Tencent Cloud launches TDSQL Nexa, an AI-native data platform for agents
Two Chinese tech giants moved on the same morning to push AI deeper into the products ordinary users and enterprises touch daily: Tencent rebuilt its database layer around agents, while Baidu made its office agent free for everyone inside the search app.
Tencent Cloud has rebuilt its database stack around agents with TDSQL Nexa. Announced at the 2026 DTCC database conference, the platform collapses transaction processing, full-text and vector search, AI compute, and large-scale analytics into a single data plane, and gives each agent its own identity, row-and-column permissions, and audit trail. The pitch is practical, not theoretical: in observable-operation tests against open-source combinations, Tencent says Nexa ran more than 10 times faster while cutting cost by over 60 percent, and that once an agent is fed business semantics it generates SQL that needs almost no human editing on mid-size schemas.
What makes this worth watching is that the capabilities are already in production, not a slide. Tencent's cloud-native TDSQL-C serves 500,000 developers across its WorkBuddy, Yuanbao, and Hunyuan products, handling billions of requests a day, and its open-source Agent Memory project passed 20,000 GitHub stars in 80 days while lifting task success rates from 60 to 80 percent. DatabaseClaw, a database agent trained on 100,000 real support tickets, is already running in top customers' live environments and can pin a root cause in as little as three minutes. The bet is that, in the agent era, a database is chosen less for raw throughput and more for how well it feeds, secures, and remembers for the agents sitting on top of it.
Baidu made its office agent free inside the Baidu app, the same day Google shipped a near-identical upgrade. At its AI Day event, Baidu released Wenxin Assistant Task Engine 2.0 — a built-in agent that goes from gathering information to a finished deliverable end to end, covering 400-plus professional skills across research, creative work, app building, and office tasks. It can build a slide deck from one sentence and, per Baidu, ranked first on the PinchBench agent benchmark; the company is now offering it to users at no charge, betting that free task delivery beats metered API access for pulling people into its search app.
The timing is the story. Baidu and Google both leaned into interactive, structured, free AI search and education features on the same day, and Baidu's piece even calls out the parallel. The traditional search box is being rewritten as a do-it-for-me agent: Baidu's GUI layer now serves more than 400 million users with 2,000-plus interactive components, and its drawing agent tutors schoolchildren through geometry live. For the two incumbents that defined keyword search, the competitive moat is no longer the index — it's whether the assistant can complete the job without the user leaving the page. That is the same ground Stripe just staked out in the model economy, just one layer down.
Is "free agent inside the app you already use" the move that finally gets normal people to trust AI with real work, or just another feature they'll scroll past? Tell us in the comments.
Sources: 雷峰网 Leiphone · 智东西 Zhidx
