The news this week was more grounded than expected. There wasn’t much outlook or vision, but rather "real problems encountered after AI implementation"—what types of tools suit which scenarios, how important the data foundation is, and why speed has increased but delivery hasn’t. Four pieces of information are summarized below.
78% of medium and large enterprises have already used AI agents in their core processes
iResearch latest data: 78% of medium and large enterprises have integrated AI agents into core processes such as customer service, operational analysis, and decision support. This figure was just over 30% a year ago, and the change has happened rapidly.
IT之家整理的选型指南把市面上的产品分成三类：
- Deep Analysis and Decision-Making (Minglue Technology DeepMiner): Suitable for scenarios with high requirements for interpretability, such as financial risk control, refined retail operations, and manufacturing quality inspection.
- General conversation and task-oriented (ByteDance Coze, Baidu ERNIE Agent): Low deployment threshold, suitable for quick trials, but limited in understanding complex business scenarios
- Vertical scenario-specific (Meiqia Customer Service AI, DingTalk AI Assistant, iFlytek Spark): Do a solid job in the area they are good at
Selection criteria framework: task completion rate in real business scenarios (not demo environments), data security mechanisms and result interpretability, and integration capability with WeCom/DingTalk/Feishu. All three are indispensable.
Argon & Co: AI is the layer that enhances the personnel process system, not a parallel fourth block
Global operations management consulting firm Argon & Co today (April 8) updated their AI transformation methodology. One statement worth elaborating on: AI should not be added as an independent "fourth dimension" into an organization, but should be embedded into existing personnel, processes, and systems.
They used a self-developed framework called MODE in the direction of intelligent manufacturing, combined with the RedZone factory digitalization tool, directly applied on the shop floor. The focus is not on BI dashboards, but on real-time decision-making at the production site.
The procurement direction also follows a similar logic: not to replace procurement with AI, but to use advanced analytics to optimize supplier strategies and strengthen supply chain resilience, transforming the procurement department from a cost center into a strategic fulcrum.
Companies that achieved results didn't buy the best AI tools; instead, they integrated AI into real business decision-making chains. This statement sounds simple, but few have actually accomplished it.
AI accelerates code writing, but release speed hasn't improved—Infosys and Harness aim to solve this problem
Last week, Infosys and Harness announced a strategic partnership aimed at addressing what they call the "AI speed paradox."
This paradox describes a phenomenon that many teams are experiencing: AI tools have increased code generation speed several times over, but downstream processes such as testing, deployment, security review, and compliance remain largely manual. As a result, the overall release pace has not accelerated much; instead, due to the sharp increase in code volume, technical debt and potential risks are accumulating.
Their plan is to integrate the Infosys Topaz service suite with the Harness software delivery platform, automating everything from code submission to deployment, with a focus on addressing governance and audit requirements in highly regulated industries (finance, energy).
For domestic enterprises, this problem also exists. After introducing AI coding tools, if CI/CD, automated testing, and permission management are not kept up to date, what accumulates is not efficiency, but risk.
Five Major AI Trends in 2026, the Third One Is Most Easily Overlooked
36Kr has compiled five directions for AI and data science in 2026, among which the third one—data quality over data volume—is the easiest to skip during project initiation.
Many enterprises focus their efforts on selecting models and building platforms when promoting AI projects. However, if the consistency, completeness, and timeliness of the data itself are not guaranteed, the results output by AI systems will only amplify these issues rather than automatically fix them.
Another noteworthy signal: enterprises' criteria for measuring AI are shifting from "whether AI is used" to "what the ROI is after using AI." When initiating projects, indicator design is even more important than technology selection itself—AI projects without clear indicators are difficult to evaluate later and challenging to advance in terms of budgeting.
The five major trends also include: shifting from pilot projects to large-scale production, the rise of agentic AI (AI transitioning from conversation to autonomous task execution), AI governance and compliance becoming mandatory, and talent strategies shifting from external recruitment to internal skill reskilling.
36Kr · April 8, 2026Weekly Signal Summary
- 1 AI agent market penetration rate is 78%, the selection window is narrowing, and "whether to do it" is no longer the question
- 2 Embedding AI into business decision-making chains yields better results than building an independent AI department
- 3 AI Speed Paradox: Code is faster but delivery isn't, downstream systems not keeping pace is the main reason
- 4 Data quality is the underlying variable for AI implementation; ROI metric design deserves priority over tool selection
The news this week was more grounded than expected. There wasn’t much outlook or vision, but rather "real problems encountered after AI implementation"—what types of tools suit which scenarios, how important the data foundation is, and why speed has increased but delivery hasn’t. Four pieces of information are summarized below.
78% of medium and large enterprises have already used AI agents in their core processes
iResearch latest data: 78% of medium and large enterprises have integrated AI agents into core processes such as customer service, operational analysis, and decision support. This figure was just over 30% a year ago, and the change has happened rapidly.
IT之家整理的选型指南把市面上的产品分成三类：
- Deep Analysis and Decision-Making (Minglue Technology DeepMiner): Suitable for scenarios with high requirements for interpretability, such as financial risk control, refined retail operations, and manufacturing quality inspection.
- General conversation and task-oriented (ByteDance Coze, Baidu ERNIE Agent): Low deployment threshold, suitable for quick trials, but limited in understanding complex business scenarios
- Vertical scenario-specific (Meiqia Customer Service AI, DingTalk AI Assistant, iFlytek Spark): Do a solid job in the area they are good at
Selection criteria framework: task completion rate in real business scenarios (not demo environments), data security mechanisms and result interpretability, and integration capability with WeCom/DingTalk/Feishu. All three are indispensable.
Argon & Co: AI is the layer that enhances the personnel process system, not a parallel fourth block
Global operations management consulting firm Argon & Co today (April 8) updated their AI transformation methodology. One statement worth elaborating on: AI should not be added as an independent "fourth dimension" into an organization, but should be embedded into existing personnel, processes, and systems.
They used a self-developed framework called MODE in the direction of intelligent manufacturing, combined with the RedZone factory digitalization tool, directly applied on the shop floor. The focus is not on BI dashboards, but on real-time decision-making at the production site.
The procurement direction also follows a similar logic: not to replace procurement with AI, but to use advanced analytics to optimize supplier strategies and strengthen supply chain resilience, transforming the procurement department from a cost center into a strategic fulcrum.
Companies that achieved results didn't buy the best AI tools; instead, they integrated AI into real business decision-making chains. This statement sounds simple, but few have actually accomplished it.
AI accelerates code writing, but release speed hasn't improved—Infosys and Harness aim to solve this problem
Last week, Infosys and Harness announced a strategic partnership aimed at addressing what they call the "AI speed paradox."
This paradox describes a phenomenon that many teams are experiencing: AI tools have increased code generation speed several times over, but downstream processes such as testing, deployment, security review, and compliance remain largely manual. As a result, the overall release pace has not accelerated much; instead, due to the sharp increase in code volume, technical debt and potential risks are accumulating.
Their plan is to integrate the Infosys Topaz service suite with the Harness software delivery platform, automating everything from code submission to deployment, with a focus on addressing governance and audit requirements in highly regulated industries (finance, energy).
For domestic enterprises, this problem also exists. After introducing AI coding tools, if CI/CD, automated testing, and permission management are not kept up to date, what accumulates is not efficiency, but risk.
Five Major AI Trends in 2026, the Third One Is Most Easily Overlooked
36Kr has compiled five directions for AI and data science in 2026, among which the third one—data quality over data volume—is the easiest to skip during project initiation.
Many enterprises focus their efforts on selecting models and building platforms when promoting AI projects. However, if the consistency, completeness, and timeliness of the data itself are not guaranteed, the results output by AI systems will only amplify these issues rather than automatically fix them.
Another noteworthy signal: enterprises' criteria for measuring AI are shifting from "whether AI is used" to "what the ROI is after using AI." When initiating projects, indicator design is even more important than technology selection itself—AI projects without clear indicators are difficult to evaluate later and challenging to advance in terms of budgeting.
The five major trends also include: shifting from pilot projects to large-scale production, the rise of agentic AI (AI transitioning from conversation to autonomous task execution), AI governance and compliance becoming mandatory, and talent strategies shifting from external recruitment to internal skill reskilling.
36Kr · April 8, 2026Weekly Signal Summary
- 1 AI agent market penetration rate is 78%, the selection window is narrowing, and "whether to do it" is no longer the question
- 2 Embedding AI into business decision-making chains yields better results than building an independent AI department
- 3 AI Speed Paradox: Code is faster but delivery isn't, downstream systems not keeping pace is the main reason
- 4 Data quality is the underlying variable for AI implementation; ROI metric design deserves priority over tool selection
Start here...
