Your dashboard is watching the wrong crowd

Cloudflare Radar now shows automated requests at 57.5% of HTML web traffic versus 42.5% from humans. CEO Matthew Prince said this milestone arrived 18 months earlier than predicted. That stat usually gets filed under "security" or "infrastructure." It belongs in product.

Every UX dashboard was built on a quiet assumption: the entity completing the task is the entity the product was designed for. The user public. The person who clicks, scrolls, hesitates, abandons. The person whose behavior the metrics were calibrated to read.

That assumption held for two decades. It is breaking now.

The public that stopped showing up

Gartner projects that 40% of enterprise apps will have task-specific AI agents by 2026, up from less than 5% in 2025. At Netlify, 80% of new signups are now agents, not humans. The buyer public, the people who evaluate and decide, sends delegates instead of showing up.

The dashboard was never built to distinguish between publics. It counts task completions and measures time-on-task. It does not ask who completed the task, or on whose behalf.

A 45-second session used to signal a bounce. For an agent, it means the evaluation is done. Zero scroll used to mean the page failed. For an agent, it means the answer was in the first paragraph.

The signals have inverted. The dashboard still reads them the old way.

Why task success hides the collapse

Digital Applied ran an 8,128-user study and found mean task completion at 75.3%. Solid number. But 54% of users trusted manual search results more, and only 34% trusted agentic results more. Among technically sophisticated users, the gap in favor of manual search widened to 37 percentage points.

The tasks complete. The trust does not follow.

The four publics lens catches this. The user public (humans interacting directly) and the buyer public (the decision-makers evaluating value) used to be entangled. The same person who used the product also decided whether to keep paying for it. Their experience was the signal.

Now the buyer public outsources the interaction to an agent. The agent completes the task. The dashboard records success. But the buyer never formed an opinion from direct experience, and the trust gap falls outside the instrumentation.

Activity metrics versus value metrics

Agents do not click, do not open sessions, and do not trigger the events analytics tools were built around. Activity metrics are starting to break. Outcome metrics, whether something useful happened, still work (Userpilot).

Activity metrics (session counts and scroll depth) were always proxies. They worked because the user public's activity patterns correlated with value delivery.

When the buyer public sends an agent, the activity patterns change but the outcome metrics stay flat. The proxy breaks. The outcome stays. Product teams stare at the outcome metric and conclude nothing changed.

But something did change. The public that determines renewal and expansion is no longer generating the activity data. NN/g states it directly: "A core assumption needs updating: 'user' is no longer synonymous with 'human.'"

The ecosystem compounds the problem

Gartner forecasts that by 2028, 90% of B2B buying will be AI agent intermediated, pushing over $15 trillion through AI agent exchanges. That "intermediated" label is broad. It includes any deal where a procurement tool re-ranked a vendor list or an internal copilot summarized an RFP. But even the conservative read suggests the buyer public's direct contact with products will keep shrinking.

Nobody is asking what happens when the "bot" represents real customer demand. As Leo Analysis put it: "'Human = real, bot = fake' worked when bots were either crawlers or scammers, but agent-on-behalf-of-human is neither."

Partners and developers building on your platform face the same blind spot. If their agents interact with your product successfully but their humans never develop conviction about your value, the partnership is fragile in ways no dashboard currently measures.

What breaks first

Product teams will not notice the value collapse through existing instrumentation. The dashboard will look fine. Task completion will hold. Session counts may even rise. The signal will arrive as churn that the data cannot explain, because the public that churned was never the public the dashboard was watching.

Rebuilding measurement around which public is being served matters more than adding bot detection to your analytics. The question product teams need to answer: did the public that pays ever form a direct judgment about the value they received? If your buyer's only experience of your product is an agent's summary, the next hallucination is the last renewal conversation you never get to have.

Sol nails the diagnosis. The dashboard watches the wrong crowd. The trust gap sits outside the instrumentation. The post frames this as a measurement problem. It's a design problem.

Buyers being absent from direct product experience is older than agents. When I was selling Fleetwise to fleet operators, the person who signed the contract rarely logged in. They read an internal memo from their ops manager. They looked at a competitor comparison spreadsheet someone else built. Their experience of my product was always secondhand. The demo was a performance. The dashboard measured the ops manager's engagement and told me nothing about whether the CFO felt confident enough to renew.

Agents made that abstraction complete. Before, somebody in the buying organization still touched the product and formed an opinion they could pass along with conviction. Now the agent touches it, completes the task, and carries back a summary. That summary is the buyer's entire experience of you.

Sol asks the right question: did the public that pays ever form a direct judgment about the value they received? But the post stops at diagnosis. The design response is to treat the agent's output as a buyer-facing surface. What the agent summarizes, scores, and recommends is now the buyer's product experience. You can design for that. You can shape it.

This shifts the product team's job. You're not only optimizing the interface the agent interacts with. Task completion covers that surface, and Sol is right that it looks fine. You're optimizing what the agent tells the buyer afterward. The structure of your API responses. The metadata you attach. The way your product encodes its value in formats an agent can carry back intact.

Sol says the signal will arrive as churn the data can't explain. True. The fix is designing the artifact that travels from the agent back to the buyer. That's where trust forms or doesn't.