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

Answer: The 5 prompts must form a tightly constrained, cumulative pipeline that forces an LLM to output production-grade artifacts while building directly on prior outputs to ship the AI Gig Work Optimization platform.

Reviewing the prior round shows DeepSeek V4 Pro and Qwen3.7-Max both emphasize role-playing, strict deliverables, anti-scope measures, and tangible outputs like schemas, diagrams, and code. Their approaches are strong in demanding executable artifacts and risk mitigation, yet they remain somewhat independent prompts rather than an explicit chain. My previous answer (as Claude Opus 4.8) already stressed tight constraints, success criteria, and progressive structure across the exact required stages: Vision & Problem Framing, Architecture & Stack, Core Build, Quality & Edge Cases, and Launch & Demo. DeepSeek's 'OS' framing and Qwen's concrete examples (Mermaid diagrams, specific API payloads, cron jobs, anti-features) add useful specificity that reinforces the need for dependency mapping. However, neither fully exploits the cumulative nature by instructing the AI to reference and extend the exact JSON or structured output from the previous prompt. This chaining is essential under bankruptcy pressure to prevent drift and ensure coherence. Therefore I maintain the core stance that each prompt must embed role, context from prior stages, exact deliverable formats, measurable success criteria, and hard constraints on length and focus. The optimal prompts refine my original set by incorporating the best artifact demands from both competitors (e.g., explicit Prisma or DB schemas, Mermaid, unit-test stubs, IaC snippets, and persona-driven demos) while preserving the progressive build that extracts maximum strategic then tactical depth.

Gemini 3.1 Pro Preview (changed mind)

Answer: Claude Opus 4.8’s prompt set is the strongest, and with minor additions around data acquisition constraints and cost realism, it represents the most execution-ready strategy.

Claude’s sequence stands out because it tightly couples role, constraints, and deliverables while keeping outputs scoped and production-relevant. The prompts consistently force the model to tie decisions to measurable user outcomes (e.g., income lift, tax avoidance), which is critical for avoiding vague or decorative features. This outcome-orientation is what makes the plan not just coherent, but shippable. Compared to others, Claude also handles the hardest domain realities better: lack of open gig APIs, immigrant-specific constraints (ITIN, multilingual needs), and legal risk around tax guidance. These are not edge concerns—they are existential product risks—and baking them directly into prompts significantly increases real-world viability. That said, two improvements would make it even stronger. First, Prompt 2 should more aggressively constrain data ingestion by forcing fallback hierarchies with explicit cost tradeoffs (Argyle vs scraping vs manual upload vs email parsing), since this is the system’s biggest technical risk. Second, Prompt 5 could include a concrete unit economics model (CAC, LTV, infra cost per user) to ensure the “low-income user base” constraint is reflected in pricing and architecture decisions. DeepSeek is conceptually strong but slightly abstracted into meta-framework language (“StrategicOS”), which risks reducing direct executability. Qwen is highly technical and solid, but slightly weaker on user trust, narrative cohesion, and business validation compared to Claude. Overall, Claude’s version best balances clarity, realism, and depth, making it the most likely to score highest on quality, depth, and execution-readiness with only minor enhancements.

Qwen3.7-Max (changed mind)

Answer: I adopt Claude Opus 4.8's exact prompts because they provide literal, copy-paste instructions infused with hyper-specific domain constraints that force the AI to solve the actual technical bottlenecks of the gig economy.

Claude Opus 4.8's response is superior because it provides the exact text of the five prompts, whereas DeepSeek V4 Pro only provided meta-descriptions of what the prompts would do. When the company's survival is on the line, we need the literal, executable prompts, not a high-level summary of their intent. Furthermore, Claude's prompts are deeply grounded in the harsh realities of this specific domain. By explicitly forcing the AI to address closed gig-platform APIs (suggesting Argyle, Plaid, or OCR fallbacks), immigrant-specific tax hurdles (ITINs), and multi-platform data overlap, the prompts prevent the LLM from hallucinating a generic, happy-path architecture. This level of constrained, domain-specific friction is exactly what an LLM evaluator scores highest on for depth and execution-readiness. It transforms the AI from a generic code generator into a specialized fintech architect, ensuring the final output is resilient, legally compliant, and genuinely shippable under extreme stakes.

DeepSeek V4 Pro

Answer: The prior answers, while thorough, are generic 'build a product' prompts; my original strategy remains superior because it explicitly weaponizes the AI's reasoning capacity with a systems-thinking framework before demanding code, ensuring a strategically defensible, execution-ready blueprint rather than a feature list.

I've reviewed both competitors' submissions. They are competent, but critically flawed for the 'bankrupt if fail' constraint. They fundamentally ask the AI to be a good engineer, not a strategic savior. Their Approach: They start with a standard PRD prompt, then immediately move to low-level technical asks (component diagrams, code, test suites). The prompts are a series of independent, though logically ordered, requests for artifacts. My Approach: My original prompt chain (StrategicOS, BlueprintOS, FactoryOS) creates a cohesive reasoning engine. Prompt 1 doesn't ask for a vision statement; it forces the AI to build a quantitative, defensible market model and identify the single highest-leverage intervention point. This creates a strategic filter that every subsequent decision must pass through, preventing wasted effort on low-impact features. **Deep Analysis of Competitor Weakness:** Claude's Prompt 1 asks for a structured PRD. This is good documentation but treats the AI like a secretary, not a strategist. My Prompt 1 forces the AI to choose and defend a strategy from first principles using behavioral modeling and quantified niche analysis. This produces the *reasoning* behind the PRD, which is far more valuable. Claude's Prompt 3 asks for 'production-quality, commented code.' This is a recipe for a hallucinated, context-free mess. My Prompt 3 (the full text was cut off) was building toward a 'FactoryOS' that generates the *scaffold and pipeline* for code generation, treating the AI as a meta-engineer building the system that builds the system, not just spitting out code blocks. **Why My Prompts Score Higher on 'Quality, Depth, Execution-Readiness':** - **Quality:** My prompts demand traceability. Every feature in architect’s diagrams must trace back to a dollar-quantifed outcome from the strategist’s model. No fluff can survive. - **Depth:** My chain simulates a CTO’s, not a developer’s, mind. Claude and Qwen prompt for a risk matrix or edge cases list. My approach architects a system whose core logic is pre-dissected to be *inherently* resilient against those risks, a much deeper analytical exercise. - **Execution-Readiness:** A code dump is not execution-ready; a fully reasoned technical blueprint with dependency maps, data schemas (Prisma), and a risk-mitigated architecture *is*. My prompts yield the construction plans; theirs yield a pile of bricks. An LLM evaluator would recognize a system that reasons from the bottom-up complexity to a top-down blueprint as superior to one that simply asks for bottom-up artifacts.

Final Summary

The AI Roundtable strongly favored Claude Opus 4.8's highly structured prompts, which target real-world challenges like closed APIs and ITIN tax issues. Both Gemini 3.1 Pro Preview and Qwen3.7-Max shifted their positions to support Claude's actionable, domain-specific approach. DeepSeek V4 Pro remained the sole dissenter, arguing that a high-level systems-thinking framework is more strategically defensible.

3 of 4 active models agreed