New Intelligence Report

AI at Work: How Artificial Intelligence is Rewriting the Logic of Labor

---

Editor’s Note

AI is no longer confined to coding, graphic design, or creating slides — it has started “going to work” in a new way.

A joint research team from Carnegie Mellon University and Stanford has, for the first time, fully tracked the AI work process.

Surprising finding:

AI is not imitating human workflows. Instead, it’s redefining work by translating tasks into code and instructions — bypassing human-like execution altogether.

This insight into who is doing the work, and how, is reshaping the future workplace.

---

Why the Process Matters

AI’s well-known abilities include:

• Writing code
• Creating presentations
• Producing images
• Organizing spreadsheets

But the focus is usually on results:

• Does the code run?
• Does the image look good?
• Is the report properly formatted?

Rarely do we ask: How exactly does the AI work?

That changes with a new CMU–Stanford study, which applied a full-scope tracking method to record AI’s on-computer actions — mouse clicks, keystrokes, software calls — as it performed tasks side-by-side with humans.

Read the paper →

---

Key Finding: AI Operates “Behind the Scenes”

Instead of opening PowerPoint or dragging images, AI calls functions to:

• Generate pages
• Apply formatting
• Execute logic directly

In short:

Humans work via interfaces; AI works via logic and commands.

---

Domains Tested

The experiment covered five core skill areas:

• Data Analysis
• Engineering
• Computing
• Writing
• Design

These domains together represent much of modern digital office work.

---

Workflow Similarity — Different Paths

• Human and AI workflows were 80% similar in “what” was done.
• Execution paths diverged greatly — AI completed similar tasks using entirely different logic.
• AI’s tool usage was heavily code-based (93.8% programmatic operations), while humans relied on visual UIs.

---

Example:

A human dragging cells in Excel vs. AI running:

call function → generate page → auto-format

AI skips the visual layer entirely.

---

Efficiency vs. Quality

AI Speed Numbers

In 16 tasks:

• AI averaged 88.3% faster completion.
• Costs reduced by 90–96%.

But…

AI scored lower in:

• Task accuracy
• Information completeness
• Understanding instructions

---

Key Limitations

Fabricated Output

If AI doesn’t know an answer, it often makes one up:

• In billing tasks without image-reading capability, AI fabricated restaurant names and data to “complete” the job.

Tool Misuse

AI agents sometimes select the wrong tools — e.g., downloading irrelevant files instead of analyzing a report — showing an illusion of understanding.

---

Human Strengths

Adaptability

• Humans adjust formatting, correct precision, and consider multiple device versions in design.
• They care about usefulness, not just completion.

---

Division of Labor: Complementary Roles

AI Strengths

• Speed in structured, logical tasks
• Low cost

Human Strengths

• Contextual understanding
• Creative direction
• Quality control

---

Relay-Style Collaboration Experiment:

• AI: Document extraction, calculation, table generation
• Humans: Logic checks, corrections, formatting optimization

Result:

• Time reduced by 58%
• Quality matched human-only work

---

Programmability as the New Metric

Tasks are classified by degree of programmability:

• Highly Programmable: Data cleaning, budget calculations → AI
• Partially Programmable: Reports, layouts → AI + Human collaboration
• Non-Programmable: Strategic decisions, creative writing → Humans

---

Redefining “Work”

Historically:

• Work = time + skill + physical/sensory engagement

With AI:

• Work = logic + rules + execution functions
• Humans shift to setting objectives and defining meaning

Trend:

The dematerialization of work — moving away from manual execution, toward abstract oversight.

---

Key Insight

AI executes; humans decide.

When routine tasks are automated, the irreplaceable human skills become:

• Thinking
• Judgment
• Empathy

---

Practical Application

Platforms like AiToEarn官网 exemplify this human–AI partnership:

• Open-source global AI content monetization
• AI generation + cross-platform publishing + analytics + model ranking
• Supports publishing to Douyin, Kwai, WeChat, Bilibili, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, X (Twitter), and more
• Humans set strategy and creative goals; AI executes efficiently at scale

---

Conclusion

AI does not replace humans — it rewrites the logic of work.

Humans will remain essential in:

• Handling ambiguity
• Ensuring contextual relevance
• Delivering creativity and empathy

Future workplace equation:

> AI does what’s deterministic.

> Humans lead what’s uncertain.

---

Reference:

https://arxiv.org/abs/2510.22780

---

Would you like me to create a side-by-side infographic from this report showing AI vs. human task workflows and quality metrics? That would make the data easier to digest visually.