 4 earnings release

Ryan also said deal timing was noisy because customers were working through more rigorous, AI-driven evaluations. On the August 4 call he told analysts that about $8 million of expected opportunities slipped past June 30 as buyers dug into AI governance, the product road map, and how BlackLine sits in their control environment, and that roughly half of that volume had closed by early August. He put the extra time at another 40 to 45 days on cycles that already run nine months to a year in the enterprise tier.

The product is not sitting still. Verity Prepare, the reconciliations agent, reached general availability in July. Platform-pricing ARR was 17% of eligible ARR on June 30, with a dollar-based net revenue retention rate of 102% and 4,260 customers. Remaining performance obligations rose 16.8% to $1.1 billion, which is the number that says the backlog is healthier than the quarter’s 5.9% billings growth. Average deal size was up 24% year over year, and 56% of the renewal book carried multiyear terms, up from 45% a year earlier.

The second-quarter results and full-year outlook still assume that mix shift pays. Third-quarter GAAP revenue is guided at $193 million to $195 million. Non-GAAP operating margin for the year is guided at 24.1% to 24.6%. The company bought back about 1.2 million shares for $37.7 million in the quarter and added $100 million to the authorization, with $179.7 million of capacity left on June 30. Usage can double and still leave revenue looking like a mid-single-digit software company if every extra review adds six weeks.

$3.5 Billion in Loans, a Thinner Cut for Pagaya

Pagaya Technologies plugs its models into banks and lenders and takes a fee when those partners fund loans they would otherwise decline. In the second quarter it put up a record $3.5 billion in network volume, up 33% from a year earlier, and total revenue and other income of $387 million, up 19% from $326 million. Auto supplied more than 75% of the volume growth. Application-to-volume conversion stayed near 1% on a network that has evaluated more than $4.0 trillion of applications since inception and helped generate more than $45 billion of new credit.

PAGAYA’S SECOND-QUARTER PRINT

• The take rate: Fee revenue less production costs was $147 million, or 4.2% of network volume, down 61 basis points from a year earlier.
• The profit: GAAP net income was $45 million, up $29 million year over year, the sixth straight profitable quarter, with the net margin at 12% against 5%.
• The engine: Adjusted EBITDA was $124 million, a 32% margin, up 43%, while core operating expenses fell 6%.
• The book: Full-year volume is now guided at $12.5 billion to $13.25 billion, with revenue and other income of $1.425 billion to $1.525 billion and GAAP net income of $155 million to $180 million.

That 4.2% slice is the whole argument. Volume can roar, as it did, and the fee on each dollar still shrinks when partner mix shifts and the cost of capital stays high. Losses on investments in loans and securities were $42.3 million in the quarter, with a $38.0 million impairment on certain investments, which is a reminder that Pagaya still sits next to credit risk rather than above it.

Funding held. The company completed $3.7 billion of asset-backed securities across six deals, held $249 million of cash and $1.04 billion of loan and securities investments on June 30, and counted 174 institutional investors against more than 35 lending partners. Fitch revised the corporate outlook to Positive. The shares still closed September 8 at a $1.84 billion capitalization, inside a 52-week range of $10.40 to $44.99.

If a general model can score a borrower from the same bureau file a bank already owns, Pagaya’s edge has to be the network, the funding shelf, and the conversion engine, not the fact of machine underwriting. Agentic credit that prices risk at machine speed would not need a specialist to sit between the lender and the applicant, and the fee would be the first thing to go.

How Much of Docusign’s Growth Is Still a Signature?

Docusign is the largest of the three, with a $12.2 billion market cap at the September 8 close and more than 1.9 million customers. For the quarter ended July 31, revenue was $875.7 million, up 9%, including a 1.3% lift from foreign exchange. Free cash flow was $295.8 million, a 34% margin, against $217.6 million and a 27% margin a year earlier. Cash, cash equivalents, and investments were $973.1 million, and the company bought back $306.5 million of stock after $201.5 million a year earlier.

The growth the AI screen cares about is Intelligent Agreement Management, the platform Docusign launched in April 2024 around its Iris contract engine, the Navigator repository, and the Maestro workflow builder. IAM was 15.1% of total ARR on July 31, up from 12.6% on April 30. Chief executive Allan Thygesen used the print to say the agents are now running contract workflows end to end.

We said IAM would be the agreement system of action, and this quarter we delivered. Our AI agents are now securely executing contract workflows end-to-end, and the IAM platform also ingested a record volume of agreements.

Allan Thygesen, CEO, September 3 results release

IAM’S SHARE OF DOCUSIGN ARR

1. January 31, 2025: IAM is 2.3% of company ARR as the platform’s first full fiscal year ends.
2. January 31, 2026: IAM is 10.8% of ARR, more than $350 million, against $3.272 billion of company ARR.
3. April 30, 2026: IAM’s share reaches 12.6% after the first quarter of fiscal 2027.
4. July 31, 2026: IAM’s share reaches 15.1% of ARR.
5. January 31, 2027: Management guides IAM to about 18% to 19% of ARR as the fiscal year exits.

The company raised its fiscal 2027 revenue guidance to $3.499 billion to $3.507 billion, 9% at the midpoint, and lifted ARR growth to 8.50% to 9.00%. Third-quarter revenue is guided at $886 million to $890 million. Trailing-twelve-month revenue is $3.36 billion. Standalone eSignature is still expected to be the majority of revenue for the foreseeable future, which is the gravity IAM has to escape.

Docusign also opened a Model Context Protocol server so agreement data and actions can run inside tools customers already pay for, with connectors to Anthropic, Gemini, OpenAI, and Microsoft Copilot, plus Slack, Perplexity, and Google Cloud’s Gemini Enterprise for Legal. That is distribution. It is also an admission that the model layer is where a growing share of the work now starts, and that Docusign’s job is to remain the system that is allowed to file, bind, and undo the result. When an agent can execute a simple contract, the painful remaining work is permission, audit, and rollback, not the click that used to be the product.

Three Tickers, Three Different Bets on the Same Wave

Put the September 8 closes next to the latest quarter and the AI mix each company will actually admit to, and the screen looks less like a single trade than three mismatched claims on the same labor-to-software shift.

THE TAPE AGAINST THE SCREEN

BlackLine and Pagaya are almost the same size in the market, about $1.8 billion, with very different engines underneath. Docusign is roughly seven times larger and already throws off more than $300 million of free cash in a quarter, which is why it can buy back stock while it waits for IAM to become the company. None of the three is priced like a lab. All three are still sold to retail screens as if the next increment of automation accrues to the application layer by default.

Free Clones and Frontier Models Split the Same Work

A June 2026 Stanford CodeX paper by Jay Mandal and Aparna Sinha on product moats for vertical AI applications ranks what actually holds once the model is a commodity: workflows and interface first, then vertical tools, built-in compliance, a data-driven operating core, and embedded judgment. Operational excellence in distribution only strengthens those, it does not replace them. That ranking is a useful way to read these three names without pretending they are Anthropic.

WHAT STILL STANDS BETWEEN THE MODELS AND THE INVOICE

• Controls: BlackLine’s pitch is an auditable close, not a faster spreadsheet, which is why buyers now spend extra weeks on AI governance before they sign.
• The network: Pagaya’s pitch is 35-plus lenders, a 174-name funding base, and a 1% conversion filter on a torrent of applications, which is harder to clone than a scorecard.
• The binding act: Docusign’s pitch is identity, consent, and a legally durable signature sitting under IAM, which is why it is pushing MCP into other people’s agents rather than trying to out-write Claude.
• The clone layer: Open-source e-sign projects are already shopped as free stand-ins for paid signature seats, which does not kill the enterprise contract but does cap what a mid-market seat can charge.

Builders who swap paid tools for GitHub repos are not Docusign’s Fortune 500 book, and they are a real ceiling on seat expansion once “send for signature” is a weekend project. The same logic hangs over credit. If the decisioning layer becomes a call to a frontier model, Pagaya has to prove the network still earns 4.2% of volume. BlackLine has to prove Verity is a control system finance chiefs will not reroute through a general agent.

Anthropic’s explorer is explicit that in the fatter growth cases, more of the extra output goes to capital, and knowledge-worker pay stalls or falls. These three companies are capital. They are also the layer most exposed to being rented by the labs, wrapped by open source, and delayed by the same AI reviews that their own road maps made necessary. The September 8 closes already treat that as the base case. The next filings will show whether IAM’s mix, Verity’s agents, and Pagaya’s auto book change the multiple, or only the product demo.

Disclaimer: This article is news reporting and analysis for information only. It is not investment advice, a recommendation to buy or sell any security, or a personal portfolio plan, and it does not consider any reader’s objectives, risk tolerance, tax position, or time horizon. Readers should consult a licensed financial adviser or other qualified investment professional before acting on any ticker, multiple, or outlook discussed here. Share prices, market caps, guidance ranges, and product-mix figures come from company releases, filings, and market data dated through September 8 and 9, 2026, and all of those figures can move.