There are many articles about AI transforming work. Few focus on specific operating patterns. This piece covers actionable workflows that knowledge workers have found reliably effective across common job functions — the goal is concrete integration, not general possibility.
## Research and Information Processing
**The problem**: reading and synthesizing large volumes of articles, reports, and emails takes time that could go elsewhere.
**The workflow**:
1. **Perplexity AI** (or ChatGPT with search) for initial orientation: ask the research question, get a cited synthesis, establish the landscape fast.
2. **Claude** for long-document processing: upload complete PDFs or reports and ask for extraction of key findings, data points, and conclusions. The 200K context window handles 100+ page documents without chunking.
3. **Standardized summary format**: use a fixed summary prompt (“three most important findings + one follow-up question + key data points”) for batch document processing. Building a personal knowledge base becomes systematic.
Reported gains from systematic AI research assistance: 60–70% reduction in reading and synthesis time.
## Writing and Content Creation
**The workflow**:
1. **Outline first**: describe the piece’s purpose, audience, and main argument; ask for multiple outline options; choose and adjust.
2. **Section-by-section drafting**: collaborate on one section at a time rather than generating the whole piece. Produces better output and easier iteration.
3. **Revision loop**: return drafts with specific improvement criteria (“make the language more concise,” “add data support for the third argument,” “adjust to a more formal business register”).
4. **Style guide prompt**: for regular publications (blogs, reports), maintain a style guide prompt that is prepended to each generation request to keep voice consistent.
## Code and Technical Work
**The workflow**:
1. **Cursor** (AI IDE) with Claude or GPT-4: the whole codebase as context makes questions and modifications far more precise than pasting snippets into a chat.
2. **Debugging**: paste the error message + relevant code + your hypothesis together. Getting an answer takes seconds versus minutes of manual trace.
3. **Code review**: run AI review for security, performance, and readability before human review — as a preliminary filter, not a replacement.
4. **Documentation**: generate docstrings, README sections, and API documentation drafts automatically; human review for accuracy.
## Meetings and Communication
**The workflow**:
1. **Transcription + summary**: Otter.ai or Notion AI transcribes meetings; AI extracts action items and decisions.
2. **Email analysis before reply**: for complex emails, first ask the AI “what is this person’s core request?” before drafting a response. Reduces misunderstandings.
3. **Template library**: build AI-assisted templates for common communication patterns (change notifications, status updates, client refusals) to eliminate starting from scratch each time.
## Learning New Skills
**The workflow**:
1. **Personalized learning plan**: describe your existing knowledge and target skill; ask for a 4-week or 8-week plan with specific daily tasks.
2. **Immediate concept clarification**: use Feynman-style prompts (“explain X in simple terms, then give me an analogy”) for concepts you’re unclear on.
3. **Custom practice generation**: have AI generate exercises and mini-projects specific to what you’re learning at each stage.
## Principles That Apply Everywhere
Specify output format explicitly rather than letting AI guess. Treat AI output as a starting point, not a final product. Retain critical judgment — especially for data and factual claims. Build a reusable prompt library to reduce per-task design time.
For related reading, see [Prompt Engineering](https://sunqi.org/llm-prompt-engineering-en/) and [Claude AI Capabilities](https://sunqi.org/claude-ai-capabilities-en/).
