AI-Augmented Workflows

The productivity gain from AI doesn't come from asking it a question and copying the answer. It comes from redesigning how you work so that AI handles the parts it's good at (generating drafts, transforming formats, searching for patterns) while you focus on the parts it's bad at (judgment, strategy, knowing what matters).

This article covers concrete workflow patterns for four common IC (individual contributor) activities. For the tools themselves, see [Generative AI Tools for Individuals](GenerativeAiToolsForIndividuals). For how to prompt effectively within these workflows, see [Practical Prompt Engineering](PracticalPromptEngineering).

Principle: AI as Force Multiplier, Not Replacement

The workflow that doesn't work: "AI, do my job." → Bad output → Spend more time fixing than doing it yourself.

The workflow that works: "I'll do the thinking. AI does the typing." → Good output in a fraction of the time.

In every workflow below, you provide the intent, context, and quality judgment. The AI provides speed, breadth, and the ability to generate options you wouldn't have considered.

Coding Workflows

The Implementation Sprint

**Without AI**: You know what function you need. You write it line by line, look up syntax, debug typos. 30 minutes.

**With AI**: You describe what you need in natural language. The AI generates the function. You review, test, adjust. 5 minutes.

```

Prompt: "Write a Python function that takes a CSV file path and returns a

dictionary where keys are the values in the first column and values are

lists of the remaining columns. Handle the header row. Use csv.DictReader.

Include type hints and a docstring."

```

The AI generates 90% correct code. You adjust the edge case it missed (empty rows). Net time saved: 80%.

The Test Generation Pattern

**The workflow**: Write the implementation first. Then ask AI to generate tests.

```

Prompt: "Here is my function [paste code]. Write pytest tests covering:

- Normal operation with valid input

- Empty input

- Malformed input (missing columns, encoding issues)

- Edge cases you can identify from the implementation

Use parametrize where appropriate."

```

AI-generated tests are excellent scaffolding. They cover the obvious cases quickly and often catch edge cases you didn't consider. You add the domain-specific test cases that require business knowledge.

The Rubber Duck Debugger

**The workflow**: Paste your error, your code, and what you've already tried. The AI acts as an informed rubber duck.

```

Prompt: "I'm getting this error [paste traceback]. Here's the relevant code

[paste code]. I've already checked that the file exists and the permissions

are correct. What else could cause this?"

```

This is often faster than Stack Overflow because the AI sees your specific code, not a generic example. But verify its suggestions — the AI can be confidently wrong about runtime behaviour.

The Code Review Partner

```

Prompt: "Review this code for: bugs, security issues, performance problems,

and readability. Be specific — point to exact lines. Don't mention style

nits unless they affect readability."

```

AI code review catches real issues — null pointer risks, SQL injection, race conditions — that human reviewers sometimes miss because they're focused on the logic. Use it as a complement to human review, not a replacement.

Writing Workflows

The Structured First Draft

**The workflow**: You create the outline (the thinking). AI creates the first draft (the typing). You edit it into your voice.

1. **You**: Write 5-10 bullet points of what you want to cover

2. **AI**: "Expand these bullets into a 1,000-word article. Use a professional but conversational tone. Each point should be a section with a clear heading."

3. **You**: Read the draft. Delete the generic parts. Rewrite the sections that need your voice or expertise. Add specific examples from your experience.

The AI's draft is a 60% solution in 60 seconds. Your editing makes it a 95% solution in 20 minutes. Total: 21 minutes instead of 90.

The Compression Pattern

**The workflow**: You have something too long. AI condenses it.

```

Prompt: "Reduce this 3,000-word report to a 500-word executive summary.

Keep all specific numbers, dates, and recommendations.

Remove background context, caveats, and hedging language.

Use bullet points for the recommendations section."

```

AI compression is excellent because it doesn't have the emotional attachment to its own words that humans do. It cuts ruthlessly. You review to ensure nothing critical was lost.

The Multi-Version Generator

**The workflow**: You need the same content in different formats.

```

Prompt: "Here's my technical blog post. Create these versions:

A 280-character tweet

2. A 3-sentence LinkedIn post

3. A 5-bullet executive summary

4. A Slack message to my team summarising the key finding"

```

One piece of content, four formats, two minutes. Without AI, this is 30 minutes of context-switching between different writing modes.

Research Workflows

The Landscape Survey

**The workflow**: You need to understand a new field quickly.

1. **Ask for structure**: "What are the 5-7 major sub-topics within [field]? For each, give me the key concepts and the most important tensions or debates."

2. **Go deeper selectively**: "Explain [sub-topic 3] in more detail. What are the current best practices and common mistakes?"

3. **Verify independently**: Use Perplexity or direct source search to verify the key claims. The AI gives you the map; you verify the territory.

**Critical warning**: AI is a research *accelerator*, not a research *source*. It gives you the right questions to ask and a framework for organising answers. The answers themselves must be verified against primary sources.

The Document Analysis Pattern

**The workflow**: You have a long document (contract, specification, research paper) and need to extract specific information.

```

Prompt: "Here is a 40-page vendor contract. Extract and summarise:

Payment terms and amounts

2. Termination clauses and notice periods

3. Liability limitations

4. Any auto-renewal provisions

5. Anything unusual or unfavourable compared to standard contracts

Quote exact language from the contract for each point."

```

The "quote exact language" instruction is key — it forces the model to ground its answers in the actual text rather than generating plausible-sounding summaries.

The Comparative Analysis

```

Prompt: "Compare these 3 approaches to [problem]:

[Approach A description]

[Approach B description]

[Approach C description]

Create a comparison table with rows: implementation complexity, cost,

scalability, maintenance burden, risks, best suited for.

Then recommend one with reasoning."

```

Analysis Workflows

The Data Interpretation Pattern

```

Prompt: "Here is a CSV of monthly sales data for the past 3 years

[paste data]. Identify:

Overall trend (growing, declining, flat)

2. Seasonal patterns

3. Any anomalies or outliers

4. Three possible explanations for the biggest anomaly

Write Python code to generate visualisations for each finding."

```

Notice: you're asking the AI to write code that does the analysis, not to do the analysis directly. This leverages the AI's strength (code generation) while avoiding its weakness (math).

The Decision Framework Generator

```

Prompt: "I need to decide between [Option A] and [Option B] for [context].

My priorities are [list 3-5 priorities].

Create a weighted decision matrix with these priorities as criteria.

Rate each option 1-5 on each criterion.

Show the weighted totals.

Then argue for the option that lost — what am I missing?"

```

The final instruction — "argue for the loser" — is the most valuable part. It forces consideration of perspectives you've already dismissed.

Anti-Patterns: When AI Slows You Down

| Situation | Why AI Hurts | Do This Instead |

|-----------|-------------|----------------|

| You know exactly what to write/code | AI generation + review takes longer than just doing it | Just do it. AI is for when you're stuck or need volume. |

| The task requires deep domain expertise | AI output will be generic and require heavy editing | Use AI for structure/formatting only, write the substance yourself. |

| You need guaranteed accuracy | AI will be confidently wrong about specific facts | Use traditional sources, then use AI to format/organise your verified findings. |

| Quick one-line tasks | Context-switching to AI takes longer than the task | If it takes less than 2 minutes manually, just do it. |