ADIN Go-To-Market Strategy & Execution Playbook

Part 1: How ADIN Works

ADIN is a multi-agent AI assistant that learns and adapts over time. Unlike single-model chatbots, ADIN orchestrates specialized agents, maintains persistent memory, and improves from every interaction. This section explains the core systems that make ADIN unique.

Multi-Agent Orchestration

At the heart of ADIN is a powerful reasoning model that acts as the main orchestrator. Rather than handling every task itself, it delegates to specialized agents optimized for different types of work. Agents can delegate to each other up to 3 levels deep, enabling complex workflows.

Delegation Agents (17 Specialists)

The delegation system handles specific task types through focused specialists:

Core Agents form the backbone of task execution. The Researcher finds information from documents and web sources. The Analyst synthesizes and compares data, using compute for calculations and charts for visualization. The Writer formats output according to user preferences. The Verifier fact-checks claims against sources. The Planner breaks complex tasks into actionable steps with agent assignments.

Professional Specialists handle domain-specific work. The Legal Analyst reviews contracts, identifies risks, and drafts agreements using the draft_contract tool. The Legal Brief Writer researches case law via CourtListener and drafts briefs, memoranda, and legal opinions with proper citations. The Financial Analyst performs investment banking-grade analysis including DCF valuations, comparable company analysis, and SEC filings research.

Business Operations Agents cover go-to-market and operational needs. The GTM Strategist orchestrates launch plans, positioning, and pricing strategy, delegating to other specialists as needed. The Accountant handles bookkeeping, tax planning, and GAAP/IFRS compliance. The Sales Strategist builds playbooks, outreach sequences, and objection handling. The HR Advisor drafts job descriptions, benchmarks compensation, and designs org structures.

Content Agents produce high-quality written output. The Marketer creates copy and develops positioning for different audiences. The Influencer crafts Twitter threads and social content with strict style rules (no emojis, no hashtags, punchy hooks). The Journalist writes long-form articles, newsletters, and investigative pieces with rigorous sourcing. The Grader evaluates all written output on an A-F scale and provides improvement feedback.

Debate Agents (Dialectic Reasoning)

For decisions that benefit from stress-testing, ADIN spawns a structured debate between specialized agents:

The Challenger uses adversarial techniques to find flaws. It can operate in multiple styles: adversarial (find logical fallacies and counter-examples), Socratic (probing questions that expose hidden assumptions), steelman (present the strongest possible counter-arguments), or adaptive (use all techniques based on effectiveness). The Challenger's job is to break arguments, not to be agreeable.

The Advocate defends positions while remaining intellectually honest. It concedes valid points, refines arguments in response to challenges, and strengthens the core thesis. The Advocate doesn't win by ignoring good objections - it wins by addressing them.

The Synthesizer produces balanced conclusions after the debate rounds complete. It weighs the strongest arguments from both sides, identifies where consensus emerged, and notes remaining areas of genuine uncertainty. The output is a nuanced position that has survived adversarial testing.

The debate system is particularly valuable for investment decisions, strategic choices, and any situation where confirmation bias could be costly.

Ensemble Forecaster (Wisdom of Crowds)

For probability predictions, ADIN runs multiple LLMs in parallel from different providers. Each model acts as an independent calibrated forecaster, producing probability estimates without seeing the others' outputs. The final prediction uses the median of the estimates, providing robustness against any single model's biases.

This approach consistently outperforms single-model predictions in backtesting, particularly for questions where different models have different blind spots.

How It Integrates with Memory & Learning:

The Ensemble Forecaster becomes significantly more powerful when combined with ADIN's memory and adaptive learning systems:

Forecast Storage & Resolution Tracking: Every prediction ADIN makes is stored in persistent memory with a unique forecast ID, the probability distribution, key reasoning, and a resolution date. When that date arrives, ADIN can compare the predicted outcome against reality and compute a Brier score (a standard measure of forecast accuracy where 0 is perfect and 1 is worst).

Calibration Feedback Loops: Over time, ADIN detects systematic biases in its forecasting. For example, if ADIN consistently predicts bullish crypto outcomes at 65% but they only resolve true 40% of the time, the system learns to apply a calibration adjustment (e.g., "subtract 15 percentage points from bullish crypto predictions"). These calibration rules are stored in persistent memory and applied to future forecasts automatically.

Global Learning Across Users: When ADIN discovers a forecasting insight--like "momentum matters more than I weighted" or "markets are more efficient than my model assumes"--it stores this as a global lesson. Future forecasts for all users benefit from this accumulated wisdom.

Ensemble Weighting Evolution: As resolution data accumulates, ADIN can learn which models in the ensemble perform better on which types of questions. Crypto price predictions might weight Model A higher, while geopolitical forecasts might favor Model B. The ensemble becomes smarter over time.

The Self-Improvement Cycle:

This creates a true learning system: ADIN's forecasts today are informed by every prediction it has made and resolved in the past. Unlike static models, ADIN gets measurably better at forecasting over time--and that improvement compounds across the entire user base.

Model Tiering & Auto-Upgrade

ADIN routes intelligently across four model tiers based on task complexity:

Each agent has a default tier and an upgrade tier. If an agent reports low confidence in its answer, ADIN automatically retries with the more powerful model. This ensures quality without wasting compute on simple tasks.

Contextual Next Steps

After completing any task, ADIN suggests relevant next actions as clickable chips. This surfaces capabilities users might not know about and creates natural workflow continuity.

How It Works:

The suggestion system follows a pattern: one expected action plus one or two discovery hints. Expected actions are obvious follow-ups (export, share, reply). Discovery hints are invitations to capabilities the user may not know about, phrased as questions ("Track as goal?", "Schedule weekly?", "Visualize this?").

Examples:

This system teaches users what ADIN can do through contextual moments rather than documentation. Users discover new capabilities exactly when those capabilities would be useful.

Memory & Knowledge System

ADIN maintains both short-term working memory and long-term persistent memory, with a sophisticated RAG (Retrieval-Augmented Generation) system for semantic search across all user content.

Dual-Scope Memory:

Knowledge Base: Files uploaded, Gmail attachments, and AI-generated documents are automatically processed through a pipeline: extract text → chunk semantically → generate embeddings → index to vector database. This enables semantic search across all your content.

Workspace Isolation: Content can be scoped to personal (only you), team:{teamId} (shared with team), or network (public to all ADIN users).

Hybrid Search: The RAG system combines semantic search (vector similarity) with keyword filtering and uses Reciprocal Rank Fusion to merge results, graded by quality thresholds.

Context Compression System

Long conversations can exceed model context limits. ADIN handles this through a multi-layer compression system:

Rolling Context Summary: When a conversation approaches the token threshold (70% of budget), ADIN proactively summarizes older messages into a compressed summary. Recent messages (the "hot zone") are kept in full detail while older messages (the "cold zone") are replaced with the summary. This happens automatically without user intervention.

Overflow Recovery: If a context length error occurs despite proactive compression, the system attempts recovery by calling a summarization endpoint. It formats the conversation, generates a summary, and rebuilds the message array with the summary prepended and only recent messages retained. The system uses retry logic with exponential backoff and tracks recovery attempts to prevent loops.

Adaptive Trimming: The server-side chat route trims messages to fit within budget before sending to the model. If the initial attempt fails with a context overflow, it retries with a more aggressive budget. If both attempts fail, it returns a clear error asking the user to start a new conversation.

Task Classification & Adaptive Budgets

Not all requests need the same compute budget. ADIN classifies tasks by complexity and allocates steps accordingly:

Stagnation Detection prevents wasted compute by identifying when the agent is stuck in a loop: repeated identical tool calls (same tool with same args 4+ times), alternating patterns (ABABAB between two tools), or hitting the hard ceiling for the task type.

Tasks & Flows: Automation System

ADIN can run autonomously on schedules or in response to events. This is the automation layer that makes ADIN proactive rather than purely reactive.

Scheduled Tasks are simple recurring automations. Tell ADIN "email me every Sunday with a summary of my portfolio" and it creates a cron-based task that runs the agent headlessly and delivers results via email or saves to a conversation thread.

Flows are event-driven automations combining triggers (time, email, crypto price, calendar, webhook) with actions (run AI task, send email, start conversation, create Linear issue, send Discord/SMS).

Proactive Conversations: The start-conversation action is unique - ADIN initiates a chat that appears in your inbox marked as unread, and you can respond to continue the conversation. This turns ADIN from a reactive tool into a proactive assistant.

Adaptive Learning System

ADIN learns from every interaction and applies those lessons globally. This is the meta-learning layer that makes ADIN improve over time.

Global Lessons: When a strategy works (or fails), ADIN records the situation, approach, outcome, and insight. Lessons are deduplicated, ranked by helpfulness, and shared globally so all users benefit.

Failure Fingerprinting: When tools return poor results, ADIN classifies why - too_broad (add specifics), wrong_source (switch data source), stale_data (search web), etc. Each fingerprint has a standard correction applied automatically.

Strategy Selection: For complex tasks, ADIN matches the situation to proven strategies with tracked success rates. Successful strategies are reinforced; failing ones are deprecated.

Confession Protocol: If ADIN exhausts its retry budget without finding reliable information, it transparently reports what it tried, what it knows, what it doesn't, and suggestions for next steps - instead of hallucinating.

Native Widget System & Integrations

ADIN renders 40+ native widgets (weather, stocks, crypto, sports, flights, packages, timers, recipes, GitHub, fitness) and connects to 20+ OAuth integrations (Gmail, Calendar, Google Docs/Sheets/Slides, Notion, Linear, GitHub, MongoDB, Postgres, Mercury, Twitter, Discord, YouTube, Strava, Fitbit, Oura). Tools are dynamically enabled based on conversation context.

Part 2: Positioning & Competitive Differentiation

The Core Positioning

ADIN is Your AI Research Partner.

Not a chatbot. Not a search engine. A research partner that remembers everything you've learned and helps you build on it.

The key insight: most AI resets every prompt. ADIN compounds knowledge. The longer you use it, the better it gets at understanding your work, connecting dots across your research, and surfacing insights you'd otherwise miss.

The Forcing Function

The competition optimizes for transactions. ADIN optimizes for compounding. That's a category difference.

Part 3: The Knowledge Flywheel

The Core Insight

Every research task ADIN performs creates knowledge. Most AI assistants throw this away after each session. ADIN captures, structures, and compounds it into a growing global knowledge repository that:

• Improves ADIN - Better answers for all users as the knowledge base grows
• Drives SEO - Structured content that ranks for long-tail research queries
• Powers GEO - Authoritative source data that AI search engines cite
• Converts Users - "This answer was generated by ADIN" → signup funnel

The Flywheel Effect

More users → More research → More knowledge → Better SEO/GEO → More discovery → More users

This is the compounding advantage that makes ADIN defensible. The knowledge repository is a moat that deepens with every user interaction.

Privacy-First Approach

The knowledge flywheel requires trust:

• Explicit Opt-In: Users choose what to contribute to the global repository
• Anonymization: No identifying information in public content
• Value Exchange: Contributors get credit and premium features
• Deletion Rights: Users can remove their contributions at any time

Part 4: Pricing Strategy (Usage-Based Model)

ADIN Pricing Tiers

Fast vs Slow Requests

Fast Requests ($0.01 each beyond included): Simple questions, single-tool operations, widget displays. Uses fast-tier models.

Slow Requests ($0.10 each beyond included): Complex multi-step research, agent delegation chains, debate system, ensemble forecasting. Uses power-tier models.

The Conversion Funnel

Key Conversion Triggers:

• Free → Pro: Hitting the 7-day memory limit and realizing value of persistence
• Pro → Power: Hitting the 500 message limit during a research sprint

Part 5: Go-To-Market Strategy (90-Day Sprint)

The Two-Channel Strategy

ADIN's GTM combines two complementary channels:

Channel 1: Twitter Research Threads → SEO/GEO

• Publish deep research on underserved topics
• Threads drive immediate engagement + backlinks
• Content becomes SEO pages that compound over time
• AI search engines cite well-structured research

Channel 2: Meta Ads → Direct User Acquisition

• Targeted ads to knowledge workers, researchers, analysts
• Retargeting visitors from SEO/Twitter
• Conversion-optimized landing pages by use case
• CAC-efficient at scale once creative is dialed in

Channel 1: Twitter Research Strategy

The goal is to become the go-to source for deep research on topics that are underserved by existing content. Each thread serves multiple purposes: immediate engagement, backlinks from people sharing, SEO page fodder, and GEO citation material.

Topic Selection Criteria:

Underserved Topic Categories:

Startup/VC Research - Most competitive analyses are either paywalled (PitchBook, CB Insights) or shallow (blog posts). Deep, free, well-sourced competitive landscapes have massive search demand.

• "Stripe competitive landscape 2026"
• "Notion vs Coda vs Craft - full comparison"
• "AI infrastructure market map"
• "How [company] makes money - business model breakdown"

Technical Explainers - Finance and legal concepts that professionals need but struggle to find clear explanations for.

• "How DCF valuation actually works"
• "Cap table mechanics explained"
• "SAFE vs Convertible Note - real differences"
• "What is a 409A valuation?"

Market Sizing - TAM/SAM/SOM analyses are in high demand from founders and investors but rarely done well publicly.

• "[Industry] market size 2026"
• "[Category] TAM analysis"
• "[Vertical] growth projections"

Current Events Analysis - When news breaks, people search for analysis. Fast, deep analysis captures traffic.

• "What [acquisition] means for [industry]"
• "[Company] layoffs analysis"
• "Impact of [regulation] on [sector]"

Thread Format (Optimized for Engagement + SEO):

• Hook tweet: Contrarian take or surprising finding (this is 90% of the battle)
• The "so what": Why this matters, who should care
• The meat: 5-10 tweets of actual research, data, analysis
• Methodology note: "I used ADIN to research this" (subtle product placement)
• CTA: Link to full report on ADIN site (captures email, enables retargeting)

Thread → SEO Page Pipeline:

Every thread becomes a long-form page on the ADIN site:

• Expand thread into 2,000-3,000 word article
• Add charts, tables, and structured data
• Include "Generated with ADIN" CTA
• Optimize for target keywords
• Submit to Google Search Console

Publishing Cadence:

Total Month 3: 55+ threads, 30+ SEO pages

Channel 2: Meta Ads Strategy

Meta (Facebook/Instagram) is the most efficient paid channel for B2C and prosumer products. The targeting capabilities allow precise reach of knowledge workers, and the creative format (video demos) shows the product in action.

Why Meta Over Google Ads:

Google Ads make sense later when branded search volume exists. For launch, Meta creates demand more efficiently.

Audience Targeting:

Primary Audiences:

• Job titles: Analyst, Researcher, Consultant, Associate, Investment Professional
• Interests: Venture capital, Private equity, Market research, Competitive intelligence
• Behaviors: Business decision makers, Technology early adopters
• Lookalikes: Based on early Pro subscribers

Creative Strategy:

Format: 15-30 second video demos showing ADIN in action

Creative Angles to Test:

Landing Pages (by Use Case):

Each ad set links to a use-case-specific landing page:

• /research - For competitive analysis, market research
• /analysis - For financial analysis, due diligence
• /legal - For contract review, legal research
• /writing - For content creation, reports

Pages include: Hero with value prop → Demo video → Feature highlights → Social proof → CTA (Start Free)

Budget Allocation:

Total Q1 Meta Spend: ~$18,000

Target Metrics:

By Month 3, the goal is CAC < $15 for Pro subscribers, which is excellent for a $20/month product (payback < 1 month).

Channel 3: The Viral Waitlist

Before full public launch, ADIN runs a viral waitlist that generates FOMO, seeds the user base with high-distribution individuals, and creates organic growth through referral mechanics.

The Access Tiers:

Tier 1: Instant Access goes to three groups. First, anyone with 5,000+ X followers gets in immediately because they have distribution and will talk about the product. Second, anyone with a .edu email gets instant access because researchers are a core ICP and academic users drive credibility. Third, anyone referred by an existing user skips the line, which incentivizes invites.

Tier 2: The Waitlist is where everyone else lands, but it's not a static queue. Users see their position (e.g., "#847 of 12,493") and can take actions to move up. Every friend who joins via their referral link moves them up 50 spots. Tweeting about ADIN (verified via Twitter connect) moves them up 100 spots. Connecting integrations early (Gmail, Calendar, GitHub) moves them up 50 spots each because it signals intent and makes activation faster when they get access.

Tier 3: The Scout Fast Pass creates a direct bridge to ADIN Fund goals. Anyone who submits a deal to the fund gets instant ADIN access. This turns the waitlist into a deal flow generation mechanism and aligns user acquisition with the fund's 10K scout target.

The Viral Mechanics:

Scarcity Through Limited Invites: When a user gets access, they receive exactly 3 invite codes. Not 10. Not unlimited. Three. This forces them to think carefully about who to invite (quality filter), makes recipients feel special (status), and creates constant conversation about ADIN (FOMO). The Clubhouse playbook, executed precisely.

Public Waitlist Position: The waitlist page shows your position prominently. This is inherently shareable. "I'm #847 on the ADIN waitlist" becomes a tweet, a Slack message, a text to a friend. Every share is free distribution.

Referral Leaderboard: A public leaderboard shows the top 25 referrers by name and referral count. The top 10 referrers each month get lifetime Pro accounts. This gamifies referrals and gives power users a reason to go hard on distribution.

"Why I Want ADIN" Prompt: During signup, users answer "What would you use ADIN for?" The best answers get featured on the site and receive instant access. This generates testimonial content, surfaces power user candidates, and creates another shareable moment ("My answer got featured!").

The Waitlist Funnel Math:

A viral coefficient of 0.625 means every 1,000 waitlist signups generate 625 additional signups organically. Combined with paid acquisition, this creates efficient compounding growth.

Integration with Paid Channels:

The waitlist amplifies Meta ad efficiency. Ads drive users to the waitlist, not directly to signup. This has three benefits. First, lower friction since joining a waitlist is easier than creating an account. Second, FOMO since "join 12,493 others waiting" is more compelling than "sign up now." Third, viral amplification since every paid user can bring organic users via referrals.

The CAC math improves significantly:

Timeline:

Success Metrics:

Combined Funnel Projections

Month 1:

• Twitter: 10 threads → 5,000 impressions/thread → 50K total impressions → 500 profile visits → 100 signups
• SEO: 5 pages live, minimal traffic yet
• Meta: $3K spend → 3,750 clicks → 560 signups
• Total signups: 660 | Pro conversions: 33 | MRR: $660

Month 2:

• Twitter: 20 cumulative threads → 100K cumulative impressions → 300 signups
• SEO: 10 pages → early rankings → 200 signups
• Meta: $6K spend → 12,000 clicks → 2,400 signups
• Total signups: 2,900 | Pro conversions: 230 | MRR: $4,600

Month 3:

• Twitter: 40 cumulative threads → 200K cumulative impressions → 600 signups
• SEO: 25 pages → improving rankings → 800 signups
• Meta: $9K spend → 28,000 clicks → 7,000 signups
• Total signups: 8,400 | Pro conversions: 840 | MRR: $16,800

End of Month 3 Targets:

• 10,000+ registered users
• 1,000+ Pro subscribers
• $20K+ MRR
• 25+ SEO pages ranking
• Clear channel economics proven

GEO Strategy (AI Search Optimization)

As AI-powered search grows, being cited as an authoritative source matters as much as traditional SEO. ADIN's research content is structured for AI extraction:

Structured Data: Every SEO page includes schema markup for easy AI parsing.

Clear Attribution: Sources are cited inline, making it easy for AI to verify and attribute.

Authoritative Voice: Content is written as definitive analysis, not opinion or speculation.

Update Frequency: Pages are updated when new information is available, signaling freshness.

The GEO Bet: Within 2-3 years, a significant portion of "search" will be AI-mediated. Building a citation-worthy knowledge base now creates a durable advantage.

Part 6: Execution Checklist (90-Day Sprint)

Week 1-2: Foundation

• Implement usage tracking (fast vs slow request classification)
• Build usage dashboard for users
• Create Stripe integration for usage-based billing
• Launch free tier (50 messages, 7-day memory)
• Launch Pro tier ($20/month)
• Set up Twitter account, optimize bio with link
• Create SEO page template on ADIN site
• Set up Meta Ads account, pixel, conversion tracking
• Design 3 landing pages (research, analysis, general)
• Publish first 5 research threads
• Launch first Meta ad creative tests ($50/day)
• Build waitlist landing page with position tracking
• Implement referral link system and tracking
• Create "Why I Want ADIN" submission form
• Set up Twitter OAuth for follower count verification

Week 3-4: Iteration

• Publish 5 more research threads (10 total)
• Convert top 5 threads to SEO pages
• Analyze Meta ad performance, kill losers
• Create 3 new ad creatives based on learnings
• Scale Meta to $100/day on winners
• First 500 signups
• First 25 Pro conversions
• Collect user feedback, iterate on onboarding
• Launch public referral leaderboard
• Begin featuring top "Why I Want ADIN" answers
• First batch of instant access (5K+ followers, top referrers)
• Track viral coefficient, optimize share prompts

Week 5-8: Scale

• Publish 20 more threads (30 total)
• Convert 10 more threads to SEO pages (15 total)
• Scale Meta to $200/day
• Launch retargeting campaigns
• Test lookalike audiences based on Pro subscribers
• 3,000 total signups
• 250 Pro subscribers
• $5K MRR
• Controlled waitlist rollout: 500-1,000 users/week
• Launch Scout Fast Pass program
• Monitor and optimize viral coefficient (target >0.5)

Week 9-12: Optimize & Compound

• Publish 25 more threads (55 total)
• Convert 15 more threads to SEO pages (30 total)
• Scale Meta to $300/day
• Optimize landing pages based on conversion data
• Launch Team tier
• First SEO pages hitting page 1
• First GEO citations observed
• 10,000 total signups
• 1,000 Pro subscribers
• $20K MRR
• Clear path to $50K MRR by Month 6
• Transition from waitlist to open access
• Maintain referral program for "skip the line" benefit

Key Takeaways

1. Two channels, one flywheel. Twitter builds authority and seeds SEO. Meta drives volume. Both feed the same conversion funnel.

2. Content is the product demo. Every research thread shows ADIN's capabilities. The best marketing is using the product in public.

3. SEO compounds, ads don't. Meta ads are for acceleration. SEO pages are for durability. Build both, but know the difference.

4. Speed wins. 90 days is enough to prove channel economics and hit $20K MRR. Move fast, measure everything, double down on winners.

5. GEO is the next SEO. Structured, well-sourced research content will be cited by AI search. Build for that future now.

6. Usage-based pricing aligns incentives. Free tier removes friction. Low monthly cost captures the long tail. Usage-based scaling means power users pay fairly for value received.

7. The knowledge repository is the moat. Every thread, every SEO page, every user interaction deepens the knowledge base. That's the compounding advantage.

8. Viral waitlists multiply paid acquisition. A 0.5+ viral coefficient means every paid signup generates half an organic signup. Combined with FOMO mechanics, this dramatically improves CAC efficiency.

9. Scarcity creates status. Limited invites (exactly 3 per user) force careful selection, make recipients feel special, and generate constant word-of-mouth.

10. Align acquisition with strategic goals. The Scout Fast Pass turns user acquisition into deal flow generation, directly supporting the fund's 10K scout target.