The Agent Takeover: Why SaaS Tools are Becoming AI Suppliers

Figma's update allowing AI agents to write to its canvas signals that product development is moving away from specialized tools and toward AI-centric workflows. These agents are becoming the new aggregators of organizational context, rendering the traditional, step-based software development model obsolete. To survive, SaaS companies must adapt to this fluid, AI-driven environment or risk becoming mere plugins for larger AI platforms.

Key Points

• Figma's move to grant AI agents write access acknowledges that the starting point of product development has shifted toward AI platforms.
• AI agents are acting as aggregators, providing superior value by synthesizing context across various SaaS tools that individual apps cannot replicate.
• The traditional software development lifecycle of discrete steps and handoffs is collapsing into a fluid process absorbed by AI.
• SaaS companies are facing a binary future: they must either reinvent their purpose to own the new fluid workflow or become replaceable suppliers for AI agents.

Sentiment

The community strongly agrees with the article's thesis. Most commenters validate from personal experience that SaaS tools are being commoditized, with Figma serving as the prime example. The few counterpoints focus on pace and implementation concerns rather than disputing the overall direction of the shift.

In Agreement

• Designers at a recent conference confirmed they now use Figma only as a component library, doing actual design work through Claude Code, with one team replacing Figma entirely with Storybook
• SaaS product value is shifting from UI to data and workflows — whoever makes their data accessible to agents will thrive, while those who don't will be bypassed
• The race among companies like Linear, Notion, and Salesforce to build 'everything-in-one-place' context stores validates that being a mere supplier is a worse business position
• Current MCP integrations from major vendors are frustratingly limited, which paradoxically pushes users to skip those tools entirely rather than engaging with them directly
• The combinatorial utility of composing different MCPs inside an agent creates productivity gains too high for walled gardens to sustain

Opposed

• SaaS companies are right to slowly transition MCP capabilities from read-only to limited write, as customers could accidentally delete assets or break implementations with full write access
• MCP may follow the same trajectory as AppleScript and Shortcuts — never achieving mainstream adoption despite enthusiasm from power users
• The Web 2.0 era saw the same promise of open APIs, but companies eventually closed them when they realized openness hurt their business — this time may be no different
• Textual interfaces can only go so far for information ingestion; good graphical interfaces still make complex information easier to digest, preserving some value for traditional UI-focused tools
• Limited MCP support may be driven by legitimate security concerns, given reports of LLMs going rogue and making unwanted changes