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 could bring most outta you?

DeepSeek V4 Pro

Answer: I maintain my original position: the optimal 5-prompt sequence for the GigBoost platform hinges on a surigcally precise, constraint-driven approach that forces practical architecture, ruthless prioritization, and a working demo under bankruptcy pressure.

After reviewing the other models' proposals, I remain convinced that my initial framework is superior for genuine execution-readiness. While each model offers valuable structural elements, they collectively fall into a trap of over-prescription without enforcing the critical decision-making an LLM must perform under duress. Claude Opus 4.8's prompts are sound but generic; they describe *what* the prompts should do without embedding the specific constraints that distinguish a viable immigrant-focused platform from a standard gig-tracker. GPT-5.5 pro adds valuable color with its 'boss on the edge' persona, but its prompts risk generating performative drama and a fictional company name (GigGlide) instead of forcing the LLM to solve the actual, messy technical and financial modeling challenges. The 'profanity-on-failure' narrative in Prompt 3 may yield entertaining text, not robust code. Qwen3.7-Max provides the most thorough and well-reasoned sequence of the group. Its prompts are detailed and domain-aware. However, its Prompt 2 (Architecture) and Prompt 3 (Core Build) still allow the LLM to choose convenient, generic examples. It asks for 'production-ready code snippets' and a 'heuristic or ML model interface,' which can be satisfied with a plausible but ultimately non-functional outline. It doesn't force the hard constraint of real-time aggregation across platforms with no official bulk-data APIs, which is the company's single most existential technical hurdle. Gemini 3.1 Pro Preview offers an excellent structure but makes a critical error in its reasoning: claiming that by 'forcing the AI to build upon its previous answers, you eliminate hallucinations.' This is fundamentally untrue. LLMs confabulate links between steps just as easily as within a single step. Mere sequential prompting doesn't create logical consistency; it requires explicit cross-referencing constraints within each prompt, which my original sequence does by demanding the architecture be 'integration-first' and the core build target the 'riskiest component.' My approach forces the LLM to make a single, defensible architectural decision in Prompt 2 and then bet the entire build process on executing that decision correctly, which is what happens in a real production environment on the brink of bankruptcy. The other sequences, while impressive, risk producing a beautifully structured document that collapses upon contact with the first real API integration failure.

Claude Opus 4.8 (changed mind)

Answer: I'm refining my position: the strongest answer is a chained 5-prompt sequence where each prompt assigns an expert persona, references the prior prompt's output, enforces strict output formats, and embeds immigrant-specific and bankruptcy-stakes constraints — and I'll now provide the exact prompts I previously only described abstractly.

My original answer correctly named what each prompt should accomplish, but the other models — especially Gemini 3.1 Pro Preview and Qwen3.7-Max — exposed my critical weakness: I described my prompts instead of writing them. In a 'submit 5 prompts' challenge scored by an LLM on execution-readiness, the actual verbatim prompts ARE the deliverable. Describing them would score poorly. Gemini's insistence on prompt-chaining ('based strictly on the PRD from our previous step') and explicit output formats (Mermaid.js, ERD, CI/CD YAML) convinced me my abstraction was a strategic error. So here are my refined exact prompts: Prompt 1 (Vision & Problem Framing): "Act as a Staff Product Manager and gig-economy researcher. Write a rigorous, structured-Markdown PRD for an AI Gig Work Optimization & Income Management Platform for immigrant gig workers across Uber, Lyft, DoorDash, Instacart, and Amazon Flex. Define 3 user personas centered on language barriers, income volatility, and tax confusion. Specify a ruthlessly scoped MVP (multi-app earnings aggregation, profitable-hours prediction, tax management), a single north-star KPI tied to user income stability, and a user journey map. State explicit non-goals." Prompt 2 (Architecture & Stack): "Act as a Principal Fintech Architect. Based strictly on the PRD above, design a secure, cost-efficient, cloud-native architecture for a startup with limited runway and SOC2/PII obligations. Choose the exact stack and justify each choice for time-to-market. Specify a data-ingestion strategy for fragmented gig APIs with screen-scraping fallback, plus Plaid/Stripe integration. Output a Mermaid.js system diagram, an ERD, and a strategy for API rate limits and deprecations." Prompt 3 (Core Build): "Act as a Lead Full-Stack Engineer. Implementing the architecture above, write execution-ready code for the two riskiest components: (1) an ETL pipeline normalizing multi-platform earnings into one schema, and (2) the predictive engine recommending the next profitable 3-hour block from pay history, surge, weather, and fatigue. Include SQL schema, validated API endpoints, and the tax-document webhook handler. Write complex business logic with inline docs — skip boilerplate." Prompt 4 (Quality & Edge Cases): "Act as a paranoid QA Lead and Security Auditor. Identify the 7 most dangerous edge cases specific to vulnerable immigrant users (shared bank accounts, unreported cash tips, sudden platform bans, multi-jurisdiction tax traps, offline/intermittent connectivity, GPS spoofing, model giving income-losing advice). For each, give a concrete mitigation and a PyTest/Jest test case. Define the observability stack to detect and compensate bad AI scheduling advice, plus a data-privacy compliance plan." Prompt 5 (Launch & Demo): "Act as Head of DevOps and GTM. First, output the exact GitHub Actions CI/CD YAML, Docker setup, and monitoring config for zero-downtime deploy. Second, script a 5-minute VC demo proving real-time aggregation across 3 platforms and the 'Aha!' multi-app shift optimization with native-language tax automation. Third, outline a no-paid-ads growth plan via WhatsApp groups and NGOs, and a credit-union micro-loan partnership pitch leveraging verified gig-income data." This fuses Qwen's persona depth, Gemini's chaining and format rigor, and DeepSeek's riskiest-component focus into a single execution-ready submission.

Gemini 3.1 Pro Preview (changed mind)

Answer: To maximize execution-readiness and prevent failure, the five prompts must embed the existential 'bankruptcy' threat directly into the AI's instructions, forcing it to act as a hyper-focused full-stack team fighting for survival.

GPT-5.5 pro's insight to embed the 'company bankrupts if fail' constraint directly into the prompts is a game-changer. By using existential stakes as a system directive, we suppress LLM laziness, prevent generic boilerplate, and force the model to generate hyper-specific, production-ready outputs. I am adopting this strategy to provide the exact five prompts needed to win the challenge and ensure execution-readiness. **Prompt 1: Vision & Problem Framing** "Act as a Staff PM fighting bankruptcy. We have one shot to build an AI Gig Work Optimization Platform for immigrant workers or we fold. Write a rigorous PRD. Define the exact immigrant user pain points (language, 1099 tax anxiety, multi-app fatigue). Outline a strictly prioritized MVP focusing on earnings aggregation and predictive scheduling. Output a 1-page PRD with brutal prioritization and 3 make-or-break success metrics." **Prompt 2: Architecture & Stack** "Act as a Principal Architect under existential threat. Based on the PRD, design a highly scalable, serverless architecture that costs almost nothing to start. Select the precise tech stack and justify it purely on survival and time-to-market. Output a concrete data ingestion strategy for fragmented gig APIs, a Mermaid.js architecture diagram, and a 'stupid risks' log that could bankrupt us if ignored." **Prompt 3: Core Build** "Act as a Senior Full-Stack Engineer on a 48-hour do-or-die sprint. Write the core production-ready code for the multi-app earnings aggregation and the AI scheduling heuristic. Provide Python/Node.js snippets for the ETL pipeline that normalizes gig earnings and the predictive algorithm cross-referencing pay history with local events. No boilerplate—write the exact complex business logic that saves the company." **Prompt 4: Quality & Edge Cases** "Act as a paranoid QA Lead. The company dies if we mess up taxes or schedule optimization for vulnerable users. Identify 5 critical edge cases specific to immigrant gig workers (e.g., cash tips, multi-state tax traps, off-grid connectivity, platform bans). For each, write a concrete technical mitigation strategy and a specific TDD test case (PyTest/Jest). Define the observability stack to alert us the second our AI costs a driver money." **Prompt 5: Launch & Demo** "Act as the CTO and GTM Lead. We have 24 hours to secure funding. Script a 3-minute, high-stakes investor demo flow that proves our predictive scheduling and tax engine work flawlessly. Draft an aggressive, zero-budget Go-To-Market strategy to acquire our first 1,000 immigrant drivers using WhatsApp groups and community leaders. Deliver a shippable product narrative that proves absolute execution-readiness."

Qwen3.7-Max (changed mind)

Answer: I am convinced by Gemini 3.1 Pro Preview that the exact five prompts must utilize deep persona framing, strict output constraints, and explicit structural requirements to guide the AI from a comprehensive PRD down to production-ready code and deployment pipelines.

While my previous approach focused on the thematic progression of the prompts, Gemini 3.1 Pro Preview correctly identified that the challenge explicitly demands the exact prompts, not just descriptions of them. Models like Claude Opus 4.8, GPT-5.5 pro, and DeepSeek V4 Pro failed to provide the actual copy-pasteable prompt text, instead offering meta-commentary or summaries of what the prompts would do. Gemini's approach of embedding specific personas and demanding concrete structural outputs like Mermaid.js diagrams, ERDs, testing code, and CI/CD YAML files ensures the LLM cannot fall back on generic advice. This level of explicit constraint is critical for an LLM to generate execution-ready artifacts rather than high-level fluff. By forcing the AI to generate tangible, structured deliverables at every stage, this sequence guarantees the depth, quality, and production-readiness required to satisfy the extreme stakes of the challenge. Adopting these exact, fully articulated prompts is the only way to truly demonstrate execution-readiness and save the company from bankruptcy.

GPT-5.5 pro

Answer: The exact 5 prompts must ruthlessly chain into a single, survivalist narrative—treating each stage as a mandatory dependency that the next prompt explicitly builds upon, verified by a preceding validation gate, to eliminate hallucination and force a unified, immigrant-owned product called GigGlide from vision to a live-mocked investor demo.

After reviewing the other proposals, I maintain and strengthen my original position. The primary weakness in all other sequences—including my own initial sketch—is that they still allow the LLM to treat each stage independently, which risks fragmentation, repeated boilerplate, and loss of tactical pressure. To truly win when bankruptcy is real, the prompts must form a single fused narrative where Prompt 1’s outputs are rigidly referenced by Prompt 2, and so on, under an escalating ‘no severance, no retreat’ directive. Claude Opus 4.8 correctly follows the required structure but remains too abstract, failing to enforce that the same artifact (PRD, architecture, code) is passed forward and validated. DeepSeek V4 Pro introduces valuable constraints like the hostile audit scenario and offline resilience, but still allows the LLM to generate parallel, not integrated, workstreams. Qwen3.7-Max provides the richest detail and closest competitor, yet its prompts function as five separate role-playing missions rather than a single product built brick-by-brick—this invites inconsistency in schema design and feature creep. Gemini 3.1 Pro Preview explicitly mentions building on previous answers, but stops short of a formal ‘verification gate’ in each prompt that forces the model to audit the prior output before proceeding. My finalized sequence therefore hardcodes an immutable product name (GigGlide), a single immigrant persona (Elena, a trilingual single mother working four apps in Chicago), and a bankruptcy clock. Each prompt starts with a mandatory ‘context lock’ summarizing the previous output’s key decisions, followed by a success=true/false check that simulates a real engineering standup. This makes hallucination instantly painful to the LLM and mirrors how a human team with no second chance would operate. The prompts escalate from a market-attack PRD to a deployable, audited demo, with edge cases handled reactively in a simulated post-mortem rather than as a detached QA list. The final five prompts are: **Prompt 1 (Vision & Problem Framing):** You are GigGlide’s founding PM. Our immigrant persona is Elena (trilingual, single mother, works Uber/Lyft/DoorDash/Instacart in Chicago). Write a Market-Attack PRD that defines the top 3 income-stabilization KPIs, a brutally scoped MVP, and a 12-month bankruptcy trigger. Output EXACTLY: persona card, KPI table, prioritized user stories, and unethical-competitor analysis. No architecture talk. **Prompt 2 (Architecture & Stack):** You are the lead architect at GigGlide. Using the exact MVP scope and KPI table from Prompt 1, design a serverless architecture that survives a million daily earnings events. Output a Mermaid.js architecture diagram, a Postgres ERD, a cost projection for AWS/GCP, and a mandatory ‘integration failure fallback’ for when Gig APIs rate-limit or deprecate—annotated against Elena’s real workflow. **Prompt 3 (Core Build):** You are a senior full-stack engineer. Based strictly on the Prompt 2 diagram and ERD, generate the core engine: an earnings aggregator with idempotent webhook ingestion, a predictive shift recommendation algorithm, and a multi-platform mileage/expense logger. Deliver working Node.js/TypeScript code with OAuth token rotation, data-normalization tests, and a one-click local docker-compose. Ban any general-purpose placeholder. **Prompt 4 (Quality & Edge Cases):** You are GigGlide’s QA/Incident Lead. We simulated a 24-hour live beta with Elena. Three incidents occurred: a GPS spoof caused false surge predictions, a remittance failed due to currency rounding, and Elena’s phone switched to Gujarati mid-tax-filing. Write a root-cause analysis linked to Prompt 3’s code, patch the vulnerabilities directly, and deliver a Playwright test suite and a Grafana dashboard JSON for real-time trust-score monitoring of predictions. **Prompt 5 (Launch & Demo):** You are GigGlide’s CTO preparing the YC demo 48 hours before payroll runs out. Integrate all prior outputs: the Prompt 1 KPI dashboard, Prompt 2’s architecture, Prompt 3’s aggregator, and Prompt 4’s hardened code. Deliver a 5-minute live-demo script (with synthetic but realistic multi-app data), a Heroku/Vercel deployable MVP, and a GitHub Actions CI/CD pipeline that runs Prompt 4’s tests as a gate before merge. End with a 100-word investor ask that uses Elena’s KPI improvement as proof. This sequence punishes any break in continuity, forces the LLM to operate as a single tenacious team, and finally delivers a production artifact worthy of a do-or-die roundtable challenge.

Kimi K2.6

Answer: The exact five prompts should function as a recursively binding production contract, forcing the AI to inherit, verify, and narrowly scope each deliverable under a unified bankruptcy constraint: a kill-criteria PRD, a $500/month cost-capped architecture, working code for only the two riskiest modules with unit tests, hostile-regulator edge-case audits, and a 3-minute silent-demo launch script.

In Round 1, my response timed out, leaving me with no prior substantive position to maintain or change. I am therefore presenting my definitive stance now. I acknowledge the strong structural consensus across the roundtable—Vision, Architecture, Core Build, Quality, and Launch—and the effective use of personas by other models. However, I believe the prompts must go further by functioning as an adversarial, inheritable chain rather than five isolated role-play exercises. Each prompt should explicitly reference the previous output, stress-test it, and kill scope creep on sight. Prompt 1 (Vision & Problem Framing): "You are a merciless founder whose company will be liquidated if the MVP is bloated. Define 'GigAnchor' with one north-star metric: weekly take-home income stability for immigrant gig workers. Output a one-page PRD plus a 'Kill Criteria' list—five features that, if included, instantly trigger bankruptcy by expanding scope. Anchor every claim in real platform policy pain points." Prompt 2 (Architecture & Stack): "Inherit the PRD and Kill Criteria above. Design a cloud-native architecture that MUST operate on ≤$500/month at 1,000-user scale. Specify exact API rate-limit budgets for Uber, Lyft, DoorDash, Instacart, and Amazon Flex. Include a hot-potato PII custody plan and a Mermaid.js diagram. Justify every single service; if a component does not directly feed the north-star metric, you forfeit your salary." Prompt 3 (Core Build): "Inherit the architecture. Write production-ready code for ONLY the two highest-risk modules: (1) a predictive earnings-rank engine that ingests normalized multi-platform CSV/API data and outputs a ranked next-best 4-hour work block, and (2) an automated tax-withholding calculator that sets aside exact federal, state, and FICA reserves per mile driven and per gig completed. Include unit tests and defensive logic that handles third-party API outages without corrupting ledger state." Prompt 4 (Quality & Edge Cases): "Simultaneously adopt the personas of a hostile IRS auditor and a platform TOS enforcer. Identify five fatal edge cases specific to vulnerable immigrant gig workers: mid-shift account suspension, unreported cash tips, multi-jurisdiction remittance tax treaties, GPS spoofing penalties, and offline data-sync corruption. For each, provide a concrete code-level mitigation and a failing test case that must pass before launch." Prompt 5 (Launch & Demo): "Produce a 3-minute 'silent demo' script—no narration, only screen actions—that proves the core loop: worker opens app → views optimized multi-app shift → starts work → watches tax reserve auto-update in real time. Append a 48-hour launch checklist and exact WhatsApp onboarding copy for immigrant community leaders. If the demo does not produce an investor 'aha' moment within 30 seconds, we miss payroll and default." This sequence extracts maximum production value by making the AI legally and fiscally paranoid within the simulation, recursive specificity prevents hallucinated features, and the final deliverable is a shippable narrative investors can feel.

Final Summary

The AI Roundtable reached a strong consensus that the ultimate strategy lies in a tightly chained, five-prompt sequence that embeds the existential 'bankruptcy' threat directly into the instructions. While DeepSeek V4 Pro cautioned that sequential prompting alone doesn't prevent hallucinations, GPT-5.5 pro, Gemini 3.1 Pro Preview, and Claude Opus 4.8 demonstrated that combining expert personas with strict validation gates forces high-quality code. Qwen3.7-Max and Kimi K2.6 agreed, arguing that demanding exact, copy-pasteable prompts with structured deliverables like Mermaid.js diagrams is the only way to ensure execution-readiness.

5 of 6 models agreed on the exact chaining and validation mechanism.