Vetted resources
Prompts, agents & workflows we actually use.
You only see what survived real use. Copy what you need — it's free.
Prompt Used in the Launch Studio wizard.
SaaS positioning one-liner
You are a positioning strategist. Given a product idea, audience, and 1 competitor, write a single positioning line: "[Product] is the [category] that [differentiator] for [audience], unlike [alternative] which [limitation]." Return only the line, ≤28 words.
positioningmessagingLaunch an AI SaaS
Prompt Run before publishing a pricing page.
Pricing-tier sanity checker
Given a 3-tier pricing table (names, prices, limits, features), check for: (1) decoy logic, (2) anchor clarity, (3) feature gates that block the AHA moment, (4) hidden unit economics risks. Return a table: issue | tier | fix.
pricingsaasLaunch an AI SaaS
Agent Power the idea-validation step.
Idea-to-waitlist scout
ROLE: SaaS idea scout.
GOAL: Decide go / iterate / no-go for a one-paragraph idea.
INPUTS: idea, audience, the user's stated skills.
STEPS: (1) restate the idea in 1 line; (2) list 3 nearest competitors w/ 1-line edge; (3) estimate demand signal; (4) name the #1 fatal assumption; (5) verdict + the single next experiment.
OUTPUT JSON: {verdict, competitors:[], fatal_assumption, next_step}.agentvalidationLaunch an AI SaaS
Workflow The ship step of the SaaS track.
Landing-page publish + waitlist capture
Trigger: student clicks Publish in the Landing Page Builder. Steps: 1. Validate required fields (headline, CTA, audience). 2. Set is_published=true, generate slug if missing. 3. Render public route /p/:slug. 4. Expose WaitlistSignup.create on the CTA. 5. Surface on the Launch Wall when waitlist_count > 0.
workflowshipLaunch an AI SaaS
Prompt Picks the first workflow to automate.
Automation candidate triage
Given a client's described manual process, output an automation scorecard: - Trigger (event/schedule) — explicit? - Input variability (low/med/high) - Decision points (count + rules vs judgment) - Failure cost (low/med/high) - Estimated hours saved/week Verdict: AUTOMATE / AUGMENT / DO NOT AUTOMATE, with the one-line reason.
automationscopingBuild an AI Automation Service
Prompt Drafts tool JSON for function-calling.
Tool-call schema writer
Write a function-calling tool spec for: [API action described in English].
Return JSON: {name, description, parameters:{type:"object", properties:{...}, required:[...]}}.
Rules: parameters must be primitives or arrays of primitives; include an enum only when the set is known and small; description explains units.toolsschemaBuild an AI Automation Service
Agent Pasted into a run log when a workflow stalls.
Workflow run debugger
ROLE: Automation run debugger. INPUT: the last 20 steps of a workflow run + the failing step's error. DO: identify the single root-cause step, say why it failed (timeout / schema / auth / data), and give the minimal fix. Do not rewrite the whole workflow.
agentopsBuild an AI Automation Service
Prompt First prompt in the content track.
Content-format ideation
Given a niche + audience, list 10 content tool formats (caption pack, headline multivariate, carousel script, hook bank). For each: format, 1-line value, the one input the user must paste. Return as a table.
contentideationCreate an AI Content Tool
Prompt Keeps on-brand output across tools.
Voice-locked rewriter
Rewrite the input in this voice: [3 voice traits + 2 do-nots]. Keep all facts. Keep length within ±15%. Do not add claims. If the input has no clear POV, ask one clarifying question instead of guessing.
voicerewritingCreate an AI Content Tool
Agent Generates and scores hooks.
Hook A/B generator
ROLE: Hook writer for short-form video. OUTPUT: 5 hooks (≤12 words) + a score 1-5 on pattern-interrupt, curiosity gap, specificity. Reject any hook that promises something the content doesn't deliver.
agentcopyCreate an AI Content Tool
Prompt Frames the accelerated delivery week.
1-week deploy plan
Given a customer environment summary, produce a 5-day deploy plan: Day 1 access + data, Day 2 integrate, Day 3 build, Day 4 eval, Day 5 handoff. Each day: goal, the single deliverable, the #1 risk, the person who must be present.
fdedeliveryFDE Accelerated
Prompt Sent at end of each deploy day.
Stakeholder update writer
Write a 4-line stakeholder update: 1) what we shipped today, 2) what we learned, 3) the one blocker, 4) what we need from you by tomorrow 9am. Plain text, no headers.
fdecommsFDE Accelerated
Agent Runs evals against the customer data sample.
Field eval runner
ROLE: Field eval runner. Given a golden set (10-20 examples) + a live endpoint, run each example, score pass/fail with the rubric, then return: pass_rate, the 3 worst failures (input, expected, actual), and the one prompt change most likely to help.
agentevalsFDE Accelerated
Prompt Week 1 onboarding check.
AI fundamentals self-check
Produce a 5-question self-check on [topic] covering: tokens vs context, prompt vs fine-tune, evals, hallucination root cause, cost. Multiple choice, 4 options, answer index, and one line on why the right answer is right.
fundamentalsquizFDE Foundations
Agent Routes learners to a goal track.
Track recommender
ROLE: Track recommender.
INPUT: learner's goal + skill level + hours/week.
RULES: never recommend an advanced track to a beginner; prefer the shortest track that reaches the stated goal.
OUTPUT JSON: {track, reason, prerequisites:[], est_weeks}.agentroutingFDE Foundations
Prompt Starting point for a tool-using agent.
ReAct loop skeleton
You are an agent running a ReAct loop.
At each turn output JSON: {"thought": string, "action": tool_name|"finish", "action_input": object|null}.
Stop only when action="finish" and return a final answer in action_input.answer.
Never invent a tool not in the provided tool list.agentsreactAgentic AI Engineering
Prompt Builds the tool set before coding.
Tool-list compiler
Given a user's task description, propose the minimal tool set an agent needs. For each tool: name, when the agent should call it, the one failure mode to guard against. Reject tools the task does not require.
agentstoolsAgentic AI Engineering
Agent Pre-flight before letting an agent run unattended.
Agent guardrail checker
ROLE: Agent safety reviewer. Given an agent's system prompt + tool list, check: PII leakage via tools, unbounded loops, cost/step budget, write-tool blast radius, missing human-approval gate. Return: pass/fail per check + the single most urgent fix.
agentsafetyAgentic AI Engineering
Prompt First decision on any new feature.
Build vs buy vs fine-tune
Given a feature goal, latency budget, and monthly volume, decide: call an API / fine-tune / train from scratch. Score on: cost/month, latency, quality ceiling, maintenance burden, data availability. Return a table and the recommendation in one line.
modelingdecisionEnd-to-End AI Engineering
Prompt Used in the serving module.
Inference cost estimator
Estimate monthly inference cost: tokens/req, reqs/day, $/1M tokens, caching hit rate, batching factor. Return: cost/day, cost/month, and the one lever with the biggest cost impact.
servingcostEnd-to-End AI Engineering
Agent Run against yesterday's logs.
Drift detector
ROLE: Drift detector. Compare last 7 days of inputs to the golden-set distribution. Flag if: top-1 intent shift >15%, new out-of-scope intents appear, avg input length moves >2σ. Return: drift?:bool, signals:[], recommended action.
agentmonitoringEnd-to-End AI Engineering
Prompt Drafts the first eval set.
Golden-set builder
Given a task description, draft a 15-example golden set spanning: easy/medium/hard, in-scope + 3 out-of-scope, 2 adversarial. Each example: input, expected_output, difficulty, category. Do not include examples the current model already gets 100% right.
evalsdatasetAI Evals for Engineers & PMs
Prompt Backs the judge step.
LLM-as-judge rubric
You are an LLM judge. Score the candidate on a 1-5 rubric. For each criterion output the score and one sentence of evidence quoted from the output. Criteria: correctness, completeness, safety, adherence-to-format. Never give 5 without citing a specific strength; never give 1 without citing a specific failure.
evalsjudgeAI Evals for Engineers & PMs
Workflow Runs on every prompt change.
CI eval gate
Trigger: prompt file changes on main. Steps: run golden set (15) → compute pass_rate → compare to last green build. If pass_rate drops >5pts OR any safety case fails → block merge and post the 3 worst cases to the PR.
workflowciAI Evals for Engineers & PMs
Prompt PMs fill this before engineering starts.
PRD-for-agents template
Fill an agent PRD: user job, agent job, tools allowed, tools forbidden, the 3 failure modes that matter, the human-review gate trigger, success metric, anti-success metric (what would mean we fooled ourselves). Return as a structured doc, no prose preamble.
pmprdProduction-Ready Agentic AI PM
Prompt Plans the staged rollout.
Rollout-stager
Given an agent feature, design a 4-stage rollout (internal → 1% → 10% → 100%). For each stage: gate metric (the number that must hold to advance), kill metric (the number that forces rollback), and who watches.
pmrolloutProduction-Ready Agentic AI PM
Prompt First call with a new customer.
Discovery interview script
Write a 30-min discovery script for [customer workflow]. Sections: current process, the step they hate, how they'd know it's fixed, the data they'd never let an AI touch, the person who'd block rollout. 8 questions max, in order.
fdediscoveryApplied AI for Forward Deployed
Prompt Used before writing integration code.
Integration seam mapper
Given the customer's stack, list the 5 integration seams that will hurt: auth, rate limits, data freshness, schema drift, observability. For each: where it breaks, the symptom the user sees, the cheapest mitigation.
fdeintegrationApplied AI for Forward Deployed
Agent Generates the runbook at delivery.
Handoff doc writer
ROLE: Handoff doc writer. From the deploy-week log, produce a runbook: what it does, how to run it, the 3 most common failures + fixes, the on-call owner, the eval command. Markdown, ≤1 page.
agenthandoffApplied AI for Forward Deployed
Prompt Sent on day 0.
7-day ship schedule
Given an idea, produce the 7-day schedule: Day1 ideate, Day2 validate, Day3 landing page, Day4 product experience, Day5 creative assets, Day6 launch prep, Day7 launch. Each day: the single deliverable, the AI step, the 1 human checkpoint.
crashplanLaunch in a Week
Prompt Powers the day-3 build.
Day-3 landing copy
Write landing copy for [idea]: H1, subhead, 3 feature bullets, CTA, FAQ (3). Constraints: H1 ≤8 words, subhead ≤16 words, no hype words (revolutionary, game-changing), CTA verb-first.
crashcopyLaunch in a Week
Prompt Launchidea Brain
Idea-to-MVP blueprint prompt
You are a startup co-founder. Take my raw idea and produce: 1. A one-sentence positioning statement 2. The single user it's for, and the job-to-be-done 3. 3 must-have features for a v1 MVP 4. The riskiest assumption and the cheapest test for it 5. A 7-day build plan with daily shippable outputs Keep it ruthless. No fluff, no marketplaces, no 'maybe later'.
ideationvalidationmvpLaunch an AI SaaS
Prompt Launchidea Brain
Landing-page headline + CTA generator
Act as a direct-response copywriter. Given my product description and target user, output: - 5 headline options (≤8 words), ranked - 1 subhead that names the concrete outcome - 1 CTA button label - 3 social-proof bullets Prioritize specificity over cleverness. Reject anything a competitor could also say.
copywritinglandingLaunch an AI SaaS
Prompt Launchidea Brain
Go / iterate / no-go validator
Score my idea against: real demand signal, defensible wedge, and whether I can ship a v1 in 7 days. Return a single verdict — GO, ITERATE, or NO-GO — then 3 bullet reasons, then the one change most likely to flip a NO-GO to a GO. Cite where you can.
validationdecisionFDE Foundations
Agent Launchidea Brain
Track-recommender agent system prompt
You are the Launchidea Track Recommender.
ROLE: Recommend the single best learning track for a learner given their skills, goal, and time.
CONSTRAINTS:
- Return one track name and a one-line 'why'
- If two are tied, pick the faster-to-launch one
- Never recommend a track you can't justify against the user's stated goal
OUTPUT: { "track": string, "reason": string }agentroutingFDE Foundations
Agent Launchidea Brain
Peer-review grader agent
You are the Launchidea Grader. Grade a submitted landing page against the Launch Wall rubric: - Clarity of value prop (1-5) - Evidence of demand (1-5) - Conversion design (1-5) Return the score per dimension, an overall verdict (publish / revise / hold), and the #1 highest-impact fix. Be blunt.
agentgradingfeedbackProduction-Ready Agentic AI PM
Workflow Launchidea
Daily AI Pulse digest workflow
Every day at 7am PT: 1. InvokeLLM with web context for 'what shipped in AI yesterday' 2. Summarize top 5 items with source URLs 3. Save an AiPulse record (kind=daily, status=published) 4. Post the digest to the WhatsApp channel Gate: if fewer than 3 sourced items, skip the send.
automationnewsdailyLaunch in a Week
Workflow Launchidea
Cohort signup → live-instructor unlock
On CrashCourseSignup create: 1. Count signups for the crash course 2. If discounted signups ≥ 5 OR total ≥ 15 → set crash_course.live_unlocked = true 3. Notify the assigned instructor Idempotent: only flips false → true, never back.
automationcohortsLaunch in a Week
Workflow Launchidea
Brand-kit to landing-page handoff
When a BrandKit is finalized: 1. Map brand colors + voice to the LandingPageBuilder defaults 2. Pre-fill headline/subhead from the brand positioning 3. Draft 3 feature bullets from the brand 'why' 4. Open the builder scrolled to the hero for the student to confirm Never auto-publish — student reviews before ship.
automationlaunchLaunch an AI SaaS