Most founders still think they’re competing on “best interface.” Notion’s latest AI announcement makes that fantasy obsolete: the real fight is over who owns the workflow, not who has the nicest dashboard.
Last week, Notion rolled out a new developer platform that turns its workspace into an orchestration layer for AI agents, external data sources, and automated multistep workflows. This isn’t “we added an AI sidebar” — it’s “your documents, databases, and agents now coordinate inside Notion, not inside your app.” In parallel, Google’s Cloud Next updates rebranded Vertex AI into the Gemini Enterprise Agent Platform and shipped Workspace Studio so non-technical teams can build agents directly inside Gmail, Docs, Sheets, and the rest of the suite. Together, those moves say the quiet part out loud: work is moving into agent hubs, and your product is expected to plug in, not own the whole stage.
What actually changed
Notion’s update gives teams a way to connect custom AI agents, external data, and code into the same workspace where specs, tickets, and roadmaps already live. It effectively promotes Notion from collaborative notebook to coordination surface for both humans and agents. Google is doing the same thing for the “we live in Workspace” crowd: Workspace Studio lets any business user describe an automation in plain language and have an AI agent run it across Gmail, Docs, Drive, Meet, Chat, plus external tools like Asana or Salesforce. When office software becomes an agent control room, every other SaaS in that stack either becomes an input, an output, or background noise.
If you’re building an “AI copilot for X” that lives in its own siloed UI, this should worry you. These platforms are training customers to expect that agents live where the work already happens, not in Yet Another Tab. They’re also normalizing something founders underestimate: agents need structured context, clear affordances, and predictable APIs more than they need another chat box.
Why this matters for your product, not just your pitch
Founders love to say “we’re the OS for [domain].” Notion and Google just reminded everyone that the OS for most knowledge work is… the thing teams already live in eight hours a day. If you’re pre-Series B, you are not the OS. At best, you’re the specialist brain that plugs into someone else’s OS — or you’re a toy that gets replaced when the platform ships a native integration.
This shift has three concrete implications for AI product design:
- UX has to assume an agent is using your product as much as a human.
- Your real product is the workflows and surfaces that agents can reliably call, not just the screens humans click through.
- “Activation” now includes: how fast can a customer wire you into their existing workspace agents without calling engineering.
That last one is brutal. If your team needs a two-week implementation project to connect to Workspace, Notion, or whatever hub your buyers live in, you’ve already lost to a worse product with a cleaner integration story.
Designing for agents as first-class users
Designing for agents is not just “add webhooks and pray.” It means treating machine users like the most impatient power users you’ll ever have. They need: stable contracts, clear error states, idempotent actions, and predictable latency. They don’t care how pretty your sidebar is; they care whether “create_draft_invoice” behaves exactly the same every time, under load, with partial data.
On the human side, UX has to expose that agent layer in a way that feels trustworthy instead of spooky. That means:
- Explicit “what this agent can and cannot do” inside the UI.
- Clear logs and reversibility when an agent touches critical data.
- Onboarding that frames agents as teammates operating in a shared workspace, not mysterious gods behind a black box.
At Poplab, most AI founders I work with are stuck between “we want to be the OS” and “we can’t out-Notion Notion,” so we design for a third option: become the specialist system of record or decision engine that agents must call for credible outcomes. The surface still matters — but mainly as the place where humans debug and steer what agents are doing.
One thing to do this month
Don’t boil the ocean. Pick one high-value workflow where your product is already delivering real outcomes — for example, “score inbound leads,” “triage support tickets,” or “generate risk summaries.” Redesign that flow to be agent-first: a clean API, a minimal set of required inputs, clear success/failure responses, and a simple UI panel that explains what the agent did in human terms.
Then ship one deep integration into the hub your customers actually live in — Notion, Workspace, or whatever their real OS is — and measure how many accounts adopt that path over your standalone UI. That data will tell you more about your product’s future than another round of landing-page A/B tests (though if you suspect your page is lying to investors, start here).
The agent era won’t kill your product. But it will absolutely demote it if you keep designing as if humans are the only ones using it.
Leave a Reply