Pricing in the AI Era: From Inputs to Outcomes, with Paid CEO Manny Medina

Market Trends and Success Factors

• Successful AI applications currently follow a "Hedgehog" strategy, focusing on narrow, specific problems rather than broad, general-purpose solutions.
• High-performing AI agents are replacing Business Process Outsourcing (BPO) roles rather than solely targeting high-paying creative professionals like lawyers or doctors.
• Examples of successful narrow-use cases include:
  • Quandary: Automating policy renewals in the insurance sector.
  • Owl: Reviewing insurance claims data.
  • Happy Robot: Negotiating freight loads by calling truckers on behalf of brokers.
  • Expo: Performing continuous penetration testing on specific applications.
• Companies attempting to solve broad problems immediately (e.g., "AI SDRs") often face high competition and "swirl," whereas those targeting specific labor pools (disappearing due to retirement or turnover) see stickier growth and better unit economics.
• "Co-pilot" models (assisting humans) are emerging for high-value creative roles, while "autopilot" models (fully replacing humans) are succeeding in tasks nobody desires to do.

Pricing Frameworks and Strategies

• Manny Medina identifies four pricing approaches currently working for AI companies:
  • Activity-based: Charging by consumption (e.g., credits or tokens), which is the current default but leads to low margins.
  • Workflow-based: Charging for a defined sequence of activities (e.g., "document review"), allowing differentiation based on document complexity.
  • Outcome-based: Charging for measurable results (e.g., "qualified meetings booked"), potentially with a base fee plus an "outcome bonus" to align with customer value.
  • Agent-based: Pricing equivalent to a human salary (e.g., $20k/year for an agent replacing a $90k human) to allow the AI to be budgeted as headcount rather than a software tool.
• Customers naturally default to easy-to-buy models (fixed or consumption prices) for the first year; the responsibility lies with the seller to renegotiate for value-based pricing once the solution proves effective.
• Bespoke contracts are becoming viable because AI eliminates the need for rigid SKU-based pricing, allowing for direct negotiation based on specific customer definitions of success (e.g., CSAT, NPS, time to resolution).
• Charging by "agent" allows companies to tap into HR/Headcount budgets rather than smaller RevOps tool budgets, significantly expanding the potential deal size.

Cost Dynamics and Margins

• Contrary to the belief that token costs will plummet, Medina argues that inference-time compute costs for reasoning models may rise as models require deeper thinking.
• Margins in AI are currently mismatched with value because many companies lack the visibility to track specific customer profitability and unit economics.
• Cost structures are compounding as agents integrate multiple modalities (text, voice, avatar), introducing non-LLM third-party API costs that obscure true margins.
• Paid, Medina's new company, addresses this by providing a unified layer for billing, invoicing, margin management, and vendor management to give AI founders visibility into their unit economics.
• The "value capture" gap persists where customers retain most of the cost savings; successful agents must transition to pricing models that explicitly capture a share of the realized labor cost reduction.

Strategic Shifts in AI Development

• The "vibe coding" methodology is validated as a rapid prototyping tool, but the core business challenge remains building a company with clear unit economics rather than just code.
• New AI founders are advised to focus on a narrow Ideal Customer Profile (ICP) first to build excellence and organic word-of-mouth before scaling to a broader platform.
• BPOs are likely to eventually develop their own internal agents using their proprietary data, making initial BPO-targeted AI businesses face a competitive threat in the long term.
• Collaborative workflows remain sticky, but pure "workflow" software faces high churn unless it defines a unique, high-value standard for a specific vertical.
• The industry is moving from "vibe revenue" (trials and POCs) to "renewal land," which will serve as the primary filter for separating viable businesses from ineffective ones.

Founder Insights and Forward-Looking Statements

• Medina believes AGI is effectively "here" but underutilized, describing it as a "scaffold for human imagination" that allows people to see farther and achieve the previously impossible.
• He predicts that models will not commoditize soon due to the increasing cost disparity between input and output tokens in reasoning models.
• Paid is currently onboarding founders manually to ensure correct agent work capture and provide best-practice guidance on monetization and margins.
• For AI founders, the primary advice is to disregard Total Addressable Market (TAM) concerns; a small, well-served TAM with a superb experience will become a large TAM.
• Recommended reading includes the statistical NLP book by Rich Manning for understanding foundational machine learning concepts.
• Key inspirations for Medina include Jeff Bezos, Todd Olson (Pendo), Sam Altman (OpenAI), and the Collison brothers.