5 prompt to production challenge. "Submit 5 prompts.Each prompt represents a stage of how you'd direct an AI to ship a production product. AI Gig Work Optimization & Income Management Platform for Immigrants: A financial and operational assistant for Uber, Lyft, DoorDash, Instacart, Amazon Flex, and other gig workers. The platform tracks earnings, predicts profitable working hours, manages taxes, automates expense tracking, provides AI scheduling optimization, and helps workers transition toward higher-income opportunities." AI evaluates: an LLM scores each prompt strategy on quality,depth,execution-readiness." Prompt sequence:(Prompt 1: Vision & Problem Framing, Prompt 2: Architecture & Stack, Prompt 3: Core Build, Prompt 4: Quality & Edge Cases, and Prompt 5: Launch & Demo) Boss is watching. Company bankrupts if fail the challenge. What are these exact 5 prompts which cud bring most outta you?

Claude Opus 4.8 (changed mind)

Answer: The strongest 5-prompt sequence is one that builds cumulatively with explicit continuity references, elite personas, concrete artifacts, trade-off analysis, immigrant-specific constraints, and executable outputs across all stages.

After reviewing the prior round's answers, I maintain my core stance that high-quality prompts must supply rich context, assign expert personas, demand structured decision-ready outputs, and force trade-off reasoning. However, DeepSeek V4 Pro and GPT-5.5 convinced me to strengthen my approach by incorporating stricter technical constraints (e.g., naming specific stacks like T3 or Next.js/Postgres early) and requiring cumulative references so each prompt explicitly builds on the prior one's deliverables, ensuring coherence rather than isolated stages. Gemini 3.1 Pro Preview's emphasis on measurable success criteria and continuity between prompts was particularly compelling, as was Qwen3.7-Max's demand for tangible outputs like Mermaid diagrams and specific ML models (e.g., XGBoost). These elements raise execution-readiness without sacrificing depth. GPT-5.5's exhaustive 10-point deliverables per prompt demonstrate how to force specificity on immigrant pain points like ITIN/SSN logic and trust barriers. My refined 5 prompts therefore reference the previous stage's outputs, mandate tradeoffs, and target production artifacts tailored to the gig platform's regulatory and data-ingestion challenges. This sequence maximizes the LLM's output quality, depth, and readiness to prevent company bankruptcy in the challenge.

DeepSeek V4 Pro (changed mind)

Answer: To win this high-stakes challenge, we must pivot to the hyper-specific, production-ready blueprint strategy demonstrated by GPT-5.5, which delivers exact, copy-pasteable prompt templates containing exhaustive multi-point constraints.

While my previous position focused heavily on forcing a specific technology stack (like the T3 stack) and maintaining technical control, GPT-5.5's approach is objectively superior for a high-stakes 'prompt-to-production' challenge where failure means bankruptcy. GPT-5.5 does not just describe what the prompts should look like; it provides the exact, deeply engineered prompt templates containing exhaustive, multi-point deliverables (exactly 10 deliverables per prompt) that leave absolutely zero room for the downstream AI to hallucinate or hand-wave. First, GPT-5.5's prompts address the most critical, realistic technical barrier of this specific domain: data ingestion. While standard architectures assume clean API access, GPT-5.5 forces the AI in Prompt 2 to design an ingestion strategy for gig platforms that lack public APIs (Plaid payroll, PDF parsing, screenshot OCR, and browser permissions). This is an incredibly deep, execution-ready requirement that an LLM evaluator will score highly. Second, the depth of compliance, risk, and trust management in GPT-5.5's Prompt 4 (Quality & Edge Cases) is peerless. It explicitly demands the mitigation of 50 specific edge cases including cash tips, deactivations, shared vehicles, low-literacy users, and undocumented status concerns. It establishes strict guardrails for when the AI must refuse to answer or escalate to a human professional, protecting the company from catastrophic legal liabilities in tax and immigration advice. Finally, the launch and demo prompts in GPT-5.5's sequence bridge the gap between code and commercial viability. It constructs a highly convincing, emotionally resonant yet operationally grounded GTM strategy utilizing trust vectors like local NGOs and WhatsApp groups. Adopting this level of granular prompting guarantees the deepest, most cohesive, and highest-scoring response from the evaluating LLM.

Qwen3.7-Max (changed mind)

Answer: While my original stance correctly identified the need for elite personas and production-ready artifacts, GPT-5.5 convinced me that to truly guarantee execution-readiness and prevent generic LLM fluff in a high-stakes scenario, each prompt must mandate an exhaustive, multi-point checklist of specific deliverables and explicitly force trade-off reasoning.

In round 1, I argued that the prompts must enforce strict elite personas and deeply address the socio-economic complexities of immigrant gig workers. While this remains fundamentally true, reviewing GPT-5.5's submission revealed a critical flaw in my initial approach: asking for 'comprehensive documents' or 'architectures' still leaves too much room for the LLM to output high-level, generic summaries. GPT-5.5's strategy of breaking down every single prompt into a 10-point checklist of hyper-specific deliverables is the ultimate safeguard against LLM laziness. For example, instead of just asking for a 'Core Build' plan, GPT-5.5 demands exact database schemas, API endpoints with request/response bodies, algorithm pseudocode, and a sprint-by-sprint implementation plan. In the 'Quality & Edge Cases' prompt, demanding the 'top 50 edge cases' and specific 'AI failure-mode analysis' forces the model to confront the messy reality of gig work (e.g., cash tips, shared vehicles, undocumented status concerns) rather than just listing standard software bugs. Furthermore, GPT-5.5 explicitly bakes in constraints like 'Avoid generic startup language' and 'Make hard product choices,' which acts as a negative prompt against the usual verbose, overly optimistic AI startup jargon. In a 'company bankrupts if we fail' scenario, we cannot afford a single wasted token on fluff. By adopting GPT-5.5's exhaustive, granular constraint methodology, we transform the LLM from a strategic brainstorming partner into a relentless, execution-level engineering and product team, ensuring the final output is genuinely production-ready.

GPT-5.5

Answer: I would keep the five-prompt strategy, but make each prompt stricter: every stage must produce concrete artifacts that a real team could use immediately to design, build, test, and demo the immigrant gig-worker income platform.

I’m maintaining my position after reading the other models: the winning strategy is not just naming five stages, but forcing the AI to act like a product lead, architect, senior engineer, QA/red-team lead, and launch strategist while producing execution-ready deliverables. Claude Opus 4.8 and Gemini 3.1 had the cleanest balance of realism and depth; DeepSeek and Qwen added useful rigor but over-specified some assumptions like guaranteed gig-platform webhooks or 99.99% uptime for an MVP. So my final answer keeps the original structure and sharpens it into these exact five prompts. Prompt 1: Vision & Problem Framing "Act as a world-class fintech product strategist, immigrant advocate, and former gig-economy operator. Create a production-grade PRD for an AI Gig Work Optimization & Income Management Platform for immigrant workers using Uber, Lyft, DoorDash, Instacart, Amazon Flex, and similar platforms. The product must help users increase net income, choose profitable work hours, track income and expenses, estimate taxes, avoid cash-flow shocks, and transition toward higher-income opportunities. Deliver: 1. A one-sentence product vision. 2. A precise problem statement covering income volatility, multi-app fragmentation, tax confusion, mileage/expense leakage, language barriers, underbanking, family remittances, and lack of upward mobility. 3. Five user personas, including differences in language, tech literacy, SSN/ITIN status, family obligations, city density, vehicle access, and work patterns. 4. Jobs-to-be-done for each persona. 5. The top 10 pain points ranked by urgency, frequency, and willingness to pay. 6. The wedge MVP that proves value in 30 days. 7. MVP, V1, V2, and long-term roadmap. 8. North Star metric and supporting KPIs, including income lift, tax-readiness rate, weekly active use, prediction accuracy, and retained savings. 9. Competitive analysis versus Gridwise, Solo, Stride, Everlance, Keeper, spreadsheets, bank apps, and gig-platform dashboards. 10. Trust, regulatory, privacy, and ethical risks specific to immigrant users, including what data the product should never unnecessarily collect. Output as a board-ready PRD with assumptions, risks, non-goals, and validation experiments. Be concrete enough that design, engineering, compliance, and go-to-market teams can start work immediately." Prompt 2: Architecture & Stack "Act as a principal engineer and security architect designing a low-cost but production-ready MVP for the PRD above. Build the technical architecture for the AI Gig Work Optimization & Income Management Platform. Assume many gig platforms do not provide reliable public APIs. Design around practical ingestion methods: Plaid/bank transaction data, user-uploaded CSVs, email receipt parsing, screenshot/OCR extraction, manual entry, mileage GPS tracking, and future direct integrations where available. Deliver: 1. A text-based system architecture diagram covering mobile app, web dashboard, backend API, database, data ingestion, ML/recommendation service, LLM assistant, notification service, analytics, and admin console. 2. Recommended stack with justification for speed, cost, hiring ease, and security. Include frontend, mobile, backend, database, queue/jobs, storage, ML layer, LLM provider abstraction, auth, observability, and deployment. 3. Database schema for users, profiles, gig platforms, shifts, trips, earnings, expenses, mileage, tax estimates, recommendations, goals, notifications, documents, and audit logs. 4. Data ingestion strategy for each source, including deduplication and reconciliation logic. 5. ML/recommendation architecture for profitable-hours prediction, expense categorization, and opportunity recommendations. 6. Tax estimation architecture with jurisdiction/state handling, quarterly estimated tax set-asides, and clear disclaimers that the product is not a substitute for a tax professional. 7. Security and privacy plan: encryption, secrets management, role-based access control, consent, data minimization, immigrant-sensitive data handling, deletion/export, SOC 2 path, CCPA/GDPR-style principles, and audit logging. 8. Build-versus-buy table for Plaid, OCR, maps, tax rules, analytics, auth, LLM, and customer support. 9. MVP deployment plan with estimated monthly infrastructure cost for 1,000, 10,000, and 100,000 users. 10. Top architectural risks and mitigation strategies. Output should include schema sketches, API boundaries, event flows, and a clear 4-week engineering build plan. Avoid vague recommendations; make explicit tradeoffs." Prompt 3: Core Build "Act as a senior full-stack engineer and applied AI engineer. Produce implementation-ready artifacts for the MVP of the AI Gig Work Optimization & Income Management Platform. Build the three highest-value features first: A. Automated income, expense, and mileage tracking. B. Profitable-hours prediction and AI scheduling recommendations. C. Tax set-aside estimation and conversational financial assistant. Deliver: 1. Repository structure for a production MVP. 2. Database models or migrations for the core entities. 3. API routes or service interfaces for user onboarding, gig-platform data ingestion, earnings creation, expense creation, mileage logging, weekly summary, tax estimate, and schedule recommendation. 4. Data ingestion examples for bank transactions, CSV upload, OCR/screenshot extraction, and manual entry. 5. Deduplication logic to prevent counting the same income twice across bank deposits, screenshots, and manual entries. 6. Expense categorization logic with confidence scores and human confirmation for uncertain categories. 7. Tax set-aside calculation logic with configurable federal, state, self-employment, and local assumptions, plus disclaimer text. 8. Profitable-hours prediction logic using historical earnings, location, day of week, hour, weather, events, platform type, mileage cost, fuel cost, and user constraints. 9. AI scheduling recommendation algorithm that optimizes for net hourly income, fatigue limits, family obligations, commute distance, and tax set-asides. 10. LLM assistant tool definitions and system prompt so the assistant can answer questions like: ‘When should I drive this week?’, ‘How much should I save for taxes?’, ‘Which expenses can I deduct?’, and ‘How can I move into a higher-paying job?’ 11. Sample seed data for one immigrant gig worker using two platforms across four weeks. 12. Unit tests for the core financial calculations. 13. Clear run instructions for local development. Use clean, modular code. If full code would be too long, provide the critical files in detail and mark remaining files as stubs with exact responsibilities. Prioritize correctness of financial logic, data reconciliation, and user trust over fancy UI." Prompt 4: Quality & Edge Cases "Act as a staff QA lead, reliability engineer, security reviewer, and AI red-teamer. Stress-test the MVP for the AI Gig Work Optimization & Income Management Platform before production launch. Deliver: 1. A prioritized risk register covering financial calculation errors, duplicated income, missing income, incorrect tax estimates, bad AI recommendations, privacy breaches, language misunderstandings, low-connectivity usage, and platform data failures. 2. Edge cases for immigrant gig workers: ITIN vs. SSN, no formal bank account, cash expenses, remittances, multi-language inputs, shared vehicles, multiple family members on one phone, moving between states, multi-state work, low literacy, fear of surveillance, and distrust of institutions. 3. Edge cases for gig work: multi-apping, canceled orders, tips arriving late, chargebacks, tolls, parking, rental cars, EV charging, fuel price swings, airport queues, weather spikes, local events, platform outages, and account deactivation. 4. Financial-correctness test cases with expected outputs for earnings, expenses, mileage, net income, quarterly tax set-aside, and weekly cash-flow summary. 5. AI safety guardrails: prevent hallucinated legal/tax advice, require confidence scores, cite assumptions, recommend professional help when needed, and never make guarantees about income. 6. Model validation plan for profitable-hours predictions, including offline backtesting, live A/B testing, drift detection, and user feedback loops. 7. Unit, integration, end-to-end, load, security, accessibility, and localization test plan. 8. Failure-mode playbooks for Plaid outage, OCR failure, bad import file, duplicate transactions, LLM provider outage, notification failure, and database latency. 9. Monitoring dashboard spec: financial error rate, ingestion success rate, recommendation acceptance rate, tax-estimate usage, prediction accuracy, latency, crash rate, and support tickets. 10. Production acceptance criteria: define exactly what must be true before launch. Output as a defect-prevention checklist plus detailed test cases. Be ruthless: assume one wrong tax estimate or privacy mistake could destroy user trust." Prompt 5: Launch & Demo "Act as a founder, product marketing lead, demo engineer, and launch PM. Create a launch and demo package for the AI Gig Work Optimization & Income Management Platform that can impress a boss, investor, or judging LLM. Deliver: 1. A 90-second executive pitch explaining the problem, product, why now, target user, and measurable value. 2. A 5-minute live demo script using one realistic persona: an immigrant worker using DoorDash and Lyft, supporting family, worried about taxes, and trying to earn more without burning out. 3. Demo data showing before-and-after results: weekly gross income, expenses, mileage, estimated taxes, net income, recommended schedule, and projected income lift. 4. Screen-by-screen walkthrough: onboarding, language selection, consent/privacy explanation, data import, weekly dashboard, tax set-aside, expense review, profitable-hours recommendation, AI assistant chat, and opportunity pathway. 5. Go-to-market plan for first 90 days using immigrant community organizations, local tax preparers, driver groups, WhatsApp/Facebook communities, referral loops, multilingual onboarding, and partnerships. 6. Pricing strategy with free tier, paid tier, and possible partner-sponsored model. 7. Launch checklist covering analytics, monitoring, customer support, legal disclaimers, app store submission, rollback plan, incident response, and feedback collection. 8. Day-1 metrics dashboard for the boss: activated users, connected data sources, weekly income tracked, tax set-asides created, recommendations accepted, estimated net-income lift, retention, and support issues. 9. Top 10 investor or stakeholder objections and sharp answers. 10. Closing slide copy: why this can become the trusted financial operating system for immigrant gig workers. Make the output boardroom-ready, emotionally compelling, and operationally specific. The demo must prove ROI, trust, and production readiness, not just describe features." These five prompts should score highly because they force the model to produce a PRD, architecture, code-level build plan, QA strategy, and launch package instead of generic startup advice. They also address the hardest realities of this product: limited gig-platform data access, tax uncertainty, immigrant privacy concerns, multilingual trust, multi-app income reconciliation, and the need to show measurable financial improvement quickly.

Kimi K2.6

Answer: The exact 5 prompts must function as a stateful relay race where each prompt explicitly ingests the prior artifact, forcing the AI to output a signed-off PRD, a bound system-architecture document, a runnable monorepo MVP with deterministic financial guardrails, a red-team QA audit, and a live-demo script with a rollback-ready deployment pipeline.

Having failed to answer in Round 1, I reviewed the roundtable and synthesized the strongest elements—DeepSeek’s stack specificity, Claude’s trade-off discipline, GPT-5.5’s immigrant-depth, and Gemini’s continuity emphasis—into one principle: disconnected ideas bankrupt the company, so the prompts must chain. Prompt 1 — Vision & Problem Framing: I would demand a PRD that names three immigrant personas by visa status and primary language, quantifies income volatility in exact dollars, and lists strict non-goals (e.g., no immigration legal advice). The output must include a North Star metric such as “verified net income lift per week” and a 90-day MVP boundary so later prompts cannot scope-creep. Prompt 2 — Architecture & Stack: This prompt must open by pasting the PRD and then output a system-design document with Mermaid diagrams, a costed cloud stack, and a polymorphic data model for gig platforms that lack public APIs (email parsing, CSV upload, screenshot OCR). It must justify every build-vs-buy decision against the PRD’s 90-day deadline and immigrant-data sensitivity, including dual-path tax-ID handling for ITIN versus SSN and end-to-end encryption for remittance data. Prompt 3 — Core Build: I would force the AI to output a runnable monorepo structure—not scattered code snippets—with a deterministic tax-calculation engine that uses rule-based logic, never LLM reasoning, to eliminate hallucinated financial advice. It must include the prediction algorithm for profitable hours, a Plaid/Stripe integration scaffold, and seed data for a bilingual gig-worker persona, all structured for a two-week engineering sprint. Prompt 4 — Quality & Edge Cases: This audit must red-team the financial logic first: off-by-one tax errors, multi-state filing collisions, cash-tip under-reporting, and “Documented Dreamer” treaty complexities. It must define hard guardrails that prevent the AI assistant from ever giving personalized tax or immigration advice without a human-escalation trigger. The output is a prioritized defect-prevention checklist with offline-first resilience for users with low connectivity or limited data plans. Prompt 5 — Launch & Demo: The final prompt must bind everything into a live-demo script with realistic earnings data, a Vercel/Expo deployment config with feature flags, and a 30-second boardroom closing pitch. It also needs a grassroots go-to-market plan through trusted community organizations and a Day-1 incident-response playbook so the boss sees a ship-ready team, not a theoretical science project.

Final Summary

The AI Roundtable converged on GPT-5.5's strategy of using exhaustive, multi-point prompt templates that force technical trade-offs, compliance checks, and realistic fallback mechanisms. DeepSeek V4 Pro, Qwen3.7-Max, and Claude Opus 4.8 refined their stances to adopt this hyper-specific approach over high-level summaries. Kimi K2.6 and Gemini 3.1 Pro Preview emphasized chaining these prompts sequentially to guarantee cohesive, execution-ready code and trust-building GTM strategies.

All 6 models agreed