Baidu updates its Assistant to handle 90% of everyday work tasks
Baidu has updated its Assistant. The AI agent now handles complete work tasks rather than just answering individual questions. Agents already help people write reports, build presentations, research competitors and draft meeting notes. But they often break down midway through complex work. Baidu's new update aims to let the agent finish up to 90% of everyday office tasks without failures.
AI-processed from Jiqizhixin (机器之心); edited by Hamidun News
On July 10, 2026, Baidu announced an update to its AI assistant aimed at the main limitation of current agents. The new version focuses on completing full, multi-step work tasks rather than just answering individual questions.
Why agents used to break down on complex tasks
The first AI agents looked impressive in demos. People tried them for daily work: asking them to write a report, prepare a presentation, look up information on competitors, or draft meeting notes.
The first experience often inspired admiration. But once the agent was built into a regular workflow, problems surfaced:
- The agent handles a single question but cannot take on a full task
- It generates content, but inconsistently, swinging from good to bad
- It loses the thread of long instructions or forgets conditions midway through
- It misses errors or cannot fix them
The result: the worker has to check every step, recalculating figures, redoing chunks of text and clarifying details. That wipes out the time savings.
What changed in Baidu's update
The company announced several key improvements to the agent's architecture:
- Better decomposition of complex tasks: the agent breaks the work into logical substeps on its own
- Stable context management: it remembers all conditions and constraints throughout the entire chain of actions
- Built-in verification: it checks its own results and tries to fix errors before the final output
- Adaptation to the user's style: it learns from previous sessions which response format is needed
According to Baidu, the updated agent completes full tasks on the first attempt in 90% of cases. That means a person does not need to step in or recalculate the result.
Where it will be useful
The main enterprise use cases:
- Analytics and reporting: compiling quarterly reports and metric summaries
- Presentations: not just text, but structure, slide logic and even design recommendations
- Market intelligence: gathering information on competitors, their products, prices and strategies
- Documentation: formatting meeting notes, extracting action items, assigning tasks
- Data work: cleaning tables, aggregating, building pivot tables
Baidu is targeting white-collar and knowledge workers, for whom such tasks take up 30–50% of working time.
What this means
If Baidu's claims hold up, agents are moving from the phase of an "assistant for individual operations" to "full automation of a class of tasks." Instead of asking an AI for help 20 times a day with small things, a person can set one complex task and get a finished result.
This will reshape the dynamics of office work: if a single agent can replace an assistant on routine document tasks, demand for such positions will fall, while demand for agent management and complex analytics will grow.
- Nov 1, 2026Google taught its StreamVibe video agent to decide when to step in on its own: +10.7%
- Nov 1, 2026New test catches hidden AI agent collusion by measuring how a partner reacts to a paraphrase
- Nov 1, 2026Illinois compiled 12 151 open problems and tested whether AI can solve them
- Oct 31, 2026AI engineer writes a single reward function for agent teams: win rate climbs from 51% to 90%
Key Questions Answered
Why agents used to break down on complex tasks?
The first AI agents looked impressive in demos. People tried them for daily work: asking them to write a report, prepare a presentation, look up information on competitors, or draft meeting notes. The first experience often inspired admiration. But once the agent was built into a regular workflow, problems surfaced: The agent handles a single question but cannot take on a full task It generates content, but inconsistently, swinging from good to bad It loses the thread of long instructions or forgets conditions midway through It misses errors or cannot fix them The result: t
What changed in Baidu's update?
The company announced several key improvements to the agent's architecture: Better decomposition of complex tasks: the agent breaks the work into logical substeps on its own Stable context management: it remembers all conditions and constraints throughout the entire chain of actions Built-in verification: it checks its own results and tries to fix errors before the final output Adaptation to the user's style: it learns from previous sessions which response format is needed According to Baidu, the updated agent completes full tasks on the first attempt in 90% of cases. That
Where it will be useful?
The main enterprise use cases: Analytics and reporting: compiling quarterly reports and metric summaries Presentations: not just text, but structure, slide logic and even design recommendations Market intelligence: gathering information on competitors, their products, prices and strategies Documentation: formatting meeting notes, extracting action items, assigning tasks Data work: cleaning tables, aggregating, building pivot tables Baidu is targeting white-collar and knowledge workers, for whom such tasks take up 30–50% of working time.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.