Baidu updated its Assistant so it can handle 90% of work tasks
Baidu has updated its Assistant — now the AI agent handles complete work tasks instead of only answering individual questions. Agents already help people write reports, create presentations, research competitors, and draft meeting notes. But they often break down halfway through complex work. Baidu’s new update aims to let the agent complete up to 90% of everyday office tasks without failures.
AI-processed from Jiqizhixin (机器之心); edited by Hamidun News
Baidu on July 10, 2026, announced an update to its AI assistant, aimed at overcoming the main limitation of modern agents. The new version is focused on handling complete, multistep work tasks rather than simply answering individual questions.
Why Agents Used to Fail on Complex Tasks
Early AI agents made a good impression in demonstrations. People tried them for daily work: asking them to write reports, prepare presentations, research competitors, compile meeting notes.
The first experience often impressed. But once you integrated the agent into a regular workflow, problems emerged:
- Agent handles an individual question but cannot take on a complete task
- Generates content but inconsistently — ranging from good to poor
- Loses track of long instructions or forgets conditions partway through
- Fails to notice errors or cannot fix them
Result: the worker must verify each step, recalculating figures, rewriting sections of text, clarifying details. This kills the time savings.
What Changed in Baidu's Update
The company announced several key improvements to agent architecture:
- Better decomposition of complex tasks — the agent breaks work down into logical substeps on its own
- Stable context management — remembers all conditions and constraints throughout the entire action chain
- Built-in verification — checks its own results and attempts to fix errors before final output
- Adaptation to user style — learns from previous sessions what format answers should take
According to Baidu, the updated agent handles complete tasks in 90% of cases on the first attempt. This means the human does not need to intervene or recalculate the result.
Where This Will Be Useful
Main applications in corporate environments:
- Analytics and reporting — compiling quarterly reports, metric summaries
- Presentations — not just text but structure, slide logic, even design recommendations
- Market intelligence — gathering information on competitors, their products, prices, strategies
- Documentation — formatting meeting notes, extracting action items, distributing tasks
- Data work — cleaning tables, aggregating, creating pivot tables
Baidu is focusing on white-collar and knowledge workers, where such tasks occupy 30–50% of working time.
What This Means
If Baidu's claims match reality, agents are moving out of the "assistant for individual operations" phase into the "full automation of a task class" phase. Instead of asking AI for help 20 times a day for small things, a person can give one complex task and receive a finished result.
This will reshape office work dynamics: if one agent can replace an assistant on typical document tasks, demand for such positions will fall while demand for agent management and complex analytics will rise.
- Oct 28, 2026Qwen 3.8 Routed Requests to an Alibaba Cloud Proxy, With Up to 4% Failures in Tests
- Oct 28, 2026OpenAI releases GPT-6.1 Terra: half the price and nearly on par with its senior models
- Oct 28, 2026OpenAI Releases GPT-6.1 Terra: $1 per Million Input Tokens and Three Times Faster Than Astra
- Oct 28, 20262026 Open-Source Models Claim 1M Context but Read It Selectively
Key Questions Answered
Why Agents Used to Fail on Complex Tasks?
Early AI agents made a good impression in demonstrations. People tried them for daily work: asking them to write reports, prepare presentations, research competitors, compile meeting notes. The first experience often impressed. But once you integrated the agent into a regular workflow, problems emerged: Agent handles an individual question but cannot take on a complete task Generates content but inconsistently — ranging from good to poor Loses track of long instructions or forgets conditions partway through Fails to notice errors or cannot fix them Result: the worker must v
What Changed in Baidu's Update?
The company announced several key improvements to agent architecture: Better decomposition of complex tasks — the agent breaks work down into logical substeps on its own Stable context management — remembers all conditions and constraints throughout the entire action chain Built-in verification — checks its own results and attempts to fix errors before final output Adaptation to user style — learns from previous sessions what format answers should take According to Baidu, the updated agent handles complete tasks in 90% of cases on the first attempt. This means the human doe
Where This Will Be Useful?
Main applications in corporate environments: Analytics and reporting — compiling quarterly reports, metric summaries Presentations — not just text but structure, slide logic, even design recommendations Market intelligence — gathering information on competitors, their products, prices, strategies Documentation — formatting meeting notes, extracting action items, distributing tasks Data work — cleaning tables, aggregating, creating pivot tables Baidu is focusing on white-collar and knowledge workers, where such tasks occupy 30–50% of working time.
Delegate Deep Research, Coding & Multi-Step Workflows to Nomad Agent
Deploy autonomous task execution with Nomad Agent (agent.nomadnet.ai) — automated web research, data synthesis, code generation, and MCP tool orchestration.
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.