The next evolution of AI in SaaS isn’t about better models – it’s about context and action. Here’s what Brandon Foo (CEO, Paragon) and Ethan Lee (Director of Product) shared at SaaStr AI Day about what’s actually working:

1. Why LLM Wrappers Failed – And What Works Instead

The first wave of AI products were mostly “LLM wrappers” – simple chatbots built on top of models like GPT. They failed for two key reasons:

  • Zero context about your specific business or product
  • No ability to actually do anything beyond generate text

The learning? AI needs both context AND the ability to take action to deliver real value.

2. The Rise of RAG + AI Agents: The Winning Formula

Leading companies are now combining two approaches:

  • RAG (Retrieval Augmented Generation) to give AI real context about your business
  • AI agents with “tool calling” capabilities to actually take actions

Take Intercom’s AI support agent Finn – it can now access help centers, internal docs, and even process refunds through Stripe. That’s the difference between a toy and a real product.

3. The Three Types of Context Your AI Needs

Your AI agent needs access to:

  • Team documents & engineering specs (the “how it works” knowledge)
  • Systems of record like CRM (the structured data)
  • Unstructured info from Slack, Zoom, etc. (the tribal knowledge)

Without all three, you’re flying blind.

4. Real Examples of What’s Working

Companies winning with AI today:

  • Intercom: Support agent accessing docs + taking actions
  • Copy.ai: Sales automation using Gong recordings + Salesforce
  • TLDV: Meeting assistant creating Jira tickets automatically
  • Supper: AI analyst pulling from multiple data sources

The common thread? They all combine rich context with the ability to take real actions.

5. What Will Actually Matter Going Forward

As AI becomes ubiquitous, the differentiators will be:

  • Response accuracy (picking the right actions)
  • User experience in existing workflows
  • Intentional design of the interface

The models themselves won’t be the moat – it’s how you implement them.

The Bottom Line

The constraints holding back AI adoption aren’t about model capability anymore – they’re about data isolation and integration. The winners will be the ones who solve the “plumbing” of connecting AI to all your business systems.

Companies like Paragon are building the infrastructure layer to make this possible – giving AI agents access to thousands of tools through a single API. That’s what will take us from demos to real products that deliver value.

Want to learn how other companies are implementing these strategies? Join us at SaaStr Annual in May.

Related Posts

Pin It on Pinterest

Share This