The “Anthropic Sell-Off”: AI Agents, Workflow Jenga, and How New Investors Can Understand the AI Stock Race

Educational / news commentary — not financial advice.

In early February 2026, a phrase started floating around investing circles: the “Anthropic sell-off.” It’s not an official market term—more like a shorthand for a very specific fear that hit the market all at once:

If AI “agents” can do real workplace tasks end-to-end, a bunch of “middleman” software and information businesses might lose pricing power… fast.

And when Wall Street gets spooked about pricing power, it doesn’t politely adjust a few spreadsheets. It hits the sell button.

Learn Investment Fundamentals on my investing page .

1) What actually happened?

Trigger: Anthropic (maker of Claude) rolled out agentic plug-ins for Cowork—tools designed to let Claude take action inside real workflows. One of the attention-grabbers was a legal/contract workflow tool aimed at tasks like NDA and contract review (exact details vary by reporting, but the market latched onto the legal/workflow angle immediately).

Market reaction: Investors treated it like a warning flare:
“If an AI agent can handle chunks of legal review + clerical workflow, what happens to the companies who sell software/data to support those humans doing that work?”

So the selling wasn’t “Anthropic stock” (Anthropic is private). The sell-off hit public companies seen as exposed—especially legal information, publishing/data, and a broader swath of software/services names. Reuters described it as a debate over AI’s “existential threat” to parts of software and services.

2) The big metaphor: “Workflow Jenga” (aka removing the middle blocks)

Here’s the simplest way to picture the fear:

Traditional workflow tower (Jenga)

A lot of white-collar work looks like a tower of steps:

Intake request
Find the right document/data
Compare to policy/playbook
Flag issues
Create summary
Route approvals
Update systems of record (CRM, contract repository, ticketing, billing)

Middleman companies often monetize one or more of those blocks.

The “agent” threat

If an AI agent can do steps 2 → 7 in one pass (with human review at the end), it can feel like someone yanked out multiple Jenga blocks at once—and investors worry the tower (pricing model) wobbles.

That’s the emotional core of the “Anthropic sell-off.”

3) What kinds of roles could get compressed? (Real examples)

This is the part new investors should understand: it’s not always “job replaced,” it’s often “work compressed.” Entry-level and workflow-heavy roles tend to be most exposed.

Example A: Contract / NDA review support (paralegal + contract analyst)

What the role entails (today):

Intake NDA/contract from email or portal
Check it against a playbook (approved clauses, fallback language)
Highlight risky clauses (indemnity, assignment, confidentiality term, governing law, liability cap)
Produce a summary + route to counsel for final signoff
Log status in a tracker / CLM tool

How an agent can replace part of it:

Agent reads the NDA, compares it to your “standard,” flags differences, proposes edits, drafts the summary, and routes it—leaving a lawyer to review the final output. This is exactly the kind of workflow “agentic” plug-ins are designed to target.

Example B: Compliance / policy workflows (compliance analyst)

What the role entails:

Gather documents and evidence
Map them to a checklist
Write reports and remediation tasks
Coordinate approvals and filing

Agent effect:

Agent pulls evidence, drafts the report, creates tickets, and follows up automatically.

Example C: Sales ops / customer success admin work

What the role entails:

Summarize customer calls
Draft follow-ups
Update CRM fields
Trigger internal tasks (support tickets, onboarding steps)

Agent effect:

Plug-ins can draft emails, summarize meetings, and push updates into systems—reducing the “glue work” that powers SaaS seat counts.

4) Who got sold off—and what do they do?

Different outlets reported slightly different intraday numbers, but the theme was consistent: legal info + data + workflow-heavy software got hit hardest.

Here are the types of companies that were treated like “targets,” plus widely reported examples:

Legal information / professional research (front and center)

Thomson Reuters (TRI) – legal research/workflow products like Westlaw; shares dropped roughly mid-teens in the initial wave of selling (reports ranged around ~16% to ~18%).
RELX (RELX) – owns LexisNexis (legal & risk info); reported down around the mid-teens.
Wolters Kluwer (WKL.AS) – professional/legal & compliance tools; also reported meaningfully down in the same wave.

Broader “information is the product” names

Pearson (PSON.L) – education publishing/data; got swept into the concern trade.

Workflow-heavy software and services (broader contagion)

Reuters described a much broader pullback across software/services as investors debated whether AI agents could pressure business models and valuations.

5) “Can the middlemen survive?” Yes—some are built to.

This is where the investor brain has to kick in.

Even Reuters’ analysis and commentary emphasized the market is arguing about degree, not declaring automatic extinction.

Middlemen with stronger survival traits

These businesses tend to be stickier:

Proprietary data moat (exclusive datasets, content rights, citations, court data, deep archives)
Workflow embed (already integrated into how enterprises operate—compliance, audit trails, approvals)
Trust + liability (regulated environments care about defensible sources and accountability)
Distribution (they’re already the default platform inside big orgs)

In other words: even if AI gets smarter, data + trust + distribution can still win.

Middlemen more exposed

Businesses tend to be more exposed when:

Their value is mostly “UI over public data”
They rely on seat-based pricing for work that can be automated
Switching costs are low
The “job to be done” is mostly summarizing, drafting, routing, and checklisting

That’s the logic investors were expressing with their sell orders—not a guaranteed future, but a repricing of risk.

6) The chessboard: the “players” new investors should know

Think of AI like an economy-wide supply chain:

A) Model builders (the “brains”)

Anthropic (Claude) — private (no ticker), but hugely influential in the narrative.
OpenAI — private
Google (Alphabet — GOOGL/GOOG) — builds Gemini and sells AI via cloud & products
Microsoft (MSFT) — AI via Azure + Copilot ecosystem
Meta (META) — open models + AI in ads/social
Amazon (AMZN) — AI via AWS + partnerships

B) Compute & chips (the “muscle”)

Nvidia (NVDA) and the broader semiconductor stack remain core “picks and shovels.” Reuters even noted Nvidia’s CEO pushed back on the idea that AI “replaces” software—an important reminder that narratives can overshoot.

C) Cloud and infrastructure (the “factories”)

Hyperscalers + data center infrastructure are the backbone of the AI wave.

D) The application layer (where “Workflow Jenga” happens)

This is where the market is asking: which apps become AI-enabled winners vs. which ones get unbundled?

7) “Who are the horses?” Public tickers investors often use for AI exposure

Because Anthropic and OpenAI are private, public-market investors typically “bet the race” through public proxies:

The “picks & shovels” bucket (often the most durable AI exposure)

NVDA (chips/accelerators)
TSM (chip manufacturing)
AVGO (infrastructure semis)
ASML (critical chipmaking equipment)

The platform bucket (distribution + customers)

MSFT, AMZN, GOOGL, META (AI embedded into massive platforms)

The “app winners” bucket (more volatile, but potentially huge upside)

Enterprise software that successfully becomes the control panel for AI agents (not all will).

Investor takeaway: the horse race is real—but it’s also why diversification matters.

And you already nailed the psychology with your Benioff example: loyalty can flip fast when models leapfrog each other.

8) How can a new investor get exposure? (Stocks vs. ETFs)

Option 1: Pick 3–5 “horses” + diversify

A simple framework:

2 “picks & shovels”
2 “platform” names
1 “app layer” pick (higher risk)

Option 2: Use AI-focused ETFs (broad exposure, less single-stock stress)

A few widely followed AI/robotics ETFs and what they aim to do:

AIQ (Global X Artificial Intelligence & Technology ETF) — broad AI + enabling tech
BOTZ (Global X Robotics & Artificial Intelligence ETF) — more applied automation/robotics tilt
WTAI (WisdomTree Artificial Intelligence and Innovation Fund) — diversified, multi-sector AI theme
CHAT (Roundhill Generative AI & Technology ETF) — explicitly “generative AI” themed; marketed as the first generative AI ETF
Kiplinger’s January 2026 list also highlights AI ETF choices like AIQ, BOTZ, and iShares options such as ARTY, plus others.

(Reminder: ETF holdings and weights change—always check the fund’s current holdings before buying.)

Option 3: Indirect exposure to Anthropic/Claude

You can’t buy “Anthropic (CLAUDE)” as a stock today, but you can understand who benefits financially:

Amazon (AMZN) invested heavily in Anthropic and is its primary cloud/training partner; Anthropic has publicly stated Amazon’s total investment reached $8B while remaining a minority investor.
Reuters has also reported Anthropic exploring IPO preparations as early as 2026 (not a guarantee, but important context).

9) What should a new investor do with the “Anthropic sell-off” lesson?

Here’s the practical checklist I’d use anytime a “new tech just killed X” narrative hits:

Is the fear about demand… or pricing power?
Does the company own proprietary data or just wrap it?
How embedded is it in enterprise workflows?
Can it become the AI “control panel” instead of getting replaced?
Are you buying a long-term trend… or chasing a 48-hour headline?

This sell-off was a reminder that markets move on stories first, and earnings later.

admin

See Full Bio