Four real workflows — how AI is actually used in research, coding, content, and business. Not what AI can do. What you should do with it.
6 min read Applied Pick your lane
Workflow · Research
AI-powered research
AI-powered research means systematically using AI to gather, cross-reference, and synthesize information faster than you could alone. It is not asking one question and trusting whatever comes back. Ask the same question to two different models, compare their answers, and dig deeper wherever they disagree.
1
Define the question clearly
Vague research questions produce vague results. "Tell me about AI" is useless. "Compare the inference costs and latency of Claude Opus vs GPT-4o vs DeepSeek V3 for a high-volume API application processing 10K requests/day" is actionable.
2
Multi-source gathering
Use multiple AI models or tools. Ask the same question to different models and compare answers. If they agree, confidence is high. If they disagree, dig deeper.
3
Cross-verify claims
Never trust a single AI output for factual claims. Ask for sources. Check those sources. Use AI to generate leads, not conclusions.
4
Synthesize, don't summarize
Don't just collect AI responses — combine them into a coherent analysis. Identify patterns, contradictions, and gaps.
5
Document as you go
Save key findings, sources, and decisions in a structured format. Your future self will thank you.
Pro tips:
Use a model with web search for current events
Use a reasoning model (o1, o3, Claude Opus) for complex analysis
Use a fast model for initial exploration, switch to powerful model for deep dives
Always ask: "What would contradict this conclusion?"
Workflow · Coding
AI-assisted coding
AI-assisted coding is the highest-leverage use of AI for developers. But there's a right way and a wrong way to do it.
✗ The wrong way
"Build me a full-stack app with auth, database, API, and deploy it"
→ AI generates a massive amount of code → half doesn't work → you spend more time debugging than if you wrote it yourself.
✓ The right way
Plan first. Generate small pieces. Test immediately. Use AI for the boring parts, your brain for architecture.
→ Correct, maintainable, actually-shipped code.
1
Plan first
Break the project into small, specific tasks. "Create a FastAPI endpoint that accepts a username and returns their last 10 login timestamps."
2
Generate small pieces
Ask AI to generate one function, one component, one file at a time. Small outputs are more likely to be correct.
3
Test immediately
Don't accumulate untested code. Generate, test, fix, move on.
4
AI for the boring parts
Boilerplate, CRUD operations, data transformations, test cases — let AI handle the repetitive work.
5
Your brain for architecture
System design, data modeling, security decisions — these need human judgment. AI is a tool, not an architect.
Pro tips:
Provide context about your existing codebase (file structure, tech stack, conventions)
Ask AI to explain its code, not just generate it
Use "think step by step" for complex logic
When stuck: paste the error message and ask AI to diagnose, not just fix
Workflow · Content
The content workflow
AI can draft content 10x faster than you can write it. But raw AI output is almost never publishable. The workflow is: AI drafts, human edits.
1
Define the brief
Topic, audience, tone, length, key points. The more specific your brief, the less editing you'll need.
2
AI generates the first draft
This is the fast part. AI handles structure, flow, and initial content.
3
Human adds voice and expertise
The important part. Add your personal experience, specific examples, contrarian takes, genuine opinions. AI can't do this — it generates generic content by default.
4
Fact-check everything
AI will generate plausible-sounding "facts" that are completely wrong. Verify every claim, statistic, and source.
5
Remove AI-isms
Watch for: "In the ever-evolving landscape...", "It's worth noting that...", "In conclusion...", excessive hedging, and em dashes. Strip them. Replace with direct language.
The anti-slop checklist: Does this sound like it could only have been written by YOU? Are there specific examples from YOUR experience? Would YOU want to read this? Does it have a clear opinion, or is it wishy-washy? If the answer to any is "no," edit more.
Workflow · Business
Business automation
The highest-ROI use of AI in business is automating repetitive cognitive tasks. Chatbots get the headlines, but the real money is in the boring work: the four hours a day someone spends on data entry that an AI draft plus a 30-minute human review can replace.
The identification framework: for every task you do repeatedly, ask.
?
Creative judgment?
→ Human
≡
Follows a pattern?
→ AI candidate
▢
Structured in, predictable out?
→ Automate
✓
Has a "review by human" step?
→ AI draft + human approve
Common automations:
Task
AI Role
Human Role
Email responses
Draft replies
Review and send
Data analysis
Process and summarize
Interpret and decide
Report generation
Compile and format
Add insights
Customer support
Handle common questions
Escalate complex issues
Document processing
Extract and classify
Validate edge cases
Meeting summaries
Transcribe and summarize
Verify action items
Implementation pattern:
1
Start with one process
Don't try to automate everything at once. Pick the task that wastes the most time.
2
Measure the baseline
How long does it take manually? How many per week?
3
Build the simplest version
AI + a simple script. No complex infrastructure.
4
Measure the result
Time saved? Quality maintained? New issues?
5
Iterate and expand
Fix issues, then move to the next process.
The key insight: AI automation doesn't replace people. It replaces repetitive tasks. The person who used to spend 4 hours on data entry now spends 30 minutes reviewing AI-generated entries. That's 3.5 hours freed up for work that actually requires human thinking.
