Built as an interactive training module by TARAhut AI Labs
Generative AI & Prompt Engineering · Session 09: Role-Based Prompting & Multi-Step Workflows · tarahutailabs.com
© 2026 TARAhut AI Labs. All rights reserved.
Assign AI different roles within a single workflow. Build a complete content production pipeline where one AI acts as researcher, writer, editor, and publisher. "Ek AI nu multiple roles do — poori team da kaam ek prompt naal."
In Session 8, you built single personas. Now you will assign MULTIPLE roles to AI within a single workflow. The same AI becomes a researcher, then a writer, then an editor, then a publisher — each with different rules and perspectives. "Ikko AI — 4 roles — poori team replace."
Last session you built 3 permanent AI personas and stress-tested them. Today we take it further: instead of one persona per conversation, you will switch roles within a single workflow. Think of it as directing a play where one actor plays multiple characters. "Pichle session vich ek persona banaya — aaj ek workflow vich 4 roles."
"Write a blog post about AI in Punjab agriculture." — You get one perspective, one draft, and hope it is good enough. No review, no quality check, no different angles. Like having one employee do research, writing, editing, and publishing alone.
"Step 1: As a researcher, gather key facts. Step 2: As a writer, draft the article. Step 3: As an editor, critique and improve. Step 4: As a publisher, format for distribution." Each role adds a quality layer. The output is exponentially better.
Mission: Gather facts, data, and insights. Personality: Thorough, skeptical, source-focused. Output: Structured research brief with key findings, statistics, and source credibility notes. "Researcher di nazar — facts first, opinions baad vich."
Mission: Transform research into compelling content. Personality: Creative, engaging, audience-aware. Output: Draft content with storytelling, hooks, and clear structure. Takes research input and makes it human-readable. "Writer — boring facts nu interesting story banaunda hai."
Mission: Critique, improve, and polish. Personality: Critical, detail-oriented, reader-first. Output: Tracked changes, improvement suggestions, fact-check flags, and tone adjustments. The editor finds what the writer misses. "Editor — writer di galti labh ke fix karda hai."
Mission: Format and optimize for distribution. Personality: Platform-aware, SEO-focused, conversion-minded. Output: Final formatted content with metadata, headlines, social media snippets, and distribution notes. "Publisher — content nu duniya tak le ke jaanda hai."
When you ask AI to "write a blog post," it tries to research, write, edit, and format ALL AT ONCE. The result is mediocre at everything. When you separate roles, each step gets the AI's full attention. The researcher does not worry about style. The writer does not worry about SEO. Each role does ONE thing excellently. "Ek kaam ek vaari — quality 10x vaddh jaandi hai."
"Main pehla ek prompt vich sab kuchh mangda si — result mediocre aunda si." — That is the single-prompt trap. Multi-role workflows break complex tasks into steps where each step has a SPECIALIST perspective. The compound quality gain is dramatic. This is how content agencies work — you are building a one-person agency with AI.
Test both approaches on the same topic:
Next: Build the complete content production pipeline step by step.
Build a 4-stage content production pipeline for a Punjab business. Each stage uses a different AI role with specific instructions. The output of each stage feeds into the next. "Har stage da output agle stage da input ban jaanda hai."
Start by generating a research brief. Copy and paste this prompt:
Now switch the AI role to writer. Paste the research brief from Stage 1 along with this prompt:
Switch to editor role. Paste the draft from Stage 2:
Final stage — format for distribution. Paste the edited article:
A typical content agency charges ₹15,000-₹50,000 for this package (research + article + social media + SEO). You just built the same pipeline with 4 prompts. The cost? About ₹5-10 in API tokens, or free if using ChatGPT/Claude free tiers. This is the democratization of professional content production. "Char prompts = ek poori content team."
Next: Apply multi-step workflows to other business operations.
Content is just the beginning. Multi-role workflows transform every business function: sales, HR, product development, and customer success. "Content sirf shuruat hai — har business function vich multi-role workflow laga sakde ho."
Role 1: Market Analyst — Research the prospect, their industry, competitors, and pain points. Role 2: Solution Architect — Design a tailored solution matching their needs. Role 3: Proposal Writer — Draft the proposal with ROI calculations. Role 4: Reviewer — Check pricing, commitments, and risk language. "Sales proposal 4 steps vich — customized, professional, risk-checked."
Role 1: Job Analyst — Define role requirements, skills matrix, and salary benchmarks for Punjab market. Role 2: JD Writer — Create an engaging job description that attracts Punjab talent. Role 3: Screening Expert — Design interview questions targeting specific competencies. Role 4: HR Reviewer — Check for bias, legal compliance, and inclusivity.
Role 1: Market Researcher — Analyze market size, competition, and customer needs in Punjab. Role 2: Product Strategist — Define features, pricing, and positioning. Role 3: Launch Planner — Create timeline, marketing plan, and distribution strategy. Role 4: Risk Analyst — Identify potential failures and mitigation plans.
Role 1: Account Analyst — Review customer history, usage patterns, and satisfaction signals. Role 2: Success Strategist — Design retention and upsell plan. Role 3: Communication Writer — Draft personalized outreach (check-in emails, QBR decks). Role 4: Manager Review — Verify tone, accuracy, and strategic alignment.
Build a complete sales proposal for a Punjab IT services company:
The biggest mistake in multi-role workflows is skipping the reviewer. Always include a critical review role as the final step. The reviewer catches: over-promising, factual errors, tone mismatches, legal risks, and audience misalignment. It is the quality gate that makes the difference between amateur and professional output. "Review role kabhi skip na karo — ehi professional aur amateur da farak hai."
Next: Quality control techniques that make multi-step workflows reliable.
Multi-step workflows are powerful but can go wrong. Quality gates, role contracts, and feedback loops ensure consistent output. "Power de naal responsibility vi aundi hai — quality gates lagao."
Every role must have a clear "contract" — what it receives as input, what it must produce as output, and what quality standards apply. If the research brief is thin, the writer cannot produce a great article. Garbage in, garbage out applies to every stage. "Har role da contract clear hona chahida hai."
Define EXACTLY how output transfers between roles. Use consistent formats: "Research output must be in bullet points with [VERIFY] flags. Writer output must be in paragraphs with [SUBHEADING] markers." Structured handoffs prevent information loss between stages.
After the final stage, ask AI: "Review this entire workflow output. Are there any contradictions between the research findings and the article claims? Any inconsistencies in numbers, names, or facts?" This catches drift that happens across multiple stages.
Add a validation step: "Read this output as if you are a [TARGET AUDIENCE]. Would you find this useful? What questions would you still have? What would make you stop reading?" This catches content that is technically correct but audience-wrong.
The most powerful gate: take the editor/reviewer feedback and SEND IT BACK to the writer for revision. "Revise the article based on this editorial feedback: [PASTE FEEDBACK]." This iterative loop is how professional publishing works. Two rounds of revision transform good into excellent.
Run this after any multi-step workflow to catch errors:
Next: Quiz time! Prove your multi-role workflow mastery.
8 questions picked randomly from a pool of 20. Advanced-level questions about role-based prompting, content pipelines, and workflow quality. "Har sawaal tuhadi samajh test karda hai."
Next: Your homework and Session 10 preview.
"Tusi aaj ek poori content production team bana li — 4 AI roles naal." Here is what you learned and what comes next.
✅ Single-role vs multi-role workflow architecture
✅ The 4 role types: Researcher, Writer, Editor, Publisher
✅ Complete content production pipeline with 4 sequential stages
✅ Multi-role templates for Sales, HR, Product, and Customer Success
✅ The 5 quality gates: Role contracts, Handoff formats, Contradiction checks, Audience validation, Feedback loops
✅ QA Analyst role for cross-stage consistency checking
"Practice naal hi mastery aundi hai!"
"Kal assi prompt chaining sikhange — 5 linked prompts jo automatic pipeline banaunde ne. Research → Outline → Draft → Polish → Distribute. Ek chain jo poora workflow ek flow vich chala dende. From manual multi-step to automated chains."