Skip to main content

AI Agent

AI systems that are designed for a specific task in an industry. They should be an expert of the task, tailored to the industry, fits into the workflow and improves efficiency.

More Foundation models are coming into production

  • OpenAI, Anthropic, DeepSeek, Grok

Parallel to SaaS

  • SaaS boomed from XMLHttpRequest (AJAX)
  • User experience is going to be the key to its success
  • Obviously good ideas that could be mass consumer products
    • Chat, Docs, Maps, Storage, Emails
    • 0 startups won, incumbents won
  • Unobvious ideas that become mass consumer products
    • Uber, Instacart, Airbnb
    • Startups win these, incumbents loss
  • Enterprise SaaS companies
    • Salesforce, Shopify, Carta, Gusto (hr)
    • There is an optical issue with "web apps" and people didn't see the potential improvements the browser and internet can have
    • No incumbents can be that good at what each B2B SaaS company specializes in and there is a bundling vs unbundling argument
    • There is a huge portion of the pie in this field

LLM -> AI Agents

  • Can startups grow with small number of engineers with prompts/LLMs
  • Obviously good ideas that could be mass consumer products
    • General voice assistant, search,
    • Incumbents will win (perplexity vs google could be a counter example)
  • Can you replace the operation teams? More payroll be replaced by AI Agents compared to SaaS?
  • Need to sell to the company to replace a team, which is a brutal task if not selling to high enough person
  • Can this help manager level capacity
    • Because this becomes a bottleneck for a company
  • What vertical is correct?
    • Boring and admin task

Outset.ai

  • To conduct surveys and compute qualtrics
  • Language based business which is really good for LLM substitution

Momentic.ai

  • For testing and QA and is replacing the entire QA team
  • Rainforest QA was selling to the company as making the QA team "more efficient" but not enough to replace them all

Customer Support ai

  • Crowded space but most are bad/simple with 0 shot LLM prompting that can't replace any support team
  • It needs very complicated and detailed steps

Handle customer ticketing

  • Very tailored solution for one single business