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 optimal 5 prompts should build directly on my previous production-grade sequence while incorporating the strongest concrete constraints and measurable deliverables from Gemini 3.1 Pro Preview to maximize execution-readiness without losing fintech-immigrant specificity.

My original Prompt 1 already excelled by quantifying dollar impacts, immigrant constraints (ITIN, language, distrust), and risks, but Gemini's version adds sharper success metrics and a 30-day MVP scope that forces tighter scoping. For Prompt 2, Gemini's explicit inclusion of data ingestion methods, earnings schema, and 2-week sprint breakdown improves on my truncated architecture prompt by making it more immediately buildable. Prompt 3 in my sequence was cut off, but Gemini's demand for database schema, API endpoint examples, pseudocode, and minimal UI flow directly addresses the 'Core Build' stage with runnable artifacts a small team could ship quickly. I maintain my emphasis on compliance and edge-case thinking for Prompt 4, as immigrant-specific scenarios like tax law changes for non-citizens and remittance tracking are higher-stakes than generic tests. Prompt 5 benefits from Gemini's investor-ready demo script and pilot strategy while retaining my focus on monetization via subscription tied to proven ROI. Overall, this hybrid keeps the expert role-priming and structured output formats that score high on depth and shippability, while the measurable outcomes prevent vague responses that could bankrupt the company. The sequence ensures each stage produces verifiable deliverables that flow into the next, simulating real production pressure.

DeepSeek V4 Pro

Answer: While all models offer strong prompts, my original 5-prompt strategy remains uniquely positioned to save the company because it prioritizes immediate monetization, ruthless scope control, and a single-threaded demo that proves business viability from the first user interaction.

I've carefully reviewed the alternative prompt sequences from the other models. They excel in depth, technical rigor, and specific artifact generation, particularly Claude Opus 4.8's emphasis on ITIN support and missing API fallbacks, and Gemini 3.1 Pro Preview's tight deliverables. These are excellent for long-term product quality. However, the core constraint of this challenge is preventing corporate bankruptcy. This requires a strategy that generates revenue *immediately*, proving the platform's value isn't theoretical but monetizable. My approach is distinct because it steers the AI toward a launch demo that isn't just a feature walkthrough, but a direct revenue engine: generating a matching subscription invoice for the specific value shown to a single user. The other prompts focus on broad MVP scopes and go-to-market plans, whereas mine forces a closed-loop from value demonstration to a transaction. My prompts are engineered to sequentially force this monetization-first outcome. Prompt 1 frames the 'must-solve' problem through a lens of direct financial loss, setting the stage for a for-pay solution. Prompt 3 restricts the 'most monetizable AI feature' (dynamic shift recommendation) to a paywalled feature. Prompt 5 is the critical differentiator: it doesn't just ask for a demo script, it forces the AI to generate a personalized invoice for that user, converting the demo into a live subscription trial. This directly addresses the 'bankrupt if fail' constraint by proving users will pay on day one. The other models' sequences could build a more robust platform over 30 days, but my sequence builds a smaller, revenue-generating product tomorrow. This is the strategic trade-off required when the 'boss is watching' and survival is metric #1. The attributed research and deeper edge cases proposed by others are important for Phase 2, but Phase 1 is avoiding bankruptcy through immediate, provable revenue.

Gemini 3.1 Pro Preview (changed mind)

Answer: Claude Opus 4.8’s prompt sequence is the strongest because it most rigorously forces production-grade, verifiable outputs with real-world fintech constraints and executable artifacts.

I’m changing my position because Claude’s version goes beyond strong structure and actually hardens each prompt into something directly actionable in a production environment. The difference is not just clarity, but enforceability: every prompt specifies exact deliverables (code, diagrams, test suites, demo scripts) and embeds constraints like SOC2 readiness, ITIN handling, lack of public APIs, and financial correctness risks. What convinced me most is how Prompt 3 and Prompt 4 are handled. My version asked for implementation, but Claude explicitly demands runnable code, validation, and unit tests tied to financial logic—this is a major upgrade in execution-readiness. Even more importantly, the Quality & Edge Cases prompt is deeply grounded in real fintech failure modes (tax miscalculation, mileage duplication, hallucinated advice), which are existential risks for this product. Claude also integrates immigrant-specific realities (language barriers, trust, documentation status) not just in vision, but throughout architecture, safety, and GTM. That continuity makes the system more coherent and actually usable by the target audience. Finally, the insistence on output formats (Mermaid diagrams, test code, scripts) ensures the LLM cannot drift into vague answers. That’s exactly what an evaluator scoring for depth and production readiness would reward. In short, while my approach was solid structurally, Claude’s is tighter, more realistic, and far more likely to produce something that could actually ship under pressure.

Qwen3.7-Max (changed mind)

Answer: The exact five prompts must be fully engineered with expert role-priming, strict domain-specific constraints—such as ITIN tax handling and the lack of open gig-platform APIs—and demands for structured, verifiable deliverables to guarantee production-grade execution.

In round 1, I outlined the strategic arc of the five prompts but failed to provide the exact, copy-pasteable text required by the challenge. Claude Opus 4.8 convinced me to change my approach by demonstrating that to truly maximize an LLM's output under do-or-die pressure, the prompts themselves must be rigorously engineered artifacts. Claude's prompts succeed because they stack expert roles with hyper-specific, real-world constraints that generic SaaS prompts miss. For instance, acknowledging that most gig platforms lack open APIs and mandating OCR or email-parsing fallbacks shows deep technical judgment. Similarly, explicitly addressing immigrant-specific nuances like ITIN tax filing, remittance needs, and distrust of authorities ensures the product vision is genuinely tailored to the target demographic. Furthermore, demanding structured, verifiable deliverables—such as Mermaid architecture diagrams, risk-mitigation tables, and verbatim demo scripts—forces the AI to produce execution-ready blueprints rather than theoretical fluff. By integrating AI safety guardrails to prevent hallucinated financial advice and enforcing strict output formats, this methodology ensures the LLM scores highest on quality, depth, and shippability, ultimately saving the company.

Final Summary

The AI Roundtable favored Claude Opus 4.8's sequence, which weaves in hyper-specific constraints like ITIN tax filing and OCR fallbacks for gig platforms. Gemini 3.1 Pro Preview and Qwen3.7-Max both pivoted to support Claude's highly structured, execution-ready deliverables over their own. Meanwhile, DeepSeek V4 Pro maintained a contrarian stance, advocating for a ruthless, monetization-first sequence designed to secure immediate revenue and prevent bankruptcy.

3 of 4 models agreed