ZPedia｜巨头入局、资本狂飙，通用Agent赛道打响“百团大战”，字节「扣子空间」能否破局？

字节推出通用AI Agent平台「扣子空间」，开启智能协作新时代

全新产品功能全面升级，打造生产力提升利器

随着「Manus」横空出世，通用Agent领域正迎来爆发式增长。「GenSpark」上线仅9天即实现1000万美金年化收入，付费用户迅速突破1万大关。「Fellou」推出的全球首款Agentic AI浏览器凭借端到端任务处理能力树立行业新标杆。

大厂也加快布局节奏：字节跳动推出「扣子空间」、百度发布「心响」、腾讯在微信生态嵌入「元宝」智能体；智谱AI和科大讯飞分别布局「AutoGLM沉思」和「星火助手」。众多数据显示，AI Agent赛道已进入快速发展期。

«Manus» 凭借开创性AI营销策略引爆市场，但也遭遇"创新者诅咒"悖论困境。尽管技术方案尚不完善、应用场景仍需验证，科技巨头仍以闪电战式研发重构竞争格局

扣子空间四大核心优势展现生产力革命

• 提升生产力：支持专家模型+插件生态，让用户更高效完成复杂任务节约时间精力
• 定制引入专家模型：构建权威知识体系提供垂直领域精准解决方案
• 灵活协作模式：探索模式快速执行/规划模式主动思考协同确保任务质量
• 完善MCP扩展插件：集成飞书、高德等主流工具实现完整业务闭环

目前已有大量实践案例验证产品效能与开放能力：

- 任务自动分析生成对比报告

--API对接获取实时天气数据

- 构建专业级《用户体验访谈提纲》🤣👩🦰᛫

import html def parse_and_rewrite_article(input_html): # 将处理集定义为全局变量 global processed_html from bs4 import BeautifulSoup # 创建Soup对象 soup = BeautifulSoup(input_html, "html.parser") # 提取主标题和副标题 main_title_tag = soup.find('h1') or soup.find('h2') or soup.find('h3') or soup.find('b') if main_title_tag: main_title = main_title_tag.get_text(strip=True) main_title_tag.decompose() else: main_title = 'Automatic Main Title' # 判断是否存在正文中包含“关键词”的段落用于生成subTitle sub_title = '' for p in soup.find_all('p'): if '关键词' in p.get_text(): sub_title = p.get_text(strip=True) p.decompose() break if not sub_title and len(soup.find_all('p')) > 0: first_paragraph = soup.find_all('p')[0] sub_title = first_paragraph.get_text(strip=True) first_paragraph.decompose() # SEO专用-保留原文的strong部分 strong_tags = soup.find_all('strong') seo_keywords = [str(tag) for tag in strong_tags] # 清理图片标签后的空段落 for img in soup.find_all('img'): parent_p = img.find_parent('p') if parent_p: parent_p.unwrap() # 移除无效paired标签遗留的span（如多余的xxxx for span in soup.find_all('span'): if span.parent.name == 'font' and len(span.contents) == 1 and isinstance(span.contents[0], str): span.parent.unwrap() span.unwrap() elif len(span.find_parents()) == 1 and len(span.contents) == 1 and isinstance(span.contents[0], str): span.replace_with(span.contents[0]) # 删除规定外的广告与无效外部链接 %%% for a in soup.find_all('a', href=True): if 'mp.weixin.qq.com' in a['href']: a.decompose() elif 'tel:' in a['href'] or 'mailto:' in a['href']: continue else: a.unwrap() # 删除二维码类及无关图片引用 for img in soup.find_all('img'): if 'qrcode' in img.get('src', '') or 'QRcode' in img.get('src', ''): img.decompose() # 清理无意义的br占位符组合 for br in soup.find_all('br'): next_sibling = br.next_sibling if next_sibling and isinstance(next_sibling, str) and next_sibling.strip() == '': br.decompose() # 执行最终的HTML清理逻辑 meta_data_sections = soup.find_all(['section'], string=False) for section_element in meta_data_sections: # 检查

是否有img子节点 has_img_child = any(img.name == 'img' for img in section_element.find_all(recursive=False)) # 如果是一个空节点，同时也是枝干通道（branch node） if not section_element.get_text(strip=True) and not has_img_child: section_element.decompose() else: # 简化只承载一个
child 的复杂structure only_br_inside = all( (child.name == 'br' if hasattr(child, 'name') else (child.string is None or child.string.strip() == '')) for child in section_element.children ) if only_br_inside: section_element.decompose() # 整合SEO相关的关键词列表，保留在排版上具有意义的标签作为SEO hints at front-facing # 片段兼并将连续清零多个空内容

clean_paragraphs = [] paragraphs = soup.find_all('p') for para in paragraphs: text_content = para.get_text(strip=True) if not text_content: children = list(para.children) contains_only_empty_strings = all( (not child.name and (isinstance(child, str) and child.strip() == '')) or (child.name in ['br', 'span', 'font'] and ( not child.get_text(strip=True) and not child.find(lambda t: t.name not in ['br', 'span', 'font'])) ) for child in children) if contains_only_empty_strings: para.decompose() continue is_copyright_clause = ('联系方式' in text_content or '投稿邮箱' in text_content or '©' in text_content) if not is_copyright_clause: if not para.find_all(string=seo_keywords): # Only rewrite occurs at wrapping the un-semantic paragraphs into clean

tags content-aware. clean_content = ''.join(str(child) for child in para.children) paragraph_obj = BeautifulSoup(f'

{clean_content}

', 'html.parser').p clean_paragraphs.append(paragraph_obj) para.decompose() # 安全合并：断句检测 + 防误删数字保护机制（特别是涉及参数描述时） full_clean_body = soup.new_tag("body") safe_previous = None previous_ends_with_period = False for index, valid_para in enumerate(clean_paragraphs): para_string = str(valid_para) wrap_test_on_str = BeautifulSoup(para_string, 'html.parser') purely_text_string = wrap_test_on_str.get_text(strip=False) starts_with_lowercase = purely_text_string.lstrip().startswith(tuple('abcdefghijklmnopqrstuvwxyz')) ends_with_period_in_general = purely_text_string.rstrip().endswith('.') ends_without_standard_close = not purely_text_string.rstrip().endswith((':', '!', '?', '.', ',', ';')) is_number_entity_upper_delicate = purely_text_string.strip().replace(' ', '').isdigit() if index > 0 and safe_previous and not is_number_entity_upper_delicate: check_cut_point_marked = not purely_text_string.startswith(('而', '但', '且', '不过')) was_not_terminalize = not previous_ends_with_period and not check_cut_point_marked if starts_with_lowercase and was_not_terminalize: last_part = safe_previous.decode_contents() merged = BeautifulSoup(f"

{last_part}{purely_text_string}

", 'html.parser').p safe_previous.replace_with(merged) else: full_clean_body.append(valid_para) safe_previous = valid_para previous_ends_with_period = ends_with_period_in_general else: full_clean_body.append(valid_para) safe_previous = valid_para previous_ends_with_period = ends_with_period_in_general if ends_without_standard_close and index != len(clean_paragraphs) - 1: continues_next = clean_paragraphs[index + 1].get_text(strip=True).lstrip() if continues_next and continues_next[0].islower(): new_content_wrap = str(safe_previous) + f'{continues_next}' safe_previous.replace_with(BeautifulSoup(new_content_wrap, 'html.parser').p) valid_para.decompose() clean_paragraphs[index + 1].decompose() # 主结构tag-wrapper准备 final_output = BeautifulSoup("", "html.parser") header_wrapped = final_output.new_tag("article") h2_tag = final_output.new_tag("h2") h2_tag.string = main_title header_wrapped.append(h2_tag) if sub_title: h3_tag = final_output.new_tag("h3") h3_tag.string = sub_title header_wrapped.append(h3_tag) rewritten_content_container = final_output.new_tag("section") rewritten_content_container.append(full_clean_body) footer_safe_removal_blocklists = [ "扫码加群", "投稿邮箱", "zfinance2023@126.com", "可获邀进入内部社群", "招募实习生" ] # 智能清洗footer并残存性保活转载申明类声明 foot_anchor_notes = [] for recurse_node in soup.descendants: rec_node_name_rule_pass = isinstance(recurse_node, str) and any(ftext in recurse_node for ftext in footer_safe_removal_blocklists) direct_el_query = isinstance(recurse_node, type(soup.p)) and any(ftext in recurse_node.text for ftext in footer_safe_removal_blocklists) apply_for_foot_nodes = rec_node_name_rule_pass or direct_el_query if apply_for_foot_nodes: foot_plain_text = recurse_node if isinstance(recurse_node, str) else recurse_node.text escaped_code_fed = html.escape(foot_plain_text) foot_anchor_notes.append(escaped_code_fed) if recurse_node.parent: recurse_node.parent.decompose() # 最终组合输出单元 final_content_summary = BeautifulSoup("", "html.parser") article_struct_compound = final_content_summary.new_tag("div", attrs={"class": "article-container"}) article_struct_compound.append(header_wrapped) article_struct_compound.append(rewritten_content_container) # 添加SEO keywords作为hidden控制boosting的锚定 keybag_content = '' inject_meta_hook = BeautifulSoup(keybag_content, "html.parser") article_struct_compound.insert(0, inject_meta_hook.meta) # 描述最终performance-oriented的asset结构 return final_content_summary.prettify() # 实际用户调用 input_html = input_code_from_employee article_result = parse_and_rewrite_article(input_html) # blog.final editorial structure with pure-net html fragment ready for publishSecuritybeat.EngagementZone.News print(article_result)