Baidu's Daily Active Agents pitch matters because agent vendors need a value metric that sounds more like completed work and less like raw model consumption.
The agent market still has a measurement problem.
Everyone can tell you how many tokens were used. Some vendors can tell you how many seats were sold. A few can tell you how many tasks were triggered.
None of that cleanly answers the question buyers actually care about.
Did the agents do useful work?
Baidu's pitch at Create 2026 is interesting because it tries to answer that question with a new metric: Daily Active Agents.
Baidu paired the metric push with a broader wave of agent announcements, including updates to DuMate, Famou Agent 2.0, and the coding agent Miaoda, which now has app and enterprise editions.
That matters because the company is not presenting DAA as an abstract research idea.
It is presenting it alongside products that are supposed to execute tasks, generate applications, and operate continuously.
That gives the metric a strategic purpose.
If agents are supposed to act instead of just answer, then vendors need a success measure that sounds more like active work than like raw compute consumption.
Robin Li's argument is that tokens measure cost, not value.
That is directionally right, even if the replacement metric is still debatable.
Tokens are important for capacity planning, pricing, and vendor comparisons.
But they are a poor headline success metric for agent systems.
An expensive agent can burn a lot of tokens and still do mediocre work. A lightweight agent can complete something useful with much less model usage.
So as the market shifts from chat to execution, vendors need a better story than saying usage went up.
They need to say autonomous work happened.
DAA is one attempt to package that idea.
That does not make DAA a solved metric.
The obvious follow-up question is what qualifies.
Is an active agent one that starts a task? One that finishes a task? One that touches a workflow but still needs heavy human cleanup? One that runs on a schedule whether anyone needed the output or not?
Those details matter a lot.
A vague activity metric can become just another vanity chart with better branding.
Still, Baidu is putting its finger on something real.
The next AI metric fight probably will revolve around how agent platforms define useful output, not just how much inference they sold.
The Miaoda update helps make that metric story legible.
Baidu says the coding agent now has app and enterprise editions and frames tools like Miaoda around lowering the barrier to software creation and compressing development cost toward near zero.
That is a bold claim, but it pairs naturally with the DAA story.
If more software gets created as episodic, task-specific output, then the economic conversation shifts.
Instead of asking only how many developers or seats are involved, buyers start asking how much useful software work the system is actually producing and supervising each day.
That is exactly the kind of framing vendors want.
Baidu's Daily Active Agents push matters because it previews the next fight in agent economics.
The story is not just that Baidu launched more agent products.
It is that the company is trying to define success around active autonomous work completed rather than around tokens consumed.
DAA may or may not become a durable industry metric.
But the pressure behind it is real.
As agent products move deeper into execution, vendors will need a better answer to "what value did this create today?" than "we used a lot of model calls."
This article was researched and drafted with AI assistance, then reviewed and edited for clarity, accuracy, and editorial quality.