On April 14, Figma filed a brief 8-Kwith the SEC: director Mike Krieger resigned,effective immediately.。

That same day, The Information broke an exclusive: Anthropic’s upcoming Opus 4.7 model would ship with a full suite of design tools,directly competing with Figma’s core product.。

Three days later, on April 17, Claude Design launched officially. Figma fell roughly -7%, closing at $18.92(reported by StocksToTrade / CoinCentral). From its April 8 peak through the close on April 17, it dropped roughly $23.15 . -18%Adobe, Wix, and GoDaddy also declined in tandem, by 3-5% (per GuruFocus and others).

It wasn’t just one stock that fell.The market is repricing the entire premise of ‘AI companies shipping downstream applications’.

Krieger’s personal trajectory is short but telling.

He co-founded Instagram in 2010 with Kevin Systrom,2012 selling it for approximately 10 billion dollars to Facebook in 2012 (per Wikipedia / CNBC). In early 2023, he and Systrom relaunched with Artifact—a news app powered by AI—which shut down in early 2024 and was acquired by Yahoo in April. In May 2024, he joined Anthropic asChief Product Officer(Anthropic official announcement, May 15, 2024).

In July 2025, he joined Figma’s board (Figma official blog, July 21, 2025). A product leader joining a competitor’s board is common industry practice—signaling mutual respect and ongoing dialogue. But the moment the SEC filing landed on April 14, everyone understood:He didn’t resign under pressure—he had to resign.Staying seated would have created an overt conflict of interest ahead of next week’s product launch.

From Instagram to Artifact to Anthropic, Krieger has pursued one consistent mission: compressing model capability into consumer-grade experiences. The person who best understands ‘consumer products’ is now pushing AI companies’ boundaries—from the model layer all the way to the application layer.

Anthropic’s release sequence over the past 14 months traces a cleardownward path。

The first cut wascoding—Claude Code debuted as a research preview in 2025 and reached $ 10 billion in annualized revenue by November of that year (Anthropic internal data). This validated a core proposition:model companies can build tools directly—and monetize them.。

The second cut targetedvertical professional services—in late January 2026, Anthropic launched 11 Cowork plugins covering legal contract review, financial analysis, and sales pipeline management. Bloomberg reported the move triggered roughly $ 2850 billion in market-cap erosion across software, financial services, and asset management. For the first time, markets systematically priced in a discount for ‘AI companies eating SaaS’.

The third cut is Claude Design. It targetsproductivity and collaboration tools—design, prototyping, presentations, brand systems.

Between cuts, intervals shrink: 11 months from coding to Cowork, less than three months from Cowork to Design.

Last year, Anthropic’s head of developer relations, Alex Albert, made a widely quoted remark: By 2026, AI agents willreplacemost traditional software products (CNBC, April 9). The emphasis was on “CRUD apps”—applications built around create, read, update, and delete logic, whose business logic collapses under LLMs. CRUD apps—applications built around create, read, update, and delete logic, whose business logic collapses under LLMs.

Many dismissed the comment as typical AI-company hype. A year later,Anthropic is delivering it in product form:：

· Claude Code delivers on ‘editor / IDE-adjacent tools’

· Cowork plugins deliver on ‘vertical professional services’

· Claude Design delivers on ‘design collaboration’

Claude Design doesn’t target CRUD apps—it targets thecentral hub of collaboration workflows. Figma’s real moat has never been ‘vector drawing’—it’s‘everyone on the team is already there’. Anthropic’s precision lies in bypassing Figma’s core strength: instead of redrawing pixels, it lets users ‘generate a prototype in one sentence, export to PDF/PPTX, hand off to Canva for collaboration, or send directly to Claude Code for implementation’ (Anthropic official page, April 17, 2026).

It sidesteps Figma’s thickest wall—and opens shop on the other side.

For two decades, incumbent application-layer companies have relied on four pillars:network effects, collaboration lock-in, data accumulation, and brand inertia。

AI companies aren’t attacking those four walls.They’re relocating the ‘starting point’ of work.。

A product manager used to open Figma to build a prototype. The moment they opened Figma, collaboration, comments, version history, brand libraries, and plugin ecosystems were all instantly available—that was when Figma’s moat activated.

Now they type ‘Build a SaaS landing page for SMBs, using our Notion brand guide’ into Claude.ai. Opus 4.7 reads their codebase and design files, extracts design tokens, and delivers an interactive prototype in seconds. To collaborate? Export to Canva. To engineer? Hand off to Claude Code with one click.

Figma can still compete on features—but not on starting point. Its product may be excellent, but users no longer boot it up to begin work.

Taken together, Anthropic’s three cuts aren’t consuming an entire SaaS shelf—they’re targetingspecific slots on that shelf. The downward pattern breaks into three layers.

1. Tools whose core value is ‘operating a model’.Cursor is the cleanest example—the product is essentially a GUI wrapper for GPT/Claude. When model companies grow their own conversational IDEs, the wrapper loses independent reason to exist. This layer began hollowing out the moment Claude Code launched—and 2026 is the year that hollowing materializes.

2. Tools whose collaboration lock-in can be bypassed via ‘export + bridging’.Figma is the archetype. Its moat runs deep in ‘team commenting, versioning, and shared brand libraries’—but Claude Design avoids that workflow entirely, crossing over via exports to Canva and handoffs to Claude Code.This category won’t die overnight—but it will face sustained pricing pressure.

3. Tools where data accumulation yields real compounding value.Handwritten notes in Notion, years of ticket history and decision context in Linear—these cannot be regenerated by LLMs. This layer remains safe—for now. But its expiration date is clear: the moment model companies embed ‘reading and inheriting your legacy data’ as a native capability, safety ends.

The wall hasn’t fallen—there’s just a highway built beside it.

As the sequence advances, the choice of landing point ceases to be random.

The true shift is in the work starting point. Previously, opening Figma activated collaboration, comments, versioning, and brand libraries all at once. Now the starting point has moved to ‘describing the goal inside the model’—and the originalentry momentwhere the moat engaged has vanished. Anthropic’s three cuts—on coding, professional services, and collaborative design—all land where the entry point used to live: inside the splash screen of a specific SaaS.

By this logic, the next cut is already visible: large-market-cap, CRUD-heavy domains whose moats rely on ‘everyone on the team uses this.’CRM, email, customer support—all three fit. Salesforce, Zendesk, and HubSpot share the same collaboration-wall structure as Figma—and the same bypass mechanism applies.

Once the work starting point shifts, tool proficiency itself depreciates.